Model benchmark: testaa kaikki Ollama-mallit järjestelmällisesti

Ajaa täyden pipeline-kierroksen per malli × skenaario:
1. Client-prompti → vaatimukset
2. Manager/SPEC_SYSTEM → JSON-speksi
3. Template-generointi → koodi
4. Validointi + LLM-korjaussilmukka
5. uv sync + pytest

Tuottaa vertailutaulukon: speksin laatu, testien tulos, nopeus.
Tukee suoraa Ollamaa (--ollama) ja hub-reittiä (--hub).

.gitignore

@@ -38,5 +38,11 @@ Cargo.lock
 # Ajonaikaiset tietokannat
 *.db
 
+# Lokitiedostot
+*.log
+
 # Wanha versio
 temp/
+
+# Muut
+zipit/**

TEMPLATING.md

@@ -0,0 +1,157 @@
+# Templating — rakennuspalaset koodigeneroinnissa
+
+## Perusperiaate
+
+Kielimalli päättää **mitä** rakennetaan (entiteetit, kentät, tyypit, yhteydet).
+Template-funktiot päättävät **miten** se rakennetaan (importit, engine setup, testikonfiguraatio).
+
+```
+Projektikuvaus → LLM → JSON-speksi → Templateit → Koodi → Validointi
+```
+
+LLM:n kontribuutio on yksi JSON-rakenne. Kaikki muu on determinististä —
+sama speksi tuottaa aina saman koodin.
+
+## Miksi tämä toimii
+
+Pienen kielimallin (0.5B–7B) vahvuudet ja heikkoudet ovat epäsymmetrisiä:
+
+| Tehtävä | LLM:n kyky | Ratkaisu |
+|---------|-----------|----------|
+| Tunnista entiteetit kuvauksesta | Hyvä | LLM tekee |
+| Valitse kenttätyypit | Hyvä | LLM tekee |
+| Muista importit oikein | Huono | Template tekee |
+| SQLite connect_args | Huono | Template tekee |
+| Testikonfiguraatio | Huono | Template tekee |
+| Dockerfile-rakenne | Huono | Template tekee |
+
+Annetaan mallin tehdä se missä se on hyvä. Hoidetaan loput mekaanisesti.
+
+## JSON-speksi
+
+Kielimallin ainoa tuotos on JSON joka kuvaa projektin rakenteen:
+
+```json
+{
+  "project_name": "library-app",
+  "entities": [
+    {
+      "name": "Author",
+      "table_name": "authors",
+      "fields": [
+        {"name": "name", "sa_type": "String(255)", "py_type": "str", "nullable": false, "default": null}
+      ]
+    },
+    {
+      "name": "Book",
+      "table_name": "books",
+      "fields": [
+        {"name": "title", "sa_type": "String(255)", "py_type": "str", "nullable": false, "default": null},
+        {"name": "author_id", "sa_type": "Integer", "py_type": "int", "nullable": false, "default": null}
+      ]
+    }
+  ],
+  "relationships": [
+    {"from": "Book", "field": "author_id", "to": "Author", "type": "many-to-one"}
+  ],
+  "extra_imports": []
+}
+```
+
+Speksin laatu ratkaisee kaiken. Hyvä speksi → hyvä projekti. Huono speksi →
+teknisesti toimiva mutta sisällöllisesti väärä projekti.
+
+## Architect-promptin rooli
+
+Architect-agentti (JSON-speksin generoija) on kriittisin kohta koko pipelinessa.
+Sitä ohjataan neljällä keinolla:
+
+1. **Chain-of-thought** — malli miettii ensin entiteetit, sitten kentät,
+   sitten yhteydet, vasta lopuksi JSON
+2. **Domain-esimerkit** — Todo, verkkokauppa, blogi — malli näkee miltä
+   hyvä speksi näyttää eri domaineissa
+3. **Anti-patternit** — turhat ID-kentät, Enum-tyypit, suomenkieliset nimet
+4. **Yhteyssäännöt** — jokainen `_id`-kenttä tarvitsee relationship-merkinnän
+
+Isompi malli tässä yhdessä kohdassa parantaisi kaikkien projektien laatua.
+
+## Templateit
+
+Jokainen template on funktio joka ottaa speksin ja palauttaa koodia:
+
+```
+tmplModels(spec)     → models.py      (SQLAlchemy, ForeignKey, relationship)
+tmplSchemas(spec)    → schemas.py     (Pydantic Create/Response/Detail)
+tmplMain(spec)       → main.py        (FastAPI CRUD + nested endpoints + FK-validointi)
+tmplTests(spec)      → test_main.py   (pytest + TestClient + helper-funktiot)
+tmplPyproject(spec)  → pyproject.toml (PEP 621)
+tmplDockerfile()     → Dockerfile     (uv + non-root user)
+```
+
+Templateit generoivat automaattisesti:
+- ForeignKey-constraintit ja relationship()-määrittelyt
+- Nested endpointit (`GET /authors/{id}/books/`)
+- FK-validointi (404 jos parent-entiteettiä ei ole)
+- Detail-schemat (Book + author-data mukana)
+- Test-helperit jotka luovat parent-entiteetit ensin
+- Bad FK -testit (varmistaa että orpo-validointi toimii)
+
+## Validointi
+
+Generoitu koodi validoidaan mekaanisesti ennen käyttöä:
+
+- Syntaksitarkistus (AST parse)
+- Projektin sisäiset importit (löytyykö nimi lähdetiedostosta)
+- SQLite connect_args
+- Relatiiviset importit (kielletty)
+- Testien rakenne (ei saa kopioida appia)
+- pyproject.toml (ei poetryä)
+- Dockerfile (ei poetryä, uv cache -oikeudet)
+
+Docker-testi ajaa koko projektin: build → pytest → API smoke test.
+
+## Rajoitukset
+
+Templateit kattavat rakenteellisesti tunnetut projektit:
+
+| Stack | Kattavuus |
+|-------|-----------|
+| FastAPI + SQLAlchemy CRUD | Toimii hyvin |
+| Streamlit + DuckDB dashboard | Toimii hyvin |
+| Muu | Ei templatea → ei toimi |
+
+**Ei kata:**
+- Custom business-logiikka (algoritmit, laskenta, ML)
+- Epätyypilliset arkkitehtuurit (WebSocket, graafit, tapahtumapohjaiset)
+- Frontend-sovellukset (React, Vue)
+- Mikä tahansa mitä template ei tunne
+
+Arvio: templateit kattavat ~20% kaikista mahdollisista projekteista, mutta juuri
+sen 20% mitä opiskelu- ja prototyyppiympäristöissä tarvitaan useimmin.
+
+## Laajentaminen
+
+Uuden stackin lisääminen vaatii:
+
+1. Uudet template-funktiot (käsityö, ~200–400 riviä per stack)
+2. JSON-speksin laajennos (uudet kentät jos tarvitaan)
+3. Validointisäännöt uudelle stackille
+4. Docker-testikonfiguraatio
+
+Jokainen template on staattinen — se ei opi eikä sopeudu. Kattavuus kasvaa
+vain kirjoittamalla lisää templateja.
+
+## Hybridi: seuraava askel
+
+Paras lopputulos syntyisi yhdistelmällä:
+
+```
+Speksi → Template (runko) → LLM (business-logiikka) → Validointi
+```
+
+Template tuottaa toimivan CRUD-pohjan. LLM lisää domain-kohtaisen logiikan
+pienissä palasissa (yksi funktio kerrallaan). Mekaaninen validointi
+tarkistaa jokaisen lisäyksen.
+
+Tämä palauttaa LLM:n epäluotettavuuden takaisin peliin, mutta rajattuna:
+virheet ovat paikallisia (yksi funktio) eivätkä rakenteellisia (koko projekti).

kipina-node

@@ -0,0 +1,131 @@
+#!/bin/bash
+# Kipinä Node — lataa oikea binääri ja käynnistä
+set -e
+
+BASE_URL="https://kipina.studio/download"
+HUB_URL="${KIPINA_HUB:-wss://kipina.studio/ws}"
+OLLAMA_URL="${OLLAMA_URL:-http://localhost:11434}"
+
+# Tunnista OS ja arkkitehtuuri
+OS=$(uname -s | tr '[:upper:]' '[:lower:]')
+ARCH=$(uname -m)
+
+case "$OS-$ARCH" in
+    darwin-arm64)  BINARY="kipina-node-macos-arm64" ;;
+    darwin-x86_64) BINARY="kipina-node-macos-arm64" ;; # Rosetta
+    linux-x86_64)  BINARY="kipina-node-linux-x86_64" ;;
+    linux-aarch64) BINARY="kipina-node-linux-arm64" ;;
+    *) echo "Ei tuettu: $OS-$ARCH"; exit 1 ;;
+esac
+
+echo ""
+echo "  ╔══════════════════════════════════════╗"
+echo "  ║       Kipinä Agentic Node            ║"
+echo "  ╚══════════════════════════════════════╝"
+echo ""
+echo "  OS: $OS ($ARCH)"
+echo ""
+
+# Etsi Ollama-instanssit
+CANDIDATES=(
+    "http://localhost:11434"
+    "http://127.0.0.1:11434"
+    "http://ollama:11434"
+    "http://host.docker.internal:11434"
+)
+
+# Lisää OLLAMA_URL listaan jos asetettu ja ei jo mukana
+if [ -n "$OLLAMA_URL" ]; then
+    ALREADY=false
+    for c in "${CANDIDATES[@]}"; do
+        [ "$c" = "$OLLAMA_URL" ] && ALREADY=true
+    done
+    $ALREADY || CANDIDATES=("$OLLAMA_URL" "${CANDIDATES[@]}")
+fi
+
+echo "  Etsitään Ollama-instansseja..."
+FOUND=()
+for url in "${CANDIDATES[@]}"; do
+    if curl -s --connect-timeout 1 "$url/api/tags" &>/dev/null; then
+        FOUND+=("$url")
+    fi
+done
+
+if [ ${#FOUND[@]} -eq 0 ]; then
+    # Ei löytynyt — yritä käynnistää lokaali
+    if command -v ollama &>/dev/null; then
+        echo "  Käynnistetään Ollama..."
+        ollama serve &>/dev/null &
+        sleep 3
+        if curl -s --connect-timeout 1 "http://localhost:11434/api/tags" &>/dev/null; then
+            OLLAMA_URL="http://localhost:11434"
+            echo "  ✓ Ollama käynnistetty ($OLLAMA_URL)"
+        else
+            echo "  ✗ Ollaman käynnistys epäonnistui."
+            exit 1
+        fi
+    else
+        echo ""
+        echo "  ✗ Ollamaa ei löytynyt."
+        echo "    Kontti/remote: OLLAMA_URL=http://HOST:11434 ./kipina-node"
+        echo "    Asenna:        curl -fsSL https://ollama.ai/install.sh | sh"
+        exit 1
+    fi
+elif [ ${#FOUND[@]} -eq 1 ]; then
+    OLLAMA_URL="${FOUND[0]}"
+    echo "  ✓ Ollama löytyi: $OLLAMA_URL"
+else
+    echo ""
+    echo "  Löytyi ${#FOUND[@]} Ollama-instanssia:"
+    echo ""
+    for i in "${!FOUND[@]}"; do
+        echo "    $((i+1))) ${FOUND[$i]}"
+    done
+    echo ""
+    read -p "  Valitse [1-${#FOUND[@]}]: " -r CHOICE
+    if [[ "$CHOICE" =~ ^[0-9]+$ ]] && [ "$CHOICE" -ge 1 ] && [ "$CHOICE" -le ${#FOUND[@]} ]; then
+        OLLAMA_URL="${FOUND[$((CHOICE-1))]}"
+    else
+        OLLAMA_URL="${FOUND[0]}"
+        echo "  Käytetään oletusta: $OLLAMA_URL"
+    fi
+    echo "  ✓ Valittu: $OLLAMA_URL"
+fi
+
+echo ""
+echo "  Hub: $HUB_URL"
+echo "  Ollama: $OLLAMA_URL"
+if [ -n "$KIPINA_MODEL" ]; then
+    echo "  Malli: $KIPINA_MODEL (Ympäristömuuttujasta)"
+fi
+
+# Lataa binääri
+BIN_PATH="./kipina-node-bin"
+if [ -f "$BIN_PATH" ]; then
+    echo ""
+    read -p "  Löydettiin vanha kipina-node-bin lokaalisti. Haluatko poistaa sen ja ladata uusimman version? [Y/n] " -r DEL_CHOICE
+    if [[ "$DEL_CHOICE" =~ ^[Nn]$ ]]; then
+        echo "  ✓ Käytetään lokaalia versiota."
+    else
+        rm -f "$BIN_PATH"
+        echo "  ✓ Vanha binääri poistettu ja korvataan uudella."
+    fi
+fi
+
+if [ ! -f "$BIN_PATH" ]; then
+    echo "  Ladataan tuorein $BINARY..."
+    curl -sSL "$BASE_URL/$BINARY" -o "$BIN_PATH"
+    chmod +x "$BIN_PATH"
+fi
+
+echo ""
+echo "  ✓ Siirrytään Kipinä Noden hallintaan..."
+echo "    Ctrl+C pysäyttää"
+echo ""
+
+if [ -n "$KIPINA_MODEL" ]; then
+    export OLLAMA_MODEL="$KIPINA_MODEL"
+fi
+export HUB_URL="$HUB_URL"
+export OLLAMA_URL="$OLLAMA_URL"
+exec "$BIN_PATH"

network-poc/Dockerfile.native

@@ -0,0 +1,21 @@
+# Native-node: Rust + Ollama-client (ei GPU-tunnistusta)
+FROM rust:slim AS builder
+RUN apt-get update && apt-get install -y pkg-config libssl-dev && rm -rf /var/lib/apt/lists/*
+WORKDIR /app
+COPY Cargo.toml Cargo.lock* ./
+COPY native-node/Cargo.toml native-node/Cargo.toml
+COPY native-node/src native-node/src
+# Dummy-cratet workspace-yhteensopivuuteen
+COPY hub/Cargo.toml hub/Cargo.toml
+COPY node/Cargo.toml node/Cargo.toml
+COPY cli/Cargo.toml cli/Cargo.toml
+RUN mkdir -p hub/src node/src cli/src && touch hub/src/main.rs node/src/lib.rs cli/src/main.rs
+RUN --mount=type=cache,target=/usr/local/cargo/registry \
+    --mount=type=cache,target=/app/target \
+    cargo build --release -p native-node --no-default-features \
+    && cp /app/target/release/native-node /usr/local/bin/native-node
+
+FROM debian:bookworm-slim
+RUN apt-get update && apt-get install -y ca-certificates && rm -rf /var/lib/apt/lists/*
+COPY --from=builder /usr/local/bin/native-node /usr/local/bin/native-node
+CMD ["native-node"]

network-poc/Dockerfile.prod

@@ -5,23 +5,27 @@ FROM node:22-slim AS frontend
 WORKDIR /app/frontend
 COPY frontend/package.json frontend/package-lock.json* ./
 RUN npm install --silent
-COPY frontend/ .
+# Cache-buster: git hash pakottaa rebuildin kun koodi muuttuu
-COPY frontend/public/pkg public/pkg
+ARG CACHEBUST=0
+COPY frontend/src/ ./src/
+COPY frontend/public/ ./public/
+COPY frontend/astro.config.mjs frontend/tsconfig.json ./
 RUN npm run build
 
 # --- Vaihe 2: Wasm (wasm-pack) ---
+# Cargo registry cachetetaan mount-cachella, lähdekoodi kopioidaan tuoreena
 FROM rust:slim AS wasm-builder
 RUN apt-get update && apt-get install -y curl pkg-config libssl-dev g++ && rm -rf /var/lib/apt/lists/*
 RUN curl https://rustwasm.github.io/wasm-pack/installer/init.sh -sSf | sh
 WORKDIR /app
 COPY Cargo.toml Cargo.lock* ./
 COPY node/Cargo.toml node/Cargo.toml
-COPY node/src node/src
-# Dummy-cratet jotta workspace Cargo.toml on tyytyväinen
 COPY hub/Cargo.toml hub/Cargo.toml
 COPY native-node/Cargo.toml native-node/Cargo.toml
 COPY cli/Cargo.toml cli/Cargo.toml
 RUN mkdir -p hub/src native-node/src cli/src && touch hub/src/main.rs native-node/src/main.rs cli/src/main.rs
+ARG CACHEBUST=0
+COPY node/src node/src
 RUN --mount=type=cache,target=/usr/local/cargo/registry \
     --mount=type=cache,target=/app/target \
     cd node && wasm-pack build --target web --out-dir /app/wasm-pkg
@@ -32,12 +36,12 @@ RUN apt-get update && apt-get install -y pkg-config libssl-dev && rm -rf /var/li
 WORKDIR /app
 COPY Cargo.toml Cargo.lock* ./
 COPY hub/Cargo.toml hub/Cargo.toml
-COPY hub/src hub/src
-# Tarvitaan dummy-cratet jotta workspace kompiloi
 COPY node/Cargo.toml node/Cargo.toml
 COPY native-node/Cargo.toml native-node/Cargo.toml
 COPY cli/Cargo.toml cli/Cargo.toml
 RUN mkdir -p node/src native-node/src cli/src && touch node/src/lib.rs native-node/src/main.rs cli/src/main.rs
+ARG CACHEBUST=0
+COPY hub/src hub/src
 RUN --mount=type=cache,target=/usr/local/cargo/registry \
     --mount=type=cache,target=/app/target \
     cargo build --release -p hub \
@@ -51,11 +55,6 @@ COPY --from=hub-builder /usr/local/bin/hub /usr/local/bin/hub
 COPY --from=frontend /app/frontend/dist /app/frontend/dist
 COPY --from=wasm-builder /app/wasm-pkg /app/frontend/dist/pkg
 
-# Kopioidaan GUIDE.md ja templates
-COPY frontend/public/GUIDE.md /app/frontend/dist/GUIDE.md
-COPY frontend/public/templates /app/frontend/dist/templates
-COPY frontend/public/avatars /app/frontend/dist/avatars
-
 WORKDIR /app
 ENV STATIC_DIR=/app/frontend/dist
 EXPOSE 3000

network-poc/deploy-local.sh

@@ -0,0 +1,56 @@
+#!/bin/bash
+# Kipinä Studio — paikallinen kehitysympäristö
+# Buildaa frontendin, käynnistää hubin ja native-noden (Ollama)
+# Käyttö: ./deploy-local.sh
+set -e
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+cd "$SCRIPT_DIR"
+
+cleanup() { echo ""; echo "Pysäytetään..."; kill $HUB_PID $NODE_PID 2>/dev/null; exit 0; }
+trap cleanup INT TERM
+
+# Portti vapaaksi
+lsof -ti:3000 | xargs kill -9 2>/dev/null || true
+
+# Frontend
+echo "[1/3] Frontend..."
+cd "$SCRIPT_DIR/frontend"
+[ -d node_modules ] || npm install --silent
+npm run build 2>&1 | tail -1
+cd "$SCRIPT_DIR"
+
+# Hub
+echo "[2/3] Hub..."
+STATIC_DIR="$SCRIPT_DIR/frontend/dist" cargo run -p hub 2>&1 &
+HUB_PID=$!
+until curl -sf http://localhost:3000 >/dev/null 2>&1; do sleep 1; done
+
+# Native-node
+NODE_PID=""
+if curl -sf http://localhost:11434/api/tags >/dev/null 2>&1; then
+    MODEL=$(curl -s http://localhost:11434/api/tags | python3 -c "
+import sys,json
+ms=json.load(sys.stdin).get('models',[])
+for m in ms:
+    n=m['name']
+    if '7b' in n and 'coder' in n: print(n); exit()
+for m in ms:
+    if 'coder' in m['name']: print(m['name']); exit()
+if ms: print(ms[0]['name'])
+" 2>/dev/null)
+    if [ -n "$MODEL" ]; then
+        echo "[3/3] Native-node ($MODEL)..."
+        HUB_URL=ws://localhost:3000/ws OLLAMA_MODEL="$MODEL" \
+            cargo run -p native-node --no-default-features 2>&1 &
+        NODE_PID=$!
+    else
+        echo "[3/3] Ollama: ei malleja (ollama pull qwen2.5-coder:7b)"
+    fi
+else
+    echo "[3/3] Ei Ollamaa — Wasm-fallback selaimessa"
+fi
+
+echo ""
+echo "=== http://localhost:3000 === Ctrl+C pysäyttää"
+open http://localhost:3000 2>/dev/null || xdg-open http://localhost:3000 2>/dev/null || true
+wait $HUB_PID

network-poc/deploy-remote.sh

@@ -0,0 +1,59 @@
+#!/bin/bash
+# Kipinä Studio — tuotanto-deploy kipina.studioon
+# Buildaa Docker-imagen (frontend + hub + wasm) ja vie palvelimelle
+# Käyttö: ./deploy-remote.sh
+set -e
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+cd "$SCRIPT_DIR"
+
+SERVER="ubuntu@86.50.252.98"
+REMOTE_DIR="~/code/agentic-studio/network-poc"
+SSH_OPTS="-o StrictHostKeyChecking=no"
+
+# SSH-avain — yritetään yhdistää, jos ei onnistu, pyydetään avainta
+if ! ssh $SSH_OPTS "$SERVER" "echo ok" >/dev/null 2>&1; then
+    echo "SSH-yhteys ei onnistu, lisätään avain..."
+    ssh-add "$HOME/.ssh/id_rsa" 2>/dev/null || ssh-add
+fi
+
+# Auto-commit
+if ! git diff --quiet HEAD 2>/dev/null || \
+   [ -n "$(git ls-files --others --exclude-standard 2>/dev/null)" ]; then
+    echo "Uncommitted muutoksia — commitoidaan..."
+    read -rp "  Commit-viesti: " msg
+    [ -z "$msg" ] && msg="Deploy $(date +%Y-%m-%d\ %H:%M)"
+    git add -A && git commit -m "$msg"
+fi
+
+echo "=== Kipinä Studio Deploy → kipina.studio ==="
+
+# 1. Docker-image (CACHEBUST pakottaa lähdekoodin uudelleenkopioinnin)
+echo "[1/4] Docker build..."
+docker build --platform linux/amd64 -f Dockerfile.prod \
+    --build-arg CACHEBUST="$(git rev-parse HEAD)" \
+    -t kipina-agentic:latest .
+
+# 2. Pakkaus
+echo "[2/4] Pakataan..."
+docker save kipina-agentic:latest | gzip > /tmp/kipina-agentic.tar.gz
+echo "  $(du -h /tmp/kipina-agentic.tar.gz | cut -f1)"
+
+# 3. Siirto
+echo "[3/4] Siirretään..."
+scp $SSH_OPTS /tmp/kipina-agentic.tar.gz "$SERVER:/tmp/"
+scp $SSH_OPTS docker-compose.prod.yml Caddyfile.prod "$SERVER:$REMOTE_DIR/"
+
+# 4. Käynnistys
+echo "[4/4] Käynnistetään..."
+ssh $SSH_OPTS "$SERVER" "gunzip -c /tmp/kipina-agentic.tar.gz | docker load && rm /tmp/kipina-agentic.tar.gz"
+ssh $SSH_OPTS "$SERVER" "cd $REMOTE_DIR && docker compose -f docker-compose.prod.yml down && docker compose -f docker-compose.prod.yml up -d"
+
+# Discord
+WEBHOOK="https://discord.com/api/webhooks/1489504066898755687/8U02d0wug-3MkVax0xMmRoj0s_-V1psnNLPWdSOjnGnKRBUpPjaU6XiX9Iu8DgJI69AP"
+HASH=$(git log -1 --pretty=format:"%h" 2>/dev/null || echo "?")
+MSG=$(git log -1 --pretty=format:"%s" 2>/dev/null || echo "?")
+PAYLOAD=$(python3 -c "import json,sys; print(json.dumps({'content':sys.argv[1]}))" \
+    "🚀 **Kipinä Studio julkaistu!** \`${HASH}\` ${MSG} https://kipina.studio")
+curl -sf -H "Content-Type: application/json" -d "$PAYLOAD" "$WEBHOOK" >/dev/null || true
+
+echo "=== Valmis! https://kipina.studio ==="

network-poc/deploy.sh

@@ -1,70 +0,0 @@
-#!/bin/bash
-set -e
-
-if [ "$1" == "local" ]; then
-    echo "=== Kipinä Studio Local Development ==="
-    echo "Käynnistetään kokonaisuus puhtaasti Docker-kontissa..."
-    docker compose up agentic-poc
-    exit 0
-fi
-
-SERVER="ubuntu@86.50.252.98"
-REMOTE_DIR="~/code/agentic-studio/network-poc"
-KEY="$HOME/.ssh/id_rsa"
-SSH_OPTS="-o StrictHostKeyChecking=no -i $KEY"
-
-# Varmistetaan, että SSH-avain on agentissa
-if ! ssh-add -l 2>/dev/null | grep -q id_rsa; then
-    echo "SSH-avain ei ole agentissa. Lisätään..."
-    ssh-add "$KEY"
-fi
-
-echo "=== Kipinä Studio Deploy ==="
-
-# 0. Commitoidaan uncommitted muutokset ennen deployta
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-if ! git -C "$SCRIPT_DIR" diff --quiet HEAD 2>/dev/null || \
-   [ -n "$(git -C "$SCRIPT_DIR" ls-files --others --exclude-standard 2>/dev/null)" ]; then
-    echo "[0] Uncommitted muutoksia havaittu — commitoidaan..."
-    read -rp "    Commit-viesti: " DEPLOY_MSG
-    if [ -z "$DEPLOY_MSG" ]; then
-        DEPLOY_MSG="Deploy $(date +%Y-%m-%d\ %H:%M)"
-    fi
-    git -C "$SCRIPT_DIR" add -A
-    git -C "$SCRIPT_DIR" commit -m "$DEPLOY_MSG"
-    echo "    Commitoitu: $DEPLOY_MSG"
-fi
-
-# 1. Rakennetaan Docker-image lokaalisti
-echo "[1/4] Rakennetaan image lokaalisti..."
-docker build --platform linux/amd64 -f Dockerfile.prod -t kipina-agentic:latest .
-
-# 2. Tallennetaan tiedostoon
-echo "[2/5] Pakataan image..."
-docker save kipina-agentic:latest | gzip > /tmp/kipina-agentic.tar.gz
-echo "       Koko: $(du -h /tmp/kipina-agentic.tar.gz | cut -f1)"
-
-# 3. Siirretään palvelimelle
-echo "[3/5] Siirretään palvelimelle..."
-scp $SSH_OPTS /tmp/kipina-agentic.tar.gz $SERVER:/tmp/
-scp $SSH_OPTS docker-compose.prod.yml Caddyfile.prod $SERVER:$REMOTE_DIR/
-
-# 4. Ladataan image ja käynnistetään
-echo "[4/5] Ladataan image palvelimella..."
-ssh $SSH_OPTS $SERVER "gunzip -c /tmp/kipina-agentic.tar.gz | docker load && rm /tmp/kipina-agentic.tar.gz"
-
-echo "[5/5] Käynnistetään palvelut uudelleen..."
-ssh $SSH_OPTS $SERVER "cd $REMOTE_DIR && docker compose -f docker-compose.prod.yml down && docker compose -f docker-compose.prod.yml up -d"
-
-echo "=== Valmis! https://kipina.studio ==="
-
-# Discord-notifikaatio
-DISCORD_WEBHOOK="https://discord.com/api/webhooks/1489504066898755687/8U02d0wug-3MkVax0xMmRoj0s_-V1psnNLPWdSOjnGnKRBUpPjaU6XiX9Iu8DgJI69AP"
-COMMIT_HASH=$(git -C "$SCRIPT_DIR" log -1 --pretty=format:"%h" 2>/dev/null || echo "?")
-COMMIT_MSG=$(git -C "$SCRIPT_DIR" log -1 --pretty=format:"%s" 2>/dev/null || echo "?")
-# python3 escapettaa erikoismerkit JSON-turvallisesti
-PAYLOAD=$(python3 -c "import json,sys; print(json.dumps({'content': sys.argv[1]}))" \
-  "🚀 **Kipinä Studio julkaistu!**
-> \`${COMMIT_HASH}\` ${COMMIT_MSG}
-> https://kipina.studio")
-curl -s -H "Content-Type: application/json" -d "$PAYLOAD" "$DISCORD_WEBHOOK" > /dev/null

network-poc/docker-compose.prod.yml

@@ -19,6 +19,9 @@ services:
     restart: unless-stopped
     environment:
       - DATABASE_PATH=/data/nodes.db
+      - STATIC_DIR=/app/frontend/dist
+      - ADMIN_PASSWORD=${ADMIN_PASSWORD:-}
+      - NODE_API_KEY=${NODE_API_KEY:-}
     volumes:
       - hub_data:/data
 

network-poc/frontend/public/GUIDE.md

@@ -230,6 +230,188 @@ mitä luokkia importata.
 
 ---
 
+## Rakennuspalaset vs. vapaa generointi
+
+Kielimalli voi generoida koodia kahdella perustavanlaatuisesti eri tavalla.
+Ymmärtäminen milloin kumpikin toimii on avain luotettavaan koodigenerointi-pipelineen.
+
+### Tapa 1: Vapaa generointi (naivi)
+
+LLM generoi jokaisen tiedoston tyhjästä. Prompti kuvaa mitä halutaan,
+malli tuottaa koko tiedoston — importeista lähtien.
+
+```mermaid
+flowchart LR
+    P["Prompti"] --> LLM1["LLM: models.py"]
+    LLM1 --> V1{"Validointi"}
+    V1 -->|virhe| LLM1
+    V1 -->|ok| LLM2["LLM: schemas.py"]
+    LLM2 --> V2{"Validointi"}
+    V2 -->|virhe| LLM2
+    V2 -->|ok| LLM3["LLM: main.py"]
+    LLM3 --> V3{"..."}
+
+    style V1 fill:#1a1e2e,stroke:#f85149,color:#c9d1d9
+    style V2 fill:#1a1e2e,stroke:#f85149,color:#c9d1d9
+    style V3 fill:#1a1e2e,stroke:#f85149,color:#c9d1d9
+```
+
+**Ongelma:** Pieni malli (0.5B–7B) tekee toistuvia rakenteellisia virheitä:
+
+| Virhe | Esiintymistiheys | Selitys |
+|-------|:---:|------|
+| Puuttuva import | ~60% | `from datetime import date` unohtuu |
+| SQLite `connect_args` | ~80% | Malli ei muista SQLite-erityisyyttä |
+| Väärä Enum-käyttö | ~50% | Sekoittaa `sqlalchemy.Enum` ja `enum.Enum` |
+| Poetry pyproject.toml:ssa | ~40% | Malli suosii Poetryä vaikka ohje sanoo uv |
+| Testit kopioivat koko appin | ~70% | Malli ei osaa importata, luo uudet reitit |
+
+Retry-loopilla (virhe → uusi yritys virheviestin kanssa) osa korjautuu,
+mutta **sama malli toistaa samoja virheitä** koska ne johtuvat harjoitusdatasta.
+7 tiedoston projekti vaatii 7–14 LLM-kutsua ja 80–120 sekuntia.
+
+### Tapa 2: Rakennuspalaset (template pipeline)
+
+LLM:ltä pyydetään **vain JSON-speksi** — entiteetit, kentät ja tyypit.
+Koodi kootaan mekaanisesti valmiista pohjista joiden rakenne on todistettavasti
+oikein.
+
+```mermaid
+flowchart LR
+    P["Projektin kuvaus"] --> LLM["LLM: JSON-speksi"]
+    LLM --> S["{ entities: [...] }"]
+    S --> T1["Template: models.py"]
+    S --> T2["Template: schemas.py"]
+    S --> T3["Template: main.py"]
+    S --> T4["Template: test_main.py"]
+    S --> T5["Template: Dockerfile"]
+    T1 & T2 & T3 & T4 & T5 --> D["Docker build + pytest"]
+
+    style LLM fill:#1a1e2e,stroke:#d29922,color:#c9d1d9
+    style S fill:#1a1e2e,stroke:#3fb950,color:#c9d1d9
+    style D fill:#1a1e2e,stroke:#58a6ff,color:#c9d1d9
+```
+
+**Idea:** Malli on hyvä päättämään *mitä* (entiteetit, kentät), mutta huono
+muistamaan *miten* (importit, engine setup, testikonfiguraatio). Annetaan
+mallin tehdä se missä se on hyvä, ja hoidetaan loput mekaanisesti.
+
+### LLM:n ainoa tehtävä
+
+Malli tuottaa JSON-rakenteen kuten:
+
+```json
+{
+  "project_name": "todo-app",
+  "entities": [
+    {
+      "name": "Todo",
+      "table_name": "todos",
+      "fields": [
+        {"name": "title", "sa_type": "String(255)", "py_type": "str", "nullable": false},
+        {"name": "due_date", "sa_type": "Date", "py_type": "date | None", "nullable": true},
+        {"name": "status", "sa_type": "String(20)", "py_type": "str", "default": "pending"}
+      ]
+    }
+  ],
+  "extra_imports": ["from datetime import date"]
+}
+```
+
+Tämä on yksinkertainen tehtävä jossa pienikin malli onnistuu luotettavasti:
+entiteettien tunnistus projektin kuvauksesta ja kenttätyyppien valinta.
+
+Speksi sisältää myös **taulujen väliset yhteydet** (relationships):
+
+```json
+{
+  "entities": [
+    {"name": "Author", "table_name": "authors", "fields": [...]},
+    {"name": "Book", "table_name": "books", "fields": [
+      {"name": "title", "sa_type": "String(255)", "py_type": "str", "nullable": false},
+      {"name": "author_id", "sa_type": "Integer", "py_type": "int", "nullable": false}
+    ]}
+  ],
+  "relationships": [
+    {"from": "Book", "field": "author_id", "to": "Author", "type": "many-to-one"}
+  ]
+}
+```
+
+Templateit generoivat yhteyksistä automaattisesti:
+- `ForeignKey('authors.id')` models.py:hin
+- `relationship("Book", back_populates="author")` molempiin suuntiin
+- `BookDetail`-schema jossa author-data mukana
+- `GET /authors/{id}/books/` nested endpoint
+- FK-validointi: 404 jos parent-entiteettiä ei ole
+
+### Architect-agentti: speksin laatu ratkaisee
+
+Arkkitehti on **kriittisin agentti** koko pipelinessa. Jos speksi on hyvä
+(oikeat taulut, kentät, yhteydet), kaikki muu seuraa automaattisesti.
+Jos speksi on huono, templateitkaan eivät pelasta.
+
+Arkkitehtia ohjataan:
+1. **Chain-of-thought**: "Mieti ensin taulut, sitten kentät, sitten yhteydet"
+2. **Domain-esimerkit**: Todo, verkkokauppa, blogi — malli näkee miltä hyvä speksi näyttää
+3. **Anti-patternit**: "Ei turhia ID-kenttiä, ei Enumeita, ei suomenkielisiä nimiä koodissa"
+4. **Yhteyssäännöt**: "Jokainen `_id`-kenttä tarvitsee vastaavan relationship-merkinnän"
+
+Isompi malli (tai API) tässä yhdessä kohdassa parantaa kaikkien projektien laatua
+koska speksi on ainoa paikka jossa LLM:n ymmärrys vaikuttaa.
+
+### Template täyttää loput
+
+Jokainen template on kuin madlib — aukot täytetään speksin datalla:
+
+**models.py template (yksinkertaistettu):**
+```python
+from sqlalchemy import create_engine, Column, Integer, {sa_types}, ForeignKey
+from sqlalchemy.orm import sessionmaker, relationship
+# ... aina samat importit, engine setup, SessionLocal ...
+
+class {entity.name}(Base):
+    __tablename__ = "{entity.table_name}"
+    id = Column(Integer, primary_key=True, index=True)
+    {field.name} = Column({field.sa_type}, nullable={field.nullable})
+    # FK-kentät: ForeignKey + relationship automaattisesti
+    {fk_field} = Column(Integer, ForeignKey('{parent_table}.id'))
+    {parent_lower} = relationship("{Parent}", back_populates="{children}")
+```
+
+Tulos: importit ovat aina oikein, `connect_args` on aina mukana,
+taulujen yhteydet generoituvat oikein, testit importoivat `main.py`:stä eivätkä kopioi sitä.
+
+### Vertailu: mittaustulokset
+
+| | Vapaa generointi | Rakennuspalaset |
+|---|:---:|:---:|
+| LLM-kutsuja | 7–14 | **3** (speksi + requirements + README) |
+| Aika | 80–120s | **~25s** |
+| Syntaksi OK | ~70% | **100%** |
+| Docker build | vaihteleva | **100%** |
+| Pytest läpi | 0% | **100%** |
+| API toimii | ~30% | **100%** |
+| Taulujen yhteydet (FK) | ei koskaan | **100%** |
+| Nested endpointit | ei koskaan | **automaattisesti** |
+
+### Milloin kumpikin toimii
+
+**Rakennuspalaset** kun:
+- Projektin rakenne on tunnettu (FastAPI + SQLAlchemy CRUD)
+- Laatu ja luotettavuus ovat tärkeitä
+- Malli on pieni (0.5B–7B)
+
+**Vapaa generointi** kun:
+- Projektin rakenne on epätavallinen
+- Tarvitaan custom-logiikkaa jota template ei kata
+- Malli on riittävän iso (>70B tai pilvi-API)
+
+Paras lopputulos syntyy yhdistelmällä: **rakennuspalaset perusrakenteelle,
+vapaa generointi business-logiikalle**.
+
+---
+
 ## Laadun parantaminen
 
 ### 1. Isompi malli (suurin vaikutus)

network-poc/frontend/public/download/.build-hash

@@ -0,0 +1 @@
+cf3bf54

network-poc/frontend/public/join.sh

@@ -0,0 +1,73 @@
+#!/bin/bash
+# Kipinä — liitä koneesi laskentaverkkoon
+set -e
+
+HUB_URL="${KIPINA_HUB:-wss://kipina.studio/ws}"
+MODEL="${KIPINA_MODEL:-qwen2.5-coder:3b}"
+
+echo ""
+echo "  ╔══════════════════════════════════════╗"
+echo "  ║  Kipinä Agentic Network — Node Join  ║"
+echo "  ╚══════════════════════════════════════╝"
+echo ""
+
+# 1. Ollama
+if command -v ollama &>/dev/null; then
+    echo "  ✓ Ollama löytyi: $(ollama --version 2>/dev/null || echo 'asennettu')"
+else
+    echo "  Ollama ei ole asennettu."
+    echo ""
+    read -p "  Asennetaanko Ollama? (k/e) " -n 1 -r; echo
+    if [[ $REPLY =~ ^[Kk]$ ]]; then
+        echo "  Asennetaan Ollama..."
+        curl -fsSL https://ollama.ai/install.sh | sh
+    else
+        echo "  Ollama vaaditaan laskentaan. Asenna: https://ollama.ai"
+        exit 1
+    fi
+fi
+
+# 2. Varmistetaan että Ollama on käynnissä
+if ! curl -s http://localhost:11434/api/tags &>/dev/null; then
+    echo "  Käynnistetään Ollama..."
+    ollama serve &>/dev/null &
+    sleep 3
+    if ! curl -s http://localhost:11434/api/tags &>/dev/null; then
+        echo "  ✗ Ollama ei käynnistynyt. Aja: ollama serve"
+        exit 1
+    fi
+fi
+echo "  ✓ Ollama käynnissä"
+
+# 3. Malli
+if ollama list 2>/dev/null | grep -q "$MODEL"; then
+    echo "  ✓ Malli $MODEL ladattu"
+else
+    echo "  Ladataan malli $MODEL..."
+    ollama pull "$MODEL"
+fi
+
+# 4. Native-node
+echo ""
+echo "  Yhdistetään hubiin: $HUB_URL"
+echo "  Malli: $MODEL"
+echo "  Ctrl+C pysäyttää"
+echo ""
+
+# Tarkistetaan onko native-node käännetty
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+NATIVE_BIN="$SCRIPT_DIR/target/release/native-node"
+
+if [ -f "$NATIVE_BIN" ]; then
+    HUB_URL="$HUB_URL" OLLAMA_MODEL="$MODEL" "$NATIVE_BIN"
+elif command -v cargo &>/dev/null && [ -f "$SCRIPT_DIR/native-node/Cargo.toml" ]; then
+    echo "  Käännetään native-node..."
+    cd "$SCRIPT_DIR"
+    cargo build --release -p native-node --no-default-features 2>&1 | tail -1
+    HUB_URL="$HUB_URL" OLLAMA_MODEL="$MODEL" "$NATIVE_BIN"
+else
+    echo "  ✗ native-node binääriä ei löydy eikä Rust ole asennettu."
+    echo "  Asenna Rust: curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh"
+    echo "  Tai lataa valmis binääri: https://kipina.studio/download"
+    exit 1
+fi

network-poc/frontend/public/kipina-node

@@ -0,0 +1,135 @@
+#!/bin/bash
+# Kipinä Node — lataa oikea binääri ja käynnistä
+set -e
+
+BASE_URL="https://kipina.studio/download"
+HUB_URL="${KIPINA_HUB:-wss://kipina.studio/ws}"
+OLLAMA_URL="${OLLAMA_URL:-http://localhost:11434}"
+
+# Tunnista OS ja arkkitehtuuri
+OS=$(uname -s | tr '[:upper:]' '[:lower:]')
+ARCH=$(uname -m)
+
+case "$OS-$ARCH" in
+    darwin-arm64)  BINARY="kipina-node-macos-arm64" ;;
+    darwin-x86_64) BINARY="kipina-node-macos-arm64" ;; # Rosetta
+    linux-x86_64)  BINARY="kipina-node-linux-x86_64" ;;
+    linux-aarch64) BINARY="kipina-node-linux-arm64" ;;
+    *) echo "Ei tuettu: $OS-$ARCH"; exit 1 ;;
+esac
+
+echo ""
+echo "  ╔══════════════════════════════════════╗"
+echo "  ║       Kipinä Agentic Node            ║"
+echo "  ╚══════════════════════════════════════╝"
+echo ""
+echo "  OS: $OS ($ARCH)"
+echo ""
+
+# Etsi Ollama-instanssit
+CANDIDATES=(
+    "http://localhost:11434"
+    "http://127.0.0.1:11434"
+    "http://ollama:11434"
+    "http://host.docker.internal:11434"
+)
+
+# Lisää OLLAMA_URL listaan jos asetettu ja ei jo mukana
+if [ -n "$OLLAMA_URL" ]; then
+    ALREADY=false
+    for c in "${CANDIDATES[@]}"; do
+        [ "$c" = "$OLLAMA_URL" ] && ALREADY=true
+    done
+    $ALREADY || CANDIDATES=("$OLLAMA_URL" "${CANDIDATES[@]}")
+fi
+
+echo "  Etsitään Ollama-instansseja..."
+FOUND=()
+for url in "${CANDIDATES[@]}"; do
+    if curl -s --connect-timeout 1 "$url/api/tags" &>/dev/null; then
+        FOUND+=("$url")
+    fi
+done
+
+if [ ${#FOUND[@]} -eq 0 ]; then
+    # Ei löytynyt — yritä käynnistää lokaali
+    if command -v ollama &>/dev/null; then
+        echo "  Käynnistetään Ollama..."
+        ollama serve &>/dev/null &
+        sleep 3
+        if curl -s --connect-timeout 1 "http://localhost:11434/api/tags" &>/dev/null; then
+            OLLAMA_URL="http://localhost:11434"
+            echo "  ✓ Ollama käynnistetty ($OLLAMA_URL)"
+        else
+            echo "  ✗ Ollaman käynnistys epäonnistui."
+            exit 1
+        fi
+    else
+        echo ""
+        echo "  ✗ Ollamaa ei löytynyt."
+        echo "    Kontti/remote: OLLAMA_URL=http://HOST:11434 ./kipina-node"
+        echo "    Asenna:        curl -fsSL https://ollama.ai/install.sh | sh"
+        exit 1
+    fi
+elif [ ${#FOUND[@]} -eq 1 ]; then
+    OLLAMA_URL="${FOUND[0]}"
+    echo "  ✓ Ollama löytyi: $OLLAMA_URL"
+else
+    echo ""
+    echo "  Löytyi ${#FOUND[@]} Ollama-instanssia:"
+    echo ""
+    for i in "${!FOUND[@]}"; do
+        echo "    $((i+1))) ${FOUND[$i]}"
+    done
+    echo ""
+    read -p "  Valitse [1-${#FOUND[@]}]: " -r CHOICE
+    if [[ "$CHOICE" =~ ^[0-9]+$ ]] && [ "$CHOICE" -ge 1 ] && [ "$CHOICE" -le ${#FOUND[@]} ]; then
+        OLLAMA_URL="${FOUND[$((CHOICE-1))]}"
+    else
+        OLLAMA_URL="${FOUND[0]}"
+        echo "  Käytetään oletusta: $OLLAMA_URL"
+    fi
+    echo "  ✓ Valittu: $OLLAMA_URL"
+fi
+
+echo ""
+echo "  Hub: $HUB_URL"
+echo "  Ollama: $OLLAMA_URL"
+if [ -n "$KIPINA_MODEL" ]; then
+    echo "  Malli: $KIPINA_MODEL (Ympäristömuuttujasta)"
+fi
+
+# Binäärin automaattinen päivitys — vertaa build-hashia palvelimeen
+BIN_PATH="./kipina-node-bin"
+HASH_PATH="./kipina-node-bin.hash"
+
+REMOTE_HASH=$(curl -sSL "$BASE_URL/.build-hash?v=$(date +%s)" 2>/dev/null | tr -d '[:space:]')
+LOCAL_HASH=""
+[ -f "$HASH_PATH" ] && LOCAL_HASH=$(cat "$HASH_PATH" | tr -d '[:space:]')
+
+if [ -f "$BIN_PATH" ] && [ -n "$REMOTE_HASH" ] && [ "$REMOTE_HASH" = "$LOCAL_HASH" ]; then
+    echo "  ✓ Binääri ajan tasalla (versio: $LOCAL_HASH)"
+else
+    if [ -f "$BIN_PATH" ]; then
+        echo "  ↻ Uusi versio saatavilla ($LOCAL_HASH → $REMOTE_HASH)"
+    else
+        echo "  Ladataan $BINARY..."
+    fi
+    rm -f "$BIN_PATH"
+    curl -sSL "$BASE_URL/$BINARY?v=$(date +%s)" -o "$BIN_PATH"
+    chmod +x "$BIN_PATH"
+    echo "$REMOTE_HASH" > "$HASH_PATH"
+    echo "  ✓ Päivitetty versioon $REMOTE_HASH"
+fi
+
+echo ""
+echo "  ✓ Siirrytään Kipinä Noden hallintaan..."
+echo "    Ctrl+C pysäyttää"
+echo ""
+
+if [ -n "$KIPINA_MODEL" ]; then
+    export OLLAMA_MODEL="$KIPINA_MODEL"
+fi
+export HUB_URL="$HUB_URL"
+export OLLAMA_URL="$OLLAMA_URL"
+exec "$BIN_PATH"

network-poc/frontend/public/templates/data-analytics.json

@@ -0,0 +1,33 @@
+{
+  "name": "Data Analytics Pipeline",
+  "description": "ETL, analysis, and visualization with Docker (MariaDB + Jupyter)",
+  "keywords": ["data", "analytics", "csv", "etl", "visualization", "statistics", "dashboard", "jupyter", "pandas", "matplotlib"],
+  "files": {
+    "etl.py": {
+      "description": "Data loading, cleaning, and transformation",
+      "example": "import pandas as pd\nfrom pathlib import Path\nfrom sqlalchemy import create_engine\n\nDB_URL = \"mysql+pymysql://root:secret@localhost:3306/analytics\"\nengine = create_engine(DB_URL)\n\ndef load_csv(path: str) -> pd.DataFrame:\n    df = pd.read_csv(path)\n    print(f\"Loaded {len(df)} rows from {path}\")\n    return df\n\ndef clean(df: pd.DataFrame) -> pd.DataFrame:\n    df = df.dropna(subset=[\"x\", \"y\"])\n    df = df[(df[\"x\"] >= 0) & (df[\"y\"] >= 0)]  # Remove outliers\n    df[\"timestamp\"] = pd.to_datetime(df[\"timestamp\"])\n    return df.sort_values(\"timestamp\").reset_index(drop=True)\n\ndef to_database(df: pd.DataFrame, table: str):\n    df.to_sql(table, engine, if_exists=\"replace\", index=False)\n    print(f\"Wrote {len(df)} rows to {table}\")\n\nif __name__ == \"__main__\":\n    for csv_file in sorted(Path(\"data\").glob(\"*.csv\")):\n        df = load_csv(str(csv_file))\n        df = clean(df)\n        to_database(df, \"measurements\")",
+      "instructions": "Write the ETL pipeline:\n- Load CSV files from data/ directory using pandas\n- Clean: remove nulls, filter outliers, parse timestamps\n- Transform: convert units, compute derived columns\n- Load into MariaDB via SQLAlchemy\n- Make it runnable as a standalone script"
+    },
+    "analysis.py": {
+      "description": "Statistical analysis and metrics computation",
+      "example": "import pandas as pd\nfrom sqlalchemy import create_engine\n\nDB_URL = \"mysql+pymysql://root:secret@localhost:3306/analytics\"\nengine = create_engine(DB_URL)\n\ndef load_data() -> pd.DataFrame:\n    return pd.read_sql(\"SELECT * FROM measurements\", engine)\n\ndef summary_stats(df: pd.DataFrame) -> dict:\n    return {\n        \"total_rows\": len(df),\n        \"date_range\": f\"{df['timestamp'].min()} to {df['timestamp'].max()}\",\n        \"unique_entities\": df[\"entity_id\"].nunique(),\n    }\n\ndef hourly_distribution(df: pd.DataFrame) -> pd.DataFrame:\n    df[\"hour\"] = df[\"timestamp\"].dt.hour\n    return df.groupby(\"hour\").size().reset_index(name=\"count\")\n\nif __name__ == \"__main__\":\n    df = load_data()\n    stats = summary_stats(df)\n    for k, v in stats.items():\n        print(f\"{k}: {v}\")",
+      "instructions": "Write analysis functions:\n- Load cleaned data from MariaDB\n- Compute summary statistics (counts, date ranges, distributions)\n- Time-based analysis (hourly, daily, weekly patterns)\n- Group-level metrics (per entity, per zone)\n- Return DataFrames and dicts suitable for visualization"
+    },
+    "visualize.py": {
+      "description": "Charts and visualizations with matplotlib",
+      "example": "import matplotlib.pyplot as plt\nimport pandas as pd\nfrom analysis import load_data, hourly_distribution\n\ndef plot_heatmap(df: pd.DataFrame, title: str, output: str):\n    fig, ax = plt.subplots(figsize=(12, 8))\n    scatter = ax.scatter(df[\"x\"], df[\"y\"], c=df[\"density\"], cmap=\"hot\", alpha=0.5, s=2)\n    ax.set_title(title)\n    ax.set_xlabel(\"x\")\n    ax.set_ylabel(\"y\")\n    ax.invert_yaxis()\n    plt.colorbar(scatter, label=\"Density\")\n    plt.tight_layout()\n    plt.savefig(output, dpi=150)\n    print(f\"Saved {output}\")\n\ndef plot_bar(df: pd.DataFrame, x: str, y: str, title: str, output: str):\n    fig, ax = plt.subplots(figsize=(10, 5))\n    ax.bar(df[x], df[y], color=\"steelblue\")\n    ax.set_title(title)\n    ax.set_xlabel(x)\n    ax.set_ylabel(y)\n    plt.tight_layout()\n    plt.savefig(output, dpi=150)\n\nif __name__ == \"__main__\":\n    df = load_data()\n    hourly = hourly_distribution(df)\n    plot_bar(hourly, \"hour\", \"count\", \"Hourly Distribution\", \"output/hourly.png\")",
+      "instructions": "Write visualization functions:\n- Import analysis functions for data\n- Heatmaps, bar charts, line charts as appropriate\n- Save figures to output/ directory (PNG, 150 DPI)\n- Use matplotlib with clear titles, labels, colorbars\n- Make it runnable as standalone to generate all charts"
+    },
+    "docker-compose.yml": {
+      "description": "Docker Compose stack for database and Jupyter",
+      "example": "services:\n  db:\n    image: mariadb:11\n    environment:\n      MYSQL_ROOT_PASSWORD: secret\n      MYSQL_DATABASE: analytics\n    ports:\n      - \"3306:3306\"\n    volumes:\n      - db_data:/var/lib/mysql\n\n  jupyter:\n    image: jupyter/scipy-notebook:latest\n    ports:\n      - \"8888:8888\"\n    volumes:\n      - .:/home/jovyan/work\n    environment:\n      JUPYTER_TOKEN: kipina\n    depends_on:\n      - db\n\nvolumes:\n  db_data:",
+      "instructions": "Write docker-compose.yml:\n- MariaDB service with persistent volume\n- JupyterLab service with project mounted\n- Correct environment variables\n- Port mappings for local development\n- Write ONLY the YAML, no explanations"
+    },
+    "pyproject.toml": {
+      "description": "Project dependencies",
+      "example": "[project]\nname = \"analytics\"\nversion = \"0.1.0\"\nrequires-python = \">=3.11\"\ndependencies = [\n    \"pandas\",\n    \"matplotlib\",\n    \"sqlalchemy\",\n    \"pymysql\",\n]\n\n[project.scripts]\netl = \"python etl.py\"\nanalyze = \"python analysis.py\"\nvisualize = \"python visualize.py\"",
+      "instructions": "Use [project] format (PEP 621). List all data science dependencies. Add scripts for ETL, analysis, and visualization."
+    }
+  },
+  "order": ["etl.py", "analysis.py", "visualize.py", "docker-compose.yml", "pyproject.toml"]
+}

network-poc/frontend/public/templates/fastapi-crud.json

@@ -1,6 +1,7 @@
 {
   "name": "FastAPI CRUD",
   "description": "REST API with SQLite database",
+  "keywords": ["api", "rest", "crud", "endpoint", "fastapi", "web", "backend", "server", "database", "sqlite"],
   "files": {
     "models.py": {
       "description": "SQLAlchemy models, engine, and session",

network-poc/frontend/src/components/AgentBar.astro

@@ -1,37 +1,37 @@
-<!-- Agenttigalleria + konfigurointipaneeli -->
+<!-- Agent gallery + configuration panel -->
 <div style="display:flex;gap:16px;padding:10px 0;align-items:flex-start">
     <!-- Agenttilista (drag & drop) -->
     <div id="agent-bar" style="display:flex;gap:6px;align-items:flex-end;flex-wrap:wrap">
         <!-- Renderöidään JS:stä -->
     </div>
-    <!-- + Lisää agentti -->
+    <!-- + Add agent -->
-    <div id="add-agent-btn" class="agent-avatar" onclick="addCustomAgent()" title="Lisää oma agentti" style="opacity:0.4">
+    <div id="add-agent-btn" class="agent-avatar" onclick="addCustomAgent()" title="Add custom agent" style="opacity:0.4">
         <div style="width:48px;height:48px;border-radius:50%;border:2px dashed var(--border);display:flex;align-items:center;justify-content:center;font-size:24px;color:var(--border)">+</div>
-        <span style="font-size:10px;color:#8b949e;text-align:center;display:block">Lisää</span>
+        <span style="font-size:10px;color:#8b949e;text-align:center;display:block">Add</span>
     </div>
 </div>
 
-<!-- Agentin konfigurointipaneeli (avautuu klikkaamalla avataria) -->
+<!-- Agent configuration panel (opens clicking avatar) -->
 <div id="agent-config" style="display:none;background:var(--panel);border:1px solid var(--border);border-radius:6px;padding:16px;margin-bottom:10px">
     <div style="display:flex;justify-content:space-between;align-items:center;margin-bottom:12px">
         <div style="display:flex;align-items:center;gap:10px">
             <img id="config-avatar" src="" style="width:40px;height:40px;border-radius:50%">
             <div>
-                <input id="config-name" style="background:transparent;border:none;color:var(--text);font-size:16px;font-weight:600;outline:none;width:200px" placeholder="Agentin nimi">
+                <input id="config-name" style="background:transparent;border:none;color:var(--text);font-size:16px;font-weight:600;outline:none;width:200px" placeholder="Agent Name">
                 <div id="config-role" style="font-size:11px;color:#8b949e"></div>
             </div>
         </div>
         <div style="display:flex;gap:6px">
-            <button class="btn btn-red" onclick="deleteAgent()" title="Poista agentti">Poista</button>
+            <button class="btn btn-red" onclick="deleteAgent()" title="Delete agent">Delete</button>
-            <button class="btn btn-muted" onclick="closeAgentConfig()">Sulje</button>
+            <button class="btn btn-muted" onclick="closeAgentConfig()">Close</button>
         </div>
     </div>
 
-    <!-- Malli -->
+    <!-- Model -->
     <div style="margin-bottom:10px">
-        <label style="font-size:12px;color:#8b949e;display:block;margin-bottom:4px">Kielimalli</label>
+        <label style="font-size:12px;color:#8b949e;display:block;margin-bottom:4px">Model</label>
         <select id="config-model" style="background:var(--bg);color:var(--text);border:1px solid var(--border);border-radius:4px;padding:6px 10px;font-size:13px;width:100%">
-            <option value="qwen-coder">Qwen2.5-Coder:0.5B (selain)</option>
+            <option value="qwen-coder">Qwen2.5-Coder:0.5B (browser)</option>
             <option value="qwen-coder-3b">Qwen2.5-Coder:3B (Ollama)</option>
             <option value="qwen2.5-coder:7b">Qwen2.5-Coder:7B (Ollama)</option>
             <option value="qwen2.5-coder:1.5b">Qwen2.5-Coder:1.5B (Ollama)</option>
@@ -39,41 +39,41 @@
     </div>
 
     <!-- System prompt -->
-    <div style="margin-bottom:10px" title="Agentin perusohje joka lähetetään kielimallille jokaisessa pyynnössä.&#10;&#10;Hyvän promptin rakenne:&#10;1. Rooli: 'You are an expert...'&#10;2. Säännöt: RULES/CRITICAL RULES listana&#10;3. Esimerkit: EXAMPLE OUTPUT&#10;4. Kiellot: NEVER-lista&#10;&#10;Vinkki: käytä englantia — malli ymmärtää sen paremmin ja se kuluttaa vähemmän tokeneita.">
+    <div style="margin-bottom:10px" title="System prompt sent to the LLM on every request.&#10;&#10;Good prompt structure:&#10;1. Role: 'You are an expert...'&#10;2. Rules: RULES/CRITICAL RULES as list&#10;3. Examples: EXAMPLE OUTPUT&#10;4. Restrictions: NEVER-list&#10;">
         <label style="font-size:12px;color:#8b949e;display:block;margin-bottom:4px;cursor:help">System prompt 💡</label>
-        <textarea id="config-prompt" style="width:100%;background:var(--bg);color:var(--text);border:1px solid var(--border);border-radius:4px;padding:8px;font-size:13px;font-family:'Courier New',monospace;resize:vertical;overflow:hidden;min-height:60px" placeholder="Kuvaa agentin rooli ja käyttäytyminen..."></textarea>
+        <textarea id="config-prompt" style="width:100%;background:var(--bg);color:var(--text);border:1px solid var(--border);border-radius:4px;padding:8px;font-size:13px;font-family:'Courier New',monospace;resize:vertical;overflow:hidden;min-height:60px" placeholder="Describe the agent's role and behavior..."></textarea>
     </div>
 
-    <!-- Sampling-parametrit -->
+    <!-- Sampling Parameters -->
     <div style="margin-bottom:10px">
-        <label style="font-size:12px;color:#8b949e;display:block;margin-bottom:8px">Sampling-parametrit</label>
+        <label style="font-size:12px;color:#8b949e;display:block;margin-bottom:8px">Sampling Parameters</label>
         <div style="display:grid;grid-template-columns:1fr 1fr;gap:10px">
-            <div title="Kontrolloi 'luovuutta'. Matala arvo (0.2-0.4) tuottaa ennustettavaa, toistettavaa koodia — hyvä testaajille ja reviewereille. Keskiarvo (0.6-0.8) on paras koodin generointiin. Korkea arvo (1.0+) lisää vaihtelua mutta myös virheitä.&#10;&#10;Suositus:&#10;• Manageri: 0.5 (tarkat tiedostolistat)&#10;• Koodari: 0.7 (toimiva koodi + vaihtelu)&#10;• Testaaja: 0.3 (deterministinen arviointi)">
+            <div title="Controls 'creativity'. Low value (0.2-0.4) produces predictable, repeatable code — good for testers and reviewers. Medium value (0.6-0.8) is best for generating code. High value (1.0+) adds variation but also errors.&#10;&#10;Recommendation:&#10;• Manager: 0.5 (precise file lists)&#10;• Coder: 0.7 (working code + variation)&#10;• Tester: 0.3 (deterministic evaluation)">
                 <label style="font-size:11px;color:#8b949e;cursor:help">Temperature 💡 <span id="config-temp-val" style="color:var(--accent);float:right">0.7</span></label>
                 <input type="range" id="config-temperature" min="0" max="1.5" step="0.1" value="0.7" style="width:100%;accent-color:var(--accent)">
-                <div style="font-size:10px;color:#30363d">0=tarkka · 0.7=oletus · 1.5=luova</div>
+                <div style="font-size:10px;color:#30363d">0=strict · 0.7=default · 1.5=creative</div>
             </div>
-            <div title="Vastauksen maksimipituus tokeneina (~1 token ≈ 4 merkkiä).&#10;&#10;Suositus:&#10;• Manageri: 256-512 (lyhyet tiedostolistat)&#10;• Koodari: 1024-2048 (täydet tiedostot, CRUD-endpointit)&#10;• Testaaja: 256-512 (lyhyet arvioinnit)&#10;&#10;Jos koodi katkeaa kesken, nosta tätä. Jos malli tuottaa turhaa toistoa, laske.">
+            <div title="Maximum response length in tokens (~1 token ≈ 4 chars).&#10;&#10;Recommendation:&#10;• Manager: 256-512 (short lists)&#10;• Coder: 1024-2048 (full files, CRUD endpoints)&#10;• Tester: 256-512 (short evaluations)&#10;&#10;If code cuts off early, increase this.">
                 <label style="font-size:11px;color:#8b949e;cursor:help">Max tokens 💡 <span id="config-maxtok-val" style="color:var(--accent);float:right">1024</span></label>
                 <input type="range" id="config-maxtokens" min="64" max="4096" step="64" value="1024" style="width:100%;accent-color:var(--accent)">
-                <div style="font-size:10px;color:#30363d">Vastauksen maksimipituus</div>
+                <div style="font-size:10px;color:#30363d">Maximum response length</div>
             </div>
-            <div title="Montako todennäköisintä tokenia huomioidaan valinnassa. Pieni arvo (1-10) tekee vastauksesta deterministisen. Suuri arvo (50-100) sallii harvinaisempia sanoja.&#10;&#10;Suositus:&#10;• Boilerplate-koodi: 20-30 (tutut patternit)&#10;• Yleiskoodi: 40 (hyvä oletus)&#10;• Luova teksti: 60-80&#10;&#10;Yleensä ei tarvitse muuttaa oletuksesta.">
+            <div title="How many most probable tokens are considered. Low value (1-10) makes response deterministic. High value (50-100) allows rarer words.&#10;&#10;Recommendation:&#10;• Boilerplate code: 20-30 (familiar patterns)&#10;• General code: 40 (good default)&#10;• Creative text: 60-80">
                 <label style="font-size:11px;color:#8b949e;cursor:help">Top-K 💡 <span id="config-topk-val" style="color:var(--accent);float:right">40</span></label>
                 <input type="range" id="config-topk" min="1" max="100" step="1" value="40" style="width:100%;accent-color:var(--accent)">
-                <div style="font-size:10px;color:#30363d">1=greedy · 40=oletus · 100=laaja</div>
+                <div style="font-size:10px;color:#30363d">1=greedy · 40=default · 100=wide</div>
             </div>
-            <div title="Vähentää jo tuotettujen sanojen todennäköisyyttä. Estää mallia toistamasta samaa lausetta. Liian korkea arvo (>1.5) voi rikkoa koodin koska samat avainsanat (return, if, def) ovat tarpeellisia.&#10;&#10;Suositus:&#10;• Koodi: 1.1-1.2 (lievä, sallii toiston)&#10;• Teksti: 1.15-1.3 (vahvempi)&#10;• Review: 1.0-1.1 (ei rangaistusta, lyhyet vastaukset)">
+            <div title="Reduces the probability of already generated words. Prevents model from repeating same sentences. Too high value (>1.5) can break code because common keywords (return, if, def) are necessary.&#10;&#10;Recommendation:&#10;• Code: 1.1-1.2 (mild, allows repetition)&#10;• Text: 1.15-1.3 (stronger penalty)&#10;• Review: 1.0-1.1 (no penalty, short answers)">
                 <label style="font-size:11px;color:#8b949e;cursor:help">Repetition penalty 💡 <span id="config-rep-val" style="color:var(--accent);float:right">1.15</span></label>
                 <input type="range" id="config-repeat" min="1.0" max="2.0" step="0.05" value="1.15" style="width:100%;accent-color:var(--accent)">
-                <div style="font-size:10px;color:#30363d">1.0=ei · 1.15=oletus · 2.0=vahva</div>
+                <div style="font-size:10px;color:#30363d">1.0=none · 1.15=default · 2.0=strong</div>
             </div>
         </div>
     </div>
 
-    <!-- Pipeline-järjestys -->
+    <!-- Pipeline order -->
     <div>
-        <label style="font-size:12px;color:#8b949e;display:block;margin-bottom:4px">Pipeline-järjestys <span style="color:var(--border)">(vedä järjestääksesi)</span></label>
+        <label style="font-size:12px;color:#8b949e;display:block;margin-bottom:4px">Pipeline Order <span style="color:var(--border)">(drag to sort)</span></label>
         <div id="config-pipeline" style="display:flex;gap:4px;flex-wrap:wrap"></div>
     </div>
 </div>

network-poc/frontend/src/components/Editor.astro

@@ -1,8 +1,11 @@
 <!-- Monaco Editor paneeli -->
 <div id="panel-editor" class="panel">
-    <div style="display:flex;height:calc(100vh - 200px);gap:0;border:1px solid var(--border);border-radius:6px;overflow:hidden">
+    <div style="display:flex;flex:1;min-height:0;gap:0;border:1px solid var(--border);border-radius:6px;overflow:hidden">
-        <div id="editor-filetree" style="width:200px;min-width:150px;background:var(--bg);border-right:1px solid var(--border);overflow-y:auto;font-family:'Courier New',monospace;font-size:13px">
+        <div id="editor-filetree" style="width:200px;min-width:150px;background:var(--bg);border-right:1px solid var(--border);overflow:auto;resize:horizontal;font-family:'Courier New',monospace;font-size:13px">
-            <div style="padding:10px 12px;color:#8b949e;font-size:11px;text-transform:uppercase;letter-spacing:0.5px;border-bottom:1px solid var(--border)">Tiedostot</div>
+            <div style="padding:10px 12px;color:#8b949e;font-size:11px;display:flex;justify-content:space-between;align-items:center;text-transform:uppercase;letter-spacing:0.5px;border-bottom:1px solid var(--border)">
+                <span>Tiedostot</span>
+                <button class="btn btn-green" style="padding:2px 6px;font-size:10px" onclick="downloadProjectZip()">.ZIP</button>
+            </div>
             <div id="editor-file-list" style="padding:4px 0">
                 <div style="padding:8px 16px;color:#8b949e;font-size:12px">Generoi projekti:<br><code style="color:var(--accent)">kpn project "..."</code></div>
             </div>

network-poc/frontend/src/components/Settings.astro

@@ -58,6 +58,49 @@
             </select>
         </div>
 
+        <!-- Pipeline-rajoitteet -->
+        <div class="settings-section">
+            <h3 class="settings-title">Pipeline-rajoitteet</h3>
+            <p class="settings-desc">Projektin generoinnin rajat. Suuremmat arvot = rikkaampi output, hitaampi suoritus.</p>
+            <div class="settings-grid">
+                <div>
+                    <label class="settings-label">Client: max sanat <span id="set-plc-words-val" class="settings-val">400</span></label>
+                    <input type="range" id="set-plc-words" min="100" max="800" step="50" value="400" class="settings-slider">
+                    <div class="settings-hint">Vaatimusmäärittelyn maksimipituus sanoina</div>
+                </div>
+                <div>
+                    <label class="settings-label">Client: max ominaisuudet <span id="set-plc-feats-val" class="settings-val">8</span></label>
+                    <input type="range" id="set-plc-feats" min="3" max="15" step="1" value="8" class="settings-slider">
+                    <div class="settings-hint">Montako ominaisuutta vaatimuksiin</div>
+                </div>
+                <div>
+                    <label class="settings-label">Manager: max tiedostot <span id="set-plc-mfiles-val" class="settings-val">8</span></label>
+                    <input type="range" id="set-plc-mfiles" min="3" max="15" step="1" value="8" class="settings-slider">
+                    <div class="settings-hint">Managerin suunnittelemien tiedostojen yläraja</div>
+                </div>
+                <div>
+                    <label class="settings-label">Vapaa tila: max tiedostot <span id="set-plc-ffiles-val" class="settings-val">8</span></label>
+                    <input type="range" id="set-plc-ffiles" min="3" max="15" step="1" value="8" class="settings-slider">
+                    <div class="settings-hint">Tiedostoraja kun ei mallipohjaa</div>
+                </div>
+                <div>
+                    <label class="settings-label">Review-kierrokset <span id="set-plc-review-val" class="settings-val">3</span></label>
+                    <input type="range" id="set-plc-review" min="1" max="5" step="1" value="3" class="settings-slider">
+                    <div class="settings-hint">Katselmointi-korjaus-syklien max määrä</div>
+                </div>
+                <div>
+                    <label class="settings-label">Terminaali: max rivit <span id="set-plc-term-val" class="settings-val">300</span></label>
+                    <input type="range" id="set-plc-term" min="50" max="1000" step="50" value="300" class="settings-slider">
+                    <div class="settings-hint">Terminaalin näyttämien rivien yläraja</div>
+                </div>
+                <div>
+                    <label class="settings-label">CrewAI: prompt-rivit <span id="set-plc-crew-val" class="settings-val">50</span></label>
+                    <input type="range" id="set-plc-crew" min="10" max="200" step="10" value="50" class="settings-slider">
+                    <div class="settings-hint">tasks.yaml:n promptin max rivimäärä</div>
+                </div>
+            </div>
+        </div>
+
         <!-- Reset -->
         <div style="margin-top:24px;padding-top:16px;border-top:1px solid var(--border)">
             <button class="btn btn-red" onclick="resetSettings()" style="padding:6px 16px">Palauta oletukset</button>

network-poc/frontend/src/components/StatusBar.astro

@@ -10,6 +10,39 @@
         <span id="compute-dot" class="status-dot" style="background:#30363d"></span>
         <span style="color:#8b949e">Laskenta:</span>
         <span id="compute-label" style="color:#8b949e">—</span>
-        <button id="compute-btn" class="btn btn-accent" title="Käynnistä kielimalli">Alusta</button>
+        <button id="compute-btn" class="btn btn-accent" title="Käynnistä kielimalli selaimessa">Alusta</button>
+    </span>
+    <span class="status-separator">│</span>
+    <span class="status-group">
+        <button id="join-btn" class="btn btn-green" onclick="showJoinDialog()" title="Liitä oma koneesi laskentaverkkoon (natiivi, nopea)">+ Liitä koneesi</button>
     </span>
 </div>
+
+<!-- Join-dialogi -->
+<div id="join-dialog" style="display:none;margin-top:8px;padding:16px;background:var(--panel);border:1px solid var(--border);border-radius:6px;font-size:14px">
+    <div style="display:flex;justify-content:space-between;align-items:center;margin-bottom:12px">
+        <span style="color:#e6edf3;font-weight:600;font-size:16px">Liitä koneesi laskentaverkkoon</span>
+        <button onclick="document.getElementById('join-dialog').style.display='none'" style="background:none;border:none;color:#8b949e;cursor:pointer;font-size:18px">✕</button>
+    </div>
+    <p style="color:#8b949e;margin-bottom:16px">Koneesi suorittaa tehtäviä ~10-50x nopeammin kuin selainlaskenta. Kaksi vaihetta:</p>
+
+    <!-- Vaihe 1: Ollama -->
+    <div style="margin-bottom:14px;padding:12px;background:var(--bg);border-radius:4px;border-left:3px solid var(--accent)">
+        <div style="color:#e6edf3;font-weight:600;margin-bottom:6px">1. Asenna Ollama <span style="color:#8b949e;font-weight:normal">(kielimallimoottori)</span></div>
+        <div style="display:flex;gap:6px;align-items:center;margin-bottom:6px">
+            <code style="flex:1;background:#010409;padding:8px 12px;border-radius:4px;color:var(--green);font-family:'Courier New',monospace;font-size:13px;user-select:all">curl -fsSL https://ollama.ai/install.sh | sh</code>
+            <button onclick="navigator.clipboard.writeText('curl -fsSL https://ollama.ai/install.sh | sh');this.textContent='✓';setTimeout(()=>this.textContent='Kopioi',1500)" class="btn btn-accent" style="padding:6px 10px">Kopioi</button>
+        </div>
+        <div style="color:#8b949e;font-size:12px">macOS: <code style="color:var(--accent)">brew install ollama</code> · Windows: <a href="https://ollama.ai/download" target="_blank" style="color:var(--accent)">ollama.ai/download</a> · Jos jo asennettu → siirry vaiheeseen 2.</div>
+    </div>
+
+    <!-- Vaihe 2: Kipinä-node -->
+    <div style="padding:12px;background:var(--bg);border-radius:4px;border-left:3px solid var(--green)">
+        <div style="color:#e6edf3;font-weight:600;margin-bottom:6px">2. Käynnistä Kipinä-node</div>
+        <div style="display:flex;gap:6px;align-items:center;margin-bottom:6px">
+            <code style="flex:1;background:#010409;padding:8px 12px;border-radius:4px;color:var(--green);font-family:'Courier New',monospace;font-size:13px;user-select:all">curl -sSL "https://kipina.studio/kipina-node?v=$(date +%s)" -o kipina-node && chmod +x kipina-node && ./kipina-node</code>
+            <button onclick="navigator.clipboard.writeText('curl -sSL &quot;https://kipina.studio/kipina-node?v=$(date +%s)&quot; -o kipina-node && chmod +x kipina-node && ./kipina-node');this.textContent='✓';setTimeout(()=>this.textContent='Kopioi',1500)" class="btn btn-green" style="padding:6px 10px">Kopioi</button>
+        </div>
+        <div style="color:#8b949e;font-size:12px">Lataa kielimallin (~2GB) automaattisesti ensimmäisellä kerralla. Ctrl+C pysäyttää.</div>
+    </div>
+</div>

network-poc/frontend/src/styles/global.css

@@ -1,3 +1,4 @@
+/* Oletusvärit — ylikirjoitetaan teemalla */
 :root {
     --bg: #0d1117;
     --panel: #161b22;
@@ -8,38 +9,100 @@
     --red: #f85149;
     --purple: #a371f7;
     --border: #30363d;
+    --hero-accent: #ff6b00;
+    --hero-glow: rgba(255, 107, 0, 0.15);
+}
+
+/* Gecko — lämmin kulta/oranssi (kipina.tech) */
+[data-theme="gecko"] {
+    --bg: #0a0500;
+    --panel: #1f1000;
+    --text: #fff5e6;
+    --accent: #ff7b00;
+    --green: #3fb950;
+    --yellow: #ffae00;
+    --red: #f85149;
+    --purple: #ff9d4d;
+    --border: rgba(255, 174, 0, 0.2);
+    --hero-accent: #ff7b00;
+    --hero-glow: rgba(255, 123, 0, 0.15);
+}
+
+/* Forge — kyber-sininen/syaani (kipina.tech) */
+[data-theme="forge"] {
+    --bg: #060b11;
+    --panel: #121e2d;
+    --text: #e0f2fe;
+    --accent: #00e5ff;
+    --green: #3fb950;
+    --yellow: #ff5e3a;
+    --red: #f85149;
+    --purple: #7dd3fc;
+    --border: rgba(0, 229, 255, 0.15);
+    --hero-accent: #00e5ff;
+    --hero-glow: rgba(0, 229, 255, 0.15);
+}
+
+/* Serpent — neon-turkoosi/teal (kipina.tech) */
+[data-theme="serpent"] {
+    --bg: #000808;
+    --panel: #001e1e;
+    --text: #ccffff;
+    --accent: #00ffff;
+    --green: #00ffaa;
+    --yellow: #d29922;
+    --red: #f85149;
+    --purple: #66cccc;
+    --border: rgba(0, 255, 255, 0.15);
+    --hero-accent: #00ffff;
+    --hero-glow: rgba(0, 255, 255, 0.15);
 }
 
 * { box-sizing: border-box; margin: 0; padding: 0; }
 
 body {
     font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
+    font-size: 16px;
     background: var(--bg);
     color: var(--text);
     min-height: 100vh;
 }
 
-.container { max-width: 1200px; margin: 0 auto; padding: 20px; }
+.container { 
+    max-width: 1600px; 
+    margin: 0 auto; 
+    padding: 20px 40px; 
+}
+
+#app.container {
+    height: 100vh;
+    display: flex;
+    flex-direction: column;
+    overflow: hidden;
+}
+
+#app:not(.active) { display: none; }
+#landing.hidden { display: none; }
 
 /* Tabs */
-.tabs { display: flex; gap: 4px; margin-bottom: 16px; }
+.tabs { display: flex; gap: 4px; margin-bottom: 16px; flex-shrink: 0; }
 .tab {
-    padding: 8px 16px; border-radius: 6px 6px 0 0; cursor: pointer;
+    padding: 10px 20px; border-radius: 6px 6px 0 0; cursor: pointer;
     border: 1px solid var(--border); border-bottom: none;
-    background: var(--bg); color: #8b949e; font-size: 14px;
+    background: var(--bg); color: #8b949e; font-size: 15px;
 }
 .tab.active { background: var(--panel); color: var(--accent); border-color: var(--border); }
 
 /* Panels */
 .panel { display: none; }
-.panel.active { display: block; }
+.panel.active { display: flex; flex-direction: column; flex: 1; min-height: 0; overflow-y: auto; }
 
 /* Status bar */
 .status-bar {
     display: flex; align-items: center; gap: 12px;
-    padding: 8px 14px; background: var(--bg);
+    padding: 10px 16px; background: var(--bg);
     border: 1px solid var(--border); border-radius: 6px 6px 0 0;
-    font-family: 'Courier New', monospace; font-size: 13px;
+    font-family: 'Courier New', monospace; font-size: 14px;
 }
 .status-dot {
     width: 8px; height: 8px; border-radius: 50%; display: inline-block;
@@ -50,21 +113,22 @@ body {
 /* Terminal */
 .terminal {
     background: #010409; border: 1px solid var(--border); border-top: none;
-    font-family: 'Courier New', monospace; font-size: 14px;
+    font-family: 'Courier New', monospace; font-size: 16px;
-    min-height: 300px; max-height: 60vh; overflow-y: auto;
+    flex: 1; min-height: 0; max-height: none; overflow-y: auto;
-    padding: 8px 12px;
+    padding: 12px 16px;
 }
 .terminal-line { padding: 1px 0; white-space: pre-wrap; word-break: break-word; }
 .terminal-prompt { color: var(--yellow); margin-right: 8px; }
 .terminal-input-row {
     display: flex; align-items: center; position: relative;
-    background: #010409; border: 1px solid var(--border); border-top: none;
+    background: #0d1117; border: 1px solid var(--accent); border-top: none;
-    border-radius: 0 0 6px 6px; padding: 8px 12px;
+    border-radius: 0 0 6px 6px; padding: 10px 14px;
-    font-family: 'Courier New', monospace; font-size: 14px;
+    font-family: 'Courier New', monospace; font-size: 15px;
+    box-shadow: 0 2px 8px rgba(88,166,255,0.1);
 }
 .terminal-input {
     flex: 1; background: transparent; border: none; outline: none;
-    color: var(--green); font-family: inherit; font-size: inherit;
+    color: var(--green); font-family: inherit; font-size: 16px;
 }
 .terminal-dropdown {
     display: none; position: absolute; bottom: 100%; left: 30px;
@@ -79,6 +143,12 @@ body {
 }
 .dd-item:hover, .dd-item.active { background: var(--border); color: var(--accent); }
 
+#editor-file-list .dd-item {
+    white-space: pre-wrap;
+    word-break: break-all;
+    line-height: 1.4;
+}
+
 /* Pipeline progress */
 .pipeline-bar {
     display: none; padding: 8px 14px; background: var(--bg);
@@ -100,6 +170,7 @@ body {
 .project-tab {
     padding: 4px 10px; cursor: pointer; border-radius: 4px 4px 0 0;
     font-size: 12px; color: #8b949e;
+    white-space: nowrap; flex-shrink: 0;
 }
 .project-tab.active { background: var(--panel); color: var(--accent); border: 1px solid var(--border); border-bottom: none; }
 
@@ -128,10 +199,10 @@ body {
     background: linear-gradient(145deg, rgba(33,38,45,0.4) 0%, rgba(13,17,23,0.8) 100%);
     backdrop-filter: blur(12px);
     border: 1px solid rgba(240,246,252,0.1);
-    border-radius: 12px;
+    border-radius: 14px;
-    padding: 6px 6px 4px;
+    padding: 8px 8px 6px;
     text-align: center;
-    width: 72px;
+    width: 90px;
     opacity: 0.8;
     cursor: pointer;
     transition: all 0.4s cubic-bezier(0.175, 0.885, 0.32, 1.275);
@@ -144,12 +215,12 @@ body {
     box-shadow: 0 8px 14px rgba(0,0,0,0.4);
 }
 .agent-avatar img {
-    width: 50px; height: 50px; border-radius: 12px;
+    width: 64px; height: 64px; border-radius: 14px;
     margin-bottom: 4px; border: 2px solid rgba(240,246,252,0.1);
     transition: all 0.4s ease; object-fit: cover;
 }
 .agent-avatar .avatar-name {
-    font-size: 10px; color: #8b949e; white-space: nowrap;
+    font-size: 11px; color: #8b949e; white-space: nowrap;
     overflow: hidden; text-overflow: ellipsis;
 }
 .agent-avatar.active {
@@ -163,6 +234,13 @@ body {
 .agent-avatar.active img {
     border-color: var(--accent);
     box-shadow: 0 0 25px rgba(88,166,255,0.8);
+    animation: agentBlink 1.5s infinite;
+}
+
+@keyframes agentBlink {
+    0% { opacity: 0.8; box-shadow: 0 0 15px rgba(88,166,255,0.5); }
+    50% { opacity: 1.0; box-shadow: 0 0 35px rgba(88,166,255,1.0); }
+    100% { opacity: 0.8; box-shadow: 0 0 15px rgba(88,166,255,0.5); }
 }
 
 /* Settings */
@@ -193,6 +271,218 @@ body {
     display: grid; grid-template-columns: 1fr 1fr; gap: 16px;
 }
 
+/* ===== LANDING PAGE ===== */
+
+#landing {
+    font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
+    min-height: 100vh;
+    position: relative;
+    overflow: hidden;
+}
+
+.bg-mesh {
+    position: fixed; inset: 0; z-index: -1;
+    background:
+        radial-gradient(ellipse 80% 60% at 20% 40%, var(--hero-glow) 0%, transparent 70%),
+        radial-gradient(ellipse 60% 50% at 80% 20%, rgba(88,166,255,0.06) 0%, transparent 70%),
+        var(--bg);
+}
+
+.landing-nav {
+    padding: 20px 40px;
+    display: flex; align-items: center; justify-content: space-between;
+}
+.landing-logo { text-decoration: none; font-size: 18px; font-weight: 700; }
+.logo-accent { color: var(--hero-accent); }
+.logo-sub { color: #8b949e; font-weight: 400; }
+.theme-cycle-btn {
+    background: none; border: 1px solid var(--border); border-radius: 8px;
+    width: 38px; height: 38px; font-size: 20px; cursor: pointer;
+    display: flex; align-items: center; justify-content: center;
+    transition: border-color 0.2s, transform 0.15s;
+}
+.theme-cycle-btn:hover {
+    border-color: var(--hero-accent); transform: scale(1.1);
+}
+
+/* Hero */
+.hero {
+    padding: 60px 40px 40px;
+}
+.hero-container {
+    max-width: 1200px; margin: 0 auto;
+    display: grid; grid-template-columns: 1fr 400px; gap: 60px; align-items: center;
+}
+.hero-title {
+    font-size: clamp(2rem, 4vw, 3rem); font-weight: 800;
+    line-height: 1.15; color: #e6edf3; margin-bottom: 16px;
+}
+.hero-divider {
+    width: 60px; height: 3px; background: var(--hero-accent);
+    border-radius: 2px; margin-bottom: 20px;
+}
+.hero-desc {
+    font-size: 1.05rem; color: #8b949e; line-height: 1.7; margin-bottom: 12px;
+}
+.hero-notice {
+    font-size: 0.9rem; color: #6e7681; line-height: 1.6;
+    border-left: 2px solid var(--border); padding-left: 12px; margin-bottom: 28px;
+}
+
+/* Hero input */
+.hero-input-group {
+    display: flex; gap: 8px; margin-bottom: 20px;
+}
+.hero-input {
+    flex: 1; padding: 14px 18px; font-size: 16px;
+    font-family: 'JetBrains Mono', 'Courier New', monospace;
+    background: var(--panel); color: var(--text);
+    border: 1px solid var(--border); border-radius: 8px;
+    outline: none; transition: border-color 0.2s;
+}
+.hero-input:focus {
+    border-color: var(--hero-accent); box-shadow: 0 0 0 3px var(--hero-glow);
+}
+.hero-input::placeholder { color: #484f58; }
+.hero-input.shake {
+    animation: shake 0.4s ease;
+    border-color: #f85149;
+    box-shadow: 0 0 0 3px rgba(248,81,73,0.2);
+}
+@keyframes shake {
+    0%, 100% { transform: translateX(0); }
+    20%, 60% { transform: translateX(-6px); }
+    40%, 80% { transform: translateX(6px); }
+}
+.hero-btn {
+    padding: 14px 28px; font-size: 16px; font-weight: 600;
+    font-family: 'Inter', sans-serif;
+    background: var(--hero-accent); color: #fff; border: none; border-radius: 8px;
+    cursor: pointer; transition: background 0.2s, transform 0.1s;
+    white-space: nowrap;
+}
+.hero-btn:hover { filter: brightness(0.85); transform: translateY(-1px); }
+.hero-btn:active { transform: translateY(0); }
+
+/* Example buttons */
+.hero-examples { display: flex; flex-wrap: wrap; gap: 8px; align-items: center; }
+.hero-examples-label { color: #6e7681; font-size: 14px; margin-right: 4px; }
+.example-btn {
+    padding: 8px 16px; font-size: 13px; font-family: 'Inter', sans-serif;
+    background: transparent; color: var(--accent);
+    border: 1px solid var(--border); border-radius: 6px;
+    cursor: pointer; transition: all 0.2s;
+}
+.example-btn:hover {
+    border-color: var(--accent); background: rgba(88,166,255,0.08);
+}
+
+/* Hero orb */
+.hero-orb-wrapper {
+    display: flex; justify-content: center; align-items: center;
+}
+.hero-orb {
+    width: 340px; height: 340px; border-radius: 50%;
+    background: radial-gradient(circle at 30% 30%, var(--hero-glow) 0%, transparent 70%);
+    display: flex; align-items: center; justify-content: center;
+    animation: orb-float 6s ease-in-out infinite;
+}
+.hero-orb-img {
+    width: 100%; height: 100%; object-fit: contain;
+    filter: drop-shadow(0 0 40px var(--hero-glow));
+    transition: opacity 0.2s ease;
+}
+@keyframes orb-float {
+    0%, 100% { transform: translateY(0); }
+    50% { transform: translateY(-12px); }
+}
+
+/* How section */
+.how-section {
+    padding: 60px 40px;
+    background: rgba(22,27,34,0.6);
+    border-top: 1px solid var(--border);
+}
+.how-container { max-width: 900px; margin: 0 auto; }
+.how-title {
+    text-align: center; font-size: 1.5rem; font-weight: 700;
+    color: #e6edf3; margin-bottom: 40px;
+}
+.how-steps {
+    display: grid; grid-template-columns: repeat(3, 1fr); gap: 32px;
+}
+.how-step {
+    text-align: center; padding: 24px;
+    background: var(--panel); border: 1px solid var(--border);
+    border-radius: 12px; transition: border-color 0.3s;
+}
+.how-step:hover { border-color: var(--hero-accent); }
+.how-step-num {
+    width: 40px; height: 40px; line-height: 40px;
+    border-radius: 50%; background: var(--hero-glow);
+    color: var(--hero-accent); font-weight: 700; font-size: 18px;
+    margin: 0 auto 14px;
+}
+.how-step h3 { color: #e6edf3; font-size: 1rem; margin-bottom: 8px; }
+.how-step p { color: #8b949e; font-size: 0.9rem; line-height: 1.5; }
+
+/* Landing footer */
+.landing-footer {
+    text-align: center; padding: 32px 40px;
+    color: #484f58; font-size: 13px;
+    border-top: 1px solid var(--border);
+}
+.landing-footer a { color: #8b949e; }
+
+/* Responsive */
+@media (max-width: 860px) {
+    .hero-container { grid-template-columns: 1fr; gap: 32px; }
+    .hero-orb-wrapper { order: -1; }
+    .hero-orb { width: 220px; height: 220px; }
+    .how-steps { grid-template-columns: 1fr; }
+    .hero-input-group { flex-direction: column; }
+}
+
+/* ===== OPPIMISPOLKU ===== */
+.learn-step {
+    margin: 12px 0; border: 1px solid var(--border);
+    border-radius: 8px; background: var(--panel); overflow: hidden;
+}
+.learn-step-header {
+    display: flex; align-items: center; gap: 12px;
+    padding: 12px 16px; cursor: pointer;
+    transition: background 0.15s;
+}
+.learn-step-header:hover { background: rgba(88,166,255,0.04); }
+.learn-step-num {
+    width: 28px; height: 28px; line-height: 28px; text-align: center;
+    border-radius: 50%; background: var(--hero-glow);
+    color: var(--hero-accent); font-weight: 700; font-size: 13px; flex-shrink: 0;
+}
+.learn-step-agent {
+    font-weight: 600; color: #e6edf3; font-size: 14px;
+}
+.learn-step-label {
+    color: #8b949e; font-size: 13px; margin-left: auto;
+}
+.learn-step-body {
+    display: none; padding: 0 16px 16px;
+    border-top: 1px solid var(--border);
+}
+.learn-step-body.open { display: block; }
+.learn-section-title {
+    color: var(--accent); font-size: 12px; font-weight: 600;
+    text-transform: uppercase; letter-spacing: 0.5px;
+    margin: 14px 0 6px;
+}
+.learn-code {
+    font-family: 'JetBrains Mono', 'Courier New', monospace;
+    font-size: 12px; line-height: 1.6;
+    background: #010409; border: 1px solid var(--border);
+    border-radius: 6px; padding: 12px; overflow-x: auto;
+    max-height: 300px; overflow-y: auto; white-space: pre-wrap;
+}
+
 /* Animations */
 @keyframes blink { 0%,100% { opacity:1 } 50% { opacity:0 } }
 @keyframes spin { to { transform: rotate(360deg) } }

network-poc/hub-local.log

@@ -0,0 +1,34 @@
+   Compiling hub v0.3.1 (/Users/jaakko/code/kipina-codes/playground/agentic-studio/network-poc/hub)
+    Finished `dev` profile [unoptimized + debuginfo] target(s) in 2.95s
+     Running `target/debug/hub`
+2026-04-12T04:56:09.723604Z  INFO hub: Tietokanta alustettu
+2026-04-12T04:56:09.725088Z  INFO hub: Kipinä Agent Hub v0.3.1 käynnistyy osoitteessa http://localhost:3000
+2026-04-12T04:56:18.997935Z  INFO hub: Solmu 1 yhdistyi osoitteesta 127.0.0.1
+2026-04-12T04:56:19.027478Z  INFO hub: Solmu 1 (natiivi) | 127.0.0.1 | Mac | Darwin 26.3.1 | 12 ydintä | 32768 MB RAM | varaus: 4 GB
+2026-04-12T04:56:19.029931Z  INFO hub:   GPU 0: Apple M2 Max | VRAM: 0/24576 MB | 0°C | 0%
+2026-04-12T04:56:31.260470Z  INFO hub: Solmu 2 yhdistyi osoitteesta 127.0.0.1
+2026-04-12T04:56:31.281759Z  INFO hub: Solmu 2 (selain) | 127.0.0.1 | MacIntel | 11 ydintä | ~8 GB RAM | GPU: ei GPU:ta | tehtävä: viewer | varaus: 0 GB
+2026-04-12T04:56:31.283313Z  INFO hub: Reititettiin API-pyyntö solmulle 1 (Malli: qwen-coder)
+
+━━━ Solmu 1 ━━━ qwen2.5-coder:7b-instruct-q4_K_M (Ollama) ━━━
+  Prompt: "ping"
+  Vastaus: Pong! How can I assist you today?
+  11 tokenia | 4502ms | 56.3 tok/s
+2026-04-12T04:56:36.419646Z  INFO hub: Solmu 2 (127.0.0.1) poistui verkosta.
+2026-04-12T04:56:36.433155Z  INFO hub: Solmu 3 yhdistyi osoitteesta 127.0.0.1
+2026-04-12T04:56:36.445127Z  INFO hub: Solmu 3 (selain) | 127.0.0.1 | MacIntel | 11 ydintä | ~8 GB RAM | GPU: ei GPU:ta | tehtävä: viewer | varaus: 0 GB
+2026-04-12T04:56:36.445818Z  INFO hub: Reititettiin API-pyyntö solmulle 1 (Malli: qwen-coder)
+
+━━━ Solmu 1 ━━━ qwen2.5-coder:7b-instruct-q4_K_M (Ollama) ━━━
+  Prompt: "ping"
+  Vastaus: Pong! How can I assist you today? If you have any questions or need information on a specific topic, feel free to let me know.
+  31 tokenia | 679ms | 57.5 tok/s
+2026-04-12T04:56:39.466711Z  INFO hub: Solmu 3 (127.0.0.1) poistui verkosta.
+2026-04-12T04:56:43.881216Z  INFO hub: Solmu 4 yhdistyi osoitteesta 127.0.0.1
+2026-04-12T04:56:43.894385Z  INFO hub: Solmu 4 (selain) | 127.0.0.1 | MacIntel | 3 ydintä | ~16 GB RAM | GPU: ei GPU:ta | tehtävä: viewer | varaus: 0 GB
+2026-04-12T04:56:43.894960Z  INFO hub: Reititettiin API-pyyntö solmulle 1 (Malli: qwen-coder)
+
+━━━ Solmu 1 ━━━ qwen2.5-coder:7b-instruct-q4_K_M (Ollama) ━━━
+  Prompt: "ping"
+  Vastaus: Pong! How can I assist you today?
+  11 tokenia | 333ms | 58.7 tok/s

network-poc/hub/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "hub"
-version = "0.3.0"
+version = "0.3.2"
 edition = "2024"
 
 [dependencies]
@@ -11,7 +11,6 @@ serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
-uuid = { version = "1.7.0", features = ["v4", "serde"] }
 futures = "0.3"
 rusqlite = { version = "0.31", features = ["bundled"] }
 chrono = "0.4"

network-poc/hub/src/db.rs

@@ -49,6 +49,13 @@ impl NodeDb {
                 INSERT INTO _schema_version VALUES (3);
             ");
         }
+        if version < 4 {
+            let _ = conn.execute_batch("
+                ALTER TABLE node_sessions ADD COLUMN is_paused BOOLEAN DEFAULT 0;
+                DELETE FROM _schema_version;
+                INSERT INTO _schema_version VALUES (4);
+            ");
+        }
 
         conn.execute_batch("
             CREATE TABLE IF NOT EXISTS node_sessions (
@@ -84,7 +91,10 @@ impl NodeDb {
                 has_webgpu BOOLEAN,
 
                 -- Tehtävätilastot
-                tasks_completed INTEGER DEFAULT 0
+                tasks_completed INTEGER DEFAULT 0,
+                
+                -- Ohjaustilat
+                is_paused BOOLEAN DEFAULT 0
             );
 
             CREATE TABLE IF NOT EXISTS pair_results (
@@ -183,6 +193,14 @@ impl NodeDb {
         );
     }
 
+    pub fn update_session_status(&self, node_id: u64, is_paused: bool) {
+        let conn = self.conn.lock().unwrap_or_else(|e| e.into_inner());
+        let _ = conn.execute(
+            "UPDATE node_sessions SET is_paused = ?1 WHERE node_id = ?2 AND disconnected_at IS NULL",
+            params![is_paused as i64, node_id as i64],
+        );
+    }
+
     /// Sulkee saman IP:n viewer-sessiot kun aktiivinen node liittyy
     pub fn close_viewers_by_ip(&self, ip: &str) {
         let conn = self.conn.lock().unwrap_or_else(|e| e.into_inner());
@@ -216,7 +234,7 @@ impl NodeDb {
             "SELECT id, node_id, ip, node_type, connected_at, disconnected_at,
                     platform, hostname, os, cpu_cores, cpu_model, ram_mb,
                     gpu_name, gpu_vendor, gpu_backend, vram_total_mb, gpu_temp_c, gpu_util_pct,
-                    allocated_gb, selected_task, has_webgpu, tasks_completed
+                    allocated_gb, selected_task, has_webgpu, tasks_completed, is_paused
              FROM node_sessions ORDER BY id DESC LIMIT ?1"
         ).unwrap();
 
@@ -244,6 +262,7 @@ impl NodeDb {
                 "selected_task": row.get::<_, Option<String>>(19)?,
                 "has_webgpu": row.get::<_, Option<bool>>(20)?,
                 "tasks_completed": row.get::<_, i64>(21)?,
+                "is_paused": row.get::<_, Option<bool>>(22)?.unwrap_or(false),
             }))
         }).unwrap().filter_map(|r| r.ok()).collect()
     }

network-poc/hub/src/main.rs

@@ -25,7 +25,7 @@ const ALLOWED_ORIGINS: &[&str] = &[
 ];
 
 // Sallitut viestityyypit clientilta
-const ALLOWED_MSG_TYPES: &[&str] = &["auth", "result", "pair_done", "llm_chunk", "llm_done", "llm_error", "download_progress", "user_text", "single_tokenize_done"];
+const ALLOWED_MSG_TYPES: &[&str] = &["auth", "result", "pair_done", "llm_chunk", "llm_done", "llm_error", "download_progress", "user_text", "single_tokenize_done", "status_update"];
 
 struct AppState {
     next_node_id: Mutex<u64>,
@@ -40,9 +40,14 @@ struct AppState {
     node_ips: Mutex<HashMap<u64, IpAddr>>,
     node_tasks: Mutex<HashMap<u64, String>>,  // node_id → selected_task
     node_types: Mutex<HashMap<u64, String>>,  // node_id → "native" | "browser"
+    node_paused: Mutex<std::collections::HashSet<u64>>, // node_id → onko tauolla
     node_busy: Mutex<std::collections::HashSet<u64>>,  // Solmut joilla on aktiivinen tehtävä
+    node_active_task: Mutex<HashMap<u64, String>>,   // node_id → task_id (mikä tehtävä on kesken)
     pending_task_ids: Mutex<std::collections::HashSet<String>>,  // Hubin jakamat task_id:t (gamification-validointi)
+    pending_responses: Mutex<HashMap<String, tokio::sync::oneshot::Sender<serde_json::Value>>>,  // task_id → oneshot API-vastaukselle
     api_rate_limits: Mutex<HashMap<IpAddr, (std::time::Instant, u32)>>,  // IP → (ikkuna-alku, pyyntömäärä)
+    node_models: tokio::sync::RwLock<HashMap<u64, serde_json::Value>>, // node_id → ollama tags JSON
+    node_max_param_b: tokio::sync::RwLock<HashMap<u64, u32>>, // node_id → suurimman mallin parametrit (B)
     db: db::NodeDb,
 }
 
@@ -80,6 +85,8 @@ tr:hover td { background:#1c2333; }
 .table-wrap { overflow-x:auto; max-height:70vh; overflow-y:auto; }
 .online { color:var(--green); }
 .offline { color:#8b949e; }
+.pause-btn { background:var(--panel); border:1px solid var(--border); color:var(--text); padding:4px 8px; border-radius:4px; cursor:pointer; font-size:12px; }
+.pause-btn:hover { border-color:var(--yellow); }
 </style>
 </head>
 <body>
@@ -91,6 +98,7 @@ tr:hover td { background:#1c2333; }
 <div class="tabs">
     <div class="tab active" onclick="showTab('sessions')">Sessiot</div>
     <div class="tab" onclick="showTab('pairs')">Tokenisointiparit</div>
+    <div class="tab" onclick="showTab('hardware')">Laitteisto & Mallit</div>
 </div>
 
 <div id="sessions" class="panel active">
@@ -99,12 +107,12 @@ tr:hover td { background:#1c2333; }
         <colgroup>
             <col style="width:35px"><col style="width:85px"><col style="width:95px"><col style="width:65px"><col style="width:110px"><col style="width:80px">
             <col style="width:65px"><col style="width:40px"><col style="width:70px"><col style="width:90px"><col style="width:60px">
-            <col style="width:65px"><col style="width:40px"><col style="width:130px"><col style="width:60px">
+            <col style="width:65px"><col style="width:40px"><col style="width:130px"><col style="width:60px"><col style="width:80px">
         </colgroup>
         <thead><tr>
             <th>ID</th><th>Tila</th><th>Tehtävä</th><th>Tyyppi</th><th>IP</th><th>Alusta</th>
             <th>OS</th><th>CPU</th><th>RAM</th><th>GPU</th><th>VRAM</th>
-            <th>WebGPU</th><th>Teht.</th><th>Yhdistetty</th><th>Kesto</th>
+            <th>WebGPU</th><th>Teht.</th><th>Yhdistetty</th><th>Kesto</th><th>Toiminnot</th>
         </tr></thead><tbody id="sessions-body"></tbody></table>
     </div>
 </div>
@@ -118,6 +126,19 @@ tr:hover td { background:#1c2333; }
     </div>
 </div>
 
+<div id="hardware" class="panel">
+    <div class="stats-grid" id="hardware-stats"></div>
+    <h2 style="margin-top: 10px; margin-bottom: 10px; color: var(--accent); font-size: 16px;">Käytettävissä olevat paikalliset kielimallit</h2>
+    <div class="table-wrap">
+        <table>
+            <thead><tr>
+                <th>Nimi</th><th>Koko</th><th>Parametrit</th>
+            </tr></thead>
+            <tbody id="models-body"></tbody>
+        </table>
+    </div>
+</div>
+
 <script>
 function showTab(name) {
     document.querySelectorAll('.panel').forEach(p => p.classList.remove('active'));
@@ -149,12 +170,16 @@ function duration(start, end) {
 }
 
 async function load() {
-    const [statsRes, sessionsRes, pairsRes] = await Promise.all([
+    const [statsRes, sessionsRes, pairsRes, hwRes, modelsRes] = await Promise.all([
-        fetch('/api/stats'), fetch('/api/sessions'), fetch('/api/pairs')
+        fetch('/api/stats'), fetch('/api/sessions'), fetch('/api/pairs'),
+        fetch('/api/v1/hardware').catch(() => ({json: async()=>({gpu_name:'', vram_mb:0, ram_mb:0})})),
+        fetch('/api/v1/ollama/tags').catch(() => ({json: async()=>({models:[]})}))
     ]);
     const stats = await statsRes.json();
     const sessions = await sessionsRes.json();
     const pairs = await pairsRes.json();
+    const hw = await hwRes.json().catch(() => ({gpu_name:'', vram_mb:0, ram_mb:0}));
+    const modelsData = await modelsRes.json().catch(() => ({models:[]}));
 
     // Versio
     if (stats.version) document.getElementById('admin-version').textContent = 'v' + stats.version;
@@ -173,7 +198,7 @@ async function load() {
     ].map(s => `<div class="stat-card"><div class="val">${s.v}</div><div class="label">${s.l}</div></div>`).join('');
 
     // Sessions — lajittelu: 1) aktiiviset nodet (online + ei viewer), 2) katsojat (online + viewer), 3) offline
-    const taskNames = {'tokenize':'Tokenisaatio','smollm-135m':'SmolLM 135M','qwen-05b':'Qwen2.5 0.5B','phi3-mini':'Phi-3 Mini','qwen-coder-05b':'Coder 0.5B','qwen-coder-3b':'Coder 3B','viewer':'Katsoja','codelab-viewer':'Koodilabra'};
+    const taskNames = {'tokenize':'Tokenisaatio','qwen-05b':'Qwen2.5 0.5B','qwen-coder-05b':'Coder 0.5B','qwen-coder-3b':'Coder 3B','viewer':'Katsoja','codelab-viewer':'Koodilabra'};
     sessions.sort((a, b) => {
         const aOnline = !a.disconnected_at;
         const bOnline = !b.disconnected_at;
@@ -190,9 +215,17 @@ async function load() {
     document.getElementById('sessions-body').innerHTML = sessions.map(s => {
         const online = !s.disconnected_at;
         const isViewer = s.selected_task === 'viewer';
-        const status = online
+        let status;
-            ? (isViewer ? '<span style="color:#d29922">CONNECTED</span>' : '<span class="online">ACTIVE</span>')
+        if (!online) {
-            : '<span class="offline">offline</span>';
+            status = '<span class="offline">offline</span>';
+        } else if (isViewer) {
+            status = '<span style="color:#d29922">CONNECTED</span>';
+        } else if (s.is_paused) {
+            status = '<span style="color:#8b949e">PAUSED</span>';
+        } else {
+            status = '<span class="online">ACTIVE</span>';
+        }
+        
         const typeBadge = s.node_type === 'native' ? badge('native','blue') : badge('browser','yellow');
         const taskColor = isViewer ? 'yellow' : s.selected_task === 'tokenize' ? 'green' : 'blue';
         const taskBadge = badge(taskNames[s.selected_task] || s.selected_task || '?', taskColor);
@@ -205,11 +238,16 @@ async function load() {
         const os = s.os || '-';
         const time = s.connected_at ? new Date(s.connected_at).toLocaleString('fi-FI') : '';
         const dur = duration(s.connected_at, s.disconnected_at);
+        const actionBtn = online && !isViewer 
+            ? `<button class="pause-btn" onclick="togglePause(${s.node_id}, ${s.is_paused})">${s.is_paused ? '▶ Työhön' : '⏸ Tauolle'}</button>` 
+            : '';
+            
         return `<tr>
             <td>${s.node_id}</td><td>${status}</td><td>${taskBadge}</td><td>${typeBadge}</td><td>${s.ip}</td>
             <td>${plat}</td><td>${os}</td><td>${cores}</td><td>${ram}</td>
             <td>${gpu}</td><td>${vram}</td><td>${gpuBadge}</td>
             <td>${s.tasks_completed}</td><td>${time}</td><td>${dur}</td>
+            <td>${actionBtn}</td>
         </tr>`;
     }).join('');
 
@@ -229,6 +267,35 @@ async function load() {
             <td>${p.duration_ms||0}ms</td>
         </tr>`;
     }).join('');
+
+    // Hardware
+    document.getElementById('hardware-stats').innerHTML = [
+        {v: hw.gpu_name || '-', l: 'Paikallinen GPU tila'},
+        {v: hw.vram_mb ? hw.vram_mb + ' MB' : '-', l: 'GPU Muisti (VRAM)'},
+        {v: hw.ram_mb ? hw.ram_mb + ' MB' : '-', l: 'RAM'},
+    ].map(s => `<div class="stat-card"><div class="val">${s.v}</div><div class="label">${s.l}</div></div>`).join('');
+
+    // Models
+    document.getElementById('models-body').innerHTML = (modelsData.models || []).map(m => {
+        const sizeGb = (m.size / (1024*1024*1024)).toFixed(2) + ' GB';
+        const params = m.details?.parameter_size || '-';
+        return `<tr>
+            <td><strong>${m.name}</strong></td>
+            <td>${sizeGb}</td>
+            <td>${params}</td>
+        </tr>`;
+    }).join('');
+}
+
+async function togglePause(nodeId, isPaused) {
+    try {
+        await fetch('/api/v1/control/' + nodeId, {
+            method: 'POST',
+            headers: { 'Content-Type': 'application/json' },
+            body: JSON.stringify({ action: isPaused ? 'resume' : 'pause' })
+        });
+        load(); // virkistetään
+    } catch(e) { console.error(e); }
 }
 
 load();
@@ -262,9 +329,14 @@ async fn main() {
         node_ips: Mutex::new(HashMap::new()),
         node_tasks: Mutex::new(HashMap::new()),
         node_types: Mutex::new(HashMap::new()),
+        node_paused: Mutex::new(std::collections::HashSet::new()),
         node_busy: Mutex::new(std::collections::HashSet::new()),
+        node_active_task: Mutex::new(HashMap::new()),
         pending_task_ids: Mutex::new(std::collections::HashSet::new()),
+        pending_responses: Mutex::new(HashMap::new()),
         api_rate_limits: Mutex::new(HashMap::new()),
+        node_models: tokio::sync::RwLock::new(HashMap::new()),
+        node_max_param_b: tokio::sync::RwLock::new(HashMap::new()),
         db: db::NodeDb::new(&std::env::var("DATABASE_PATH").unwrap_or_else(|_| "nodes.db".to_string())),
     });
 
@@ -351,9 +423,7 @@ async fn main() {
                         // Vapaa node -> lähetetään oikea tehtävä
                         let msg = match task.as_str() {
                             "tokenize" => Some(serde_json::json!({ "type": "pair_task", "en": en, "fi": fi })),
-                            "smollm-135m" => Some(serde_json::json!({ "type": "llm_prompt", "prompt": llm_prompts[llm_idx], "model": "smollm-135m" })),
                             "qwen-05b" => Some(serde_json::json!({ "type": "llm_prompt", "prompt": llm_prompts[llm_idx], "model": "qwen-05b" })),
-                            "phi3-mini" => Some(serde_json::json!({ "type": "llm_prompt", "prompt": llm_prompts[llm_idx], "model": "phi3-mini" })),
                             _ => None, // Coder ja viewer ei saa auto-tehtäviä
                         };
 
@@ -381,6 +451,7 @@ async fn main() {
         .route("/api/pairs", get(api_pairs))
         .route("/api/stats", get(api_stats))
         .route("/api/v1/chat/completions", axum::routing::post(api_chat_completions))
+        .route("/api/v1/control/:id", axum::routing::post(api_control_node))
         .route("/api/v1/model", axum::routing::post(api_change_model))
         .route("/api/v1/hardware", get(api_hardware))
         .route("/api/v1/ollama/tags", get(api_ollama_tags))
@@ -400,6 +471,26 @@ async fn main() {
     axum::serve(listener, app.into_make_service_with_connect_info::<SocketAddr>()).await.unwrap();
 }
 
+async fn api_control_node(
+    headers: axum::http::HeaderMap,
+    axum::extract::State(state): axum::extract::State<Arc<AppState>>,
+    axum::extract::Path(id): axum::extract::Path<u64>,
+    axum::Json(payload): axum::Json<serde_json::Value>,
+) -> axum::response::Response {
+    if !check_admin_auth(&headers) { return admin_unauthorized(); }
+    let action = payload.get("action").and_then(|v| v.as_str()).unwrap_or("");
+    if action == "pause" || action == "resume" {
+        let msg = serde_json::json!({ "type": "control", "action": action });
+        let channels = state.node_channels.read().await;
+        if let Some(tx) = channels.get(&id) {
+            let _ = tx.send(msg.to_string());
+            tracing::info!("Lähetetty control: {} solmulle {}", action, id);
+            return axum::Json(serde_json::json!({"status": "ok"})).into_response();
+        }
+    }
+    (axum::http::StatusCode::BAD_REQUEST, "Invalid action or node offline").into_response()
+}
+
 async fn api_sessions(
     headers: axum::http::HeaderMap,
     axum::extract::State(state): axum::extract::State<Arc<AppState>>,
@@ -563,6 +654,17 @@ async fn broadcast_stats(state: &Arc<AppState>) {
         "tasks": completed
     });
     let _ = state.stats_tx.send(stats_msg.to_string());
+
+    // Uutta: Laitetaan sama tieto myös kaikille yhdistyneille solmuille (viesti Hubilta Solmuille)
+    let node_status = serde_json::json!({
+        "type": "network_status",
+        "active_nodes": total_nodes,
+        "tasks": completed
+    });
+    let msg_str = node_status.to_string();
+    for tx in state.node_channels.read().await.values() {
+        let _ = tx.send(msg_str.clone());
+    }
 }
 
 /// Validoi client-viesti: pakollinen "type"-kenttä, sallittu tyyppi, validi JSON
@@ -730,6 +832,9 @@ async fn handle_socket(socket: WebSocket, state: Arc<AppState>, ip: IpAddr) {
                 }
                 state.node_tasks.lock().unwrap().insert(node_id, selected_task);
                 state.node_types.lock().unwrap().insert(node_id, node_type.to_string());
+                // Uudelleen-kirjautuessa nollataan tauko
+                state.node_paused.lock().unwrap().remove(&node_id);
+                state.db.update_session_status(node_id, false);
 
                 if node_type == "native" {
                     let sys = json.get("system");
@@ -743,6 +848,36 @@ async fn handle_socket(socket: WebSocket, state: Arc<AppState>, ip: IpAddr) {
                         node_id, ip, hostname, os, cores, ram, allocated
                     );
 
+                    // Tallennetaan välitetyt mallit muistiin + parsitaan suurin malli
+                    if let Some(models) = json.get("models") {
+                        let mut nm = state.node_models.write().await;
+                        nm.insert(node_id, models.clone());
+
+                        // Parsitaan suurin mallikoko (B) nimestä: "qwen3:32b" → 32, "qwen2.5-coder:7b" → 7
+                        let max_b = models.get("models").and_then(|v| v.as_array()).map(|arr| {
+                            arr.iter().filter_map(|m| {
+                                let name = m.get("name")?.as_str()?;
+                                // Etsitään :N tai :Nb tai -Nb muoto
+                                let lower = name.to_lowercase();
+                                for part in lower.split(&[':', '-'][..]) {
+                                    if let Some(num_str) = part.strip_suffix('b') {
+                                        if let Ok(n) = num_str.parse::<f32>() { return Some(n as u32); }
+                                    } else if let Ok(n) = part.parse::<f32>() {
+                                        if n >= 0.5 && n <= 500.0 { return Some(n as u32); }
+                                    }
+                                }
+                                // Fallback: koko tiedostosta (size / ~0.5GB per B param Q4)
+                                let size = m.get("size")?.as_u64()?;
+                                Some((size / 500_000_000) as u32) // karkea arvio
+                            }).max().unwrap_or(0)
+                        }).unwrap_or(0);
+
+                        if max_b > 0 {
+                            state.node_max_param_b.write().await.insert(node_id, max_b);
+                            tracing::info!("Solmu {} — suurin malli: ~{}B parametria", node_id, max_b);
+                        }
+                    }
+
                     if let Some(gpus) = json.get("gpus").and_then(|v| v.as_array()) {
                         for gpu in gpus {
                             tracing::info!(
@@ -780,6 +915,18 @@ async fn handle_socket(socket: WebSocket, state: Arc<AppState>, ip: IpAddr) {
             });
             let _ = state.stats_tx.send(join_msg.to_string());
 
+        } else if msg_type == "status_update" {
+            let status = json.get("status").and_then(|v| v.as_str()).unwrap_or("active");
+            if status == "paused" {
+                state.node_paused.lock().unwrap().insert(node_id);
+                state.db.update_session_status(node_id, true);
+                tracing::info!("Solmu {} ({}) asettui tauolle.", node_id, ip);
+            } else {
+                state.node_paused.lock().unwrap().remove(&node_id);
+                state.db.update_session_status(node_id, false);
+                tracing::info!("Solmu {} ({}) on taas aktiivinen.", node_id, ip);
+            }
+            broadcast_stats(&state).await;
         } else if msg_type == "result" {
             tracing::info!("Solmu {} sai tuloksen: {}", node_id, text);
             {
@@ -789,6 +936,7 @@ async fn handle_socket(socket: WebSocket, state: Arc<AppState>, ip: IpAddr) {
             broadcast_stats(&state).await;
         } else if msg_type == "pair_done" {
             state.node_busy.lock().unwrap().remove(&node_id);
+            state.node_active_task.lock().unwrap().remove(&node_id);
             {
                 let mut json = json; // Siirretään omistajuus muokkausta varten
                 if let Some(obj) = json.as_object_mut() {
@@ -875,30 +1023,44 @@ async fn handle_socket(socket: WebSocket, state: Arc<AppState>, ip: IpAddr) {
         } else if msg_type == "llm_done" {
             // Vapautetaan solmu ja tarkistetaan task_id:n aitous
             state.node_busy.lock().unwrap().remove(&node_id);
-            let valid_task = if let Some(tid) = json.get("task_id").and_then(|v| v.as_str()) {
+            state.node_active_task.lock().unwrap().remove(&node_id);
-                state.pending_task_ids.lock().unwrap().remove(tid)
+            let task_id = json.get("task_id").and_then(|v| v.as_str()).map(|s| s.to_string());
+            let valid_task = if let Some(ref tid) = task_id {
+                state.pending_task_ids.lock().unwrap().remove(tid.as_str())
             } else {
                 false
             };
+
+            // Jos API-pyyntö odottaa tätä vastausta, reititetään suoraan oneshot-kanavaan
+            let api_sender = task_id.as_ref().and_then(|tid| {
+                state.pending_responses.lock().unwrap().remove(tid)
+            });
+
             {
                 let mut json = json;
                 if let Some(obj) = json.as_object_mut() {
                     let model = obj.get("model").and_then(|v| v.as_str()).unwrap_or("?");
                     let prompt = obj.get("prompt").and_then(|v| v.as_str()).unwrap_or("");
-                    let response = obj.get("response").and_then(|v| v.as_str()).unwrap_or("");
+                    let _response = obj.get("response").and_then(|v| v.as_str()).unwrap_or("");
                     let tok_gen = obj.get("tokens_generated").and_then(|v| v.as_u64()).unwrap_or(0);
                     let duration = obj.get("duration_ms").and_then(|v| v.as_f64()).unwrap_or(0.0);
                     let tok_s = obj.get("tokens_per_sec").and_then(|v| v.as_f64()).unwrap_or(0.0);
 
                     println!();
                     println!("\x1b[35m━━━ Solmu {} ━━━ {} ━━━\x1b[0m", node_id, model);
-                    println!("  Prompt: \x1b[33m\"{}\"\x1b[0m", prompt);
+                    let prompt_preview: String = prompt.chars().take(80).collect();
-                    println!("  Vastaus: \x1b[32m{}\x1b[0m", response);
+                    println!("  Prompt: \x1b[33m\"{}...\"\x1b[0m", prompt_preview);
                     println!("  {} tokenia | {:.0}ms | \x1b[36m{:.1} tok/s\x1b[0m", tok_gen, duration, tok_s);
 
                     state.db.increment_tasks(node_id);
                     obj.insert("node_id".to_string(), serde_json::json!(node_id));
                 }
+
+                if let Some(sender) = api_sender {
+                    // API-pyyntö: reititetään vastaus suoraan odottajalle
+                    let _ = sender.send(json.clone());
+                }
+                // UI-broadcast jatkuu normaalisti
                 let _ = state.stats_tx.send(json.to_string());
 
                 let active_incentives = state.feature_flags.read().await.get("Insentiivit").copied().unwrap_or(false);
@@ -931,40 +1093,46 @@ async fn handle_socket(socket: WebSocket, state: Arc<AppState>, ip: IpAddr) {
             }
         } else if msg_type == "llm_error" {
             state.node_busy.lock().unwrap().remove(&node_id);
-            if let Some(tid) = json.get("task_id").and_then(|v| v.as_str()) {
+            state.node_active_task.lock().unwrap().remove(&node_id);
-                state.pending_task_ids.lock().unwrap().remove(tid);
+            let task_id = json.get("task_id").and_then(|v| v.as_str()).map(|s| s.to_string());
+            if let Some(ref tid) = task_id {
+                state.pending_task_ids.lock().unwrap().remove(tid.as_str());
             }
+            // Jos API-pyyntö odottaa, reititetään virhe oneshot-kanavaan
+            let api_sender = task_id.as_ref().and_then(|tid| {
+                state.pending_responses.lock().unwrap().remove(tid)
+            });
             {
                 let mut json = json;
                 if let Some(obj) = json.as_object_mut() {
                     obj.insert("node_id".to_string(), serde_json::json!(node_id));
                 }
+                if let Some(sender) = api_sender {
+                    let _ = sender.send(json.clone());
+                }
                 let _ = state.stats_tx.send(json.to_string());
             }
         } else if msg_type == "user_text" {
-            // Käyttäjän lähettämä teksti — broadcastataan pair_taskina ja llm_promptina
+            // Käyttäjän lähettämä teksti — kohdennettu reititys lähettäjäsolmulle
             let text = json.get("text").and_then(|v| v.as_str()).unwrap_or("").to_string();
             let task_type = json.get("task_type").and_then(|v| v.as_str()).unwrap_or("tokenize");
             if !text.is_empty() {
                 let preview: String = text.chars().take(80).collect();
                 tracing::info!("Solmu {} lähetti oman tekstin ({}): \"{}\"", node_id, task_type, preview);
-                match task_type {
+                let msg = match task_type {
-                    "tokenize" => {
+                    "tokenize" => serde_json::json!({
-                        let msg = serde_json::json!({
+                        "type": "single_tokenize",
-                            "type": "single_tokenize",
+                        "text": text,
-                            "text": text,
+                    }),
-                        });
+                    _ => serde_json::json!({
-                        let _ = state.stats_tx.send(msg.to_string());
+                        "type": "llm_prompt",
-                    }
+                        "prompt": text,
-                    _ => {
+                        "model": task_type,
-                        // LLM-prompti: lähetetään VAIN valitulle mallille, ei kaikille (välttää turhaa ruuhkaa ja busy-tiloja)
+                    }),
-                        let prompt = serde_json::json!({
+                };
-                            "type": "llm_prompt",
+                // Lähetetään takaisin lähettäjäsolmulle (käyttäjä haluaa oman tekstinsä tuloksen)
-                            "prompt": text,
+                if let Some(tx) = state.node_channels.read().await.get(&node_id) {
-                            "model": task_type,
+                    let _ = tx.send(msg.to_string());
-                        });
-                        let _ = state.stats_tx.send(prompt.to_string());
-                    }
                 }
             }
         }
@@ -972,6 +1140,22 @@ async fn handle_socket(socket: WebSocket, state: Arc<AppState>, ip: IpAddr) {
 
     // Yhteys katkesi — merkitään session päättyneeksi ja siivotaan atomisesti
     state.db.close_session(node_id);
+
+    // Jos solmulla oli kesken tehtävä, ilmoitetaan odottavalle API-kutsulle
+    let lost_task_id = state.node_active_task.lock().unwrap().remove(&node_id);
+    if let Some(tid) = lost_task_id {
+        tracing::warn!("Solmu {} katosi kesken tehtävän {} — palautetaan virhe API:lle", node_id, tid);
+        state.pending_task_ids.lock().unwrap().remove(&tid);
+        if let Some(resp_tx) = state.pending_responses.lock().unwrap().remove(&tid) {
+            let err = serde_json::json!({
+                "type": "llm_error",
+                "error": format!("Solmu #{} katosi kesken laskennan (task {})", node_id, tid),
+                "task_id": tid
+            });
+            let _ = resp_tx.send(err);
+        }
+    }
+
     {
         // Lukitaan kaikki kerralla, jotta solmu ei ole osittain siivottu
         let mut tasks = state.node_tasks.lock().unwrap();
@@ -989,6 +1173,9 @@ async fn handle_socket(socket: WebSocket, state: Arc<AppState>, ip: IpAddr) {
         vram.remove(&node_id);
     }
     state.node_types.lock().unwrap().remove(&node_id);
+    state.node_paused.lock().unwrap().remove(&node_id);
+    state.node_models.write().await.remove(&node_id);
+    state.node_max_param_b.write().await.remove(&node_id);
     tracing::info!("Solmu {} ({}) poistui verkosta.", node_id, ip);
     broadcast_stats(&state).await;
     sender_task.abort();
@@ -1000,6 +1187,18 @@ struct ChatCompletionRequest {
     task_id: String,
     #[serde(default)]
     max_tokens: Option<u64>,
+    #[serde(default)]
+    system_prompt: Option<String>,
+    #[serde(default)]
+    temperature: Option<f64>,
+    #[serde(default)]
+    top_k: Option<u64>,
+    #[serde(default)]
+    repeat_penalty: Option<f64>,
+    #[serde(default)]
+    stop: Option<Vec<String>>,
+    #[serde(default)]
+    capability: Option<String>, // "heavy" → priorisoi isoin malli, "light" → mikä tahansa
 }
 
 #[derive(serde::Serialize)]
@@ -1009,7 +1208,16 @@ struct ChatCompletionResponse {
     tokens_generated: u64,
 }
 
-async fn api_ollama_tags() -> axum::response::Response {
+async fn api_ollama_tags(
+    axum::extract::State(state): axum::extract::State<Arc<AppState>>,
+) -> axum::response::Response {
+    // Haetaan natiivisolmun tila muistista — priorisoidaan aito verkko-solmu
+    let node_models = state.node_models.read().await;
+    if let Some((_, models_json)) = node_models.iter().next() {
+        return axum::Json(models_json.clone()).into_response();
+    }
+
+    // Fallback: Haetaan lokaalista infra-Ollamasta ohjaimesta käsin (esim dev ympäristö)
     let ollama_url = std::env::var("OLLAMA_URL").unwrap_or_else(|_| "http://ollama:11434".to_string());
     match reqwest::get(format!("{}/api/tags", ollama_url)).await {
         Ok(resp) => {
@@ -1033,11 +1241,10 @@ async fn api_hardware(
     });
 
     let (mut vram_mb, mut gpu_name, ram_mb) = if let Some(s) = native {
-        let gpus = s.get("gpus").and_then(|v| v.as_array());
+        // Tieto on tietokannassa litteänä
-        let gpu = gpus.and_then(|g| g.first());
+        let vram = s.get("vram_total_mb").and_then(|v| v.as_u64()).unwrap_or(0);
-        let vram = gpu.and_then(|g| g.get("vram_total_mb")).and_then(|v| v.as_u64()).unwrap_or(0);
+        let name = s.get("gpu_name").and_then(|v| v.as_str()).unwrap_or("").to_string();
-        let name = gpu.and_then(|g| g.get("name")).and_then(|v| v.as_str()).unwrap_or("").to_string();
+        let ram = s.get("ram_mb").and_then(|v| v.as_u64()).unwrap_or(0);
-        let ram = s.get("system").and_then(|v| v.get("ram_total_mb")).and_then(|v| v.as_u64()).unwrap_or(0);
         (vram, name, ram)
     } else {
         (0, String::new(), 0)
@@ -1101,13 +1308,26 @@ async fn api_chat_completions(
         }
     }
 
-    // Etsitään vapaa solmu — priorisoidaan natiivisolmut (GPU) selaimen edelle
+    // Etsitään vapaa solmu — älykäs reititys kyvykkyyden mukaan
+    let want_heavy = payload.capability.as_deref() == Some("heavy");
+    // Haetaan param_b-snapshot ennen Mutex-lukituksia (async RwLock ei saa olla Mutex-scopen sisällä)
+    let param_b_snapshot: HashMap<u64, u32> = state.node_max_param_b.read().await.clone();
     let (target_node, _total_matching) = {
         let tasks = state.node_tasks.lock().unwrap();
-        let _busy = state.node_busy.lock().unwrap();
+        let busy = state.node_busy.lock().unwrap();
         let node_types = state.node_types.lock().unwrap();
-        let matching: Vec<u64> = tasks.iter().filter(|(_, task)| {
+        let paused = state.node_paused.lock().unwrap();
-            // Eksakti match tai qwen-perheen yhteensopivuus (selain: qwen-coder-05b, natiivi: qwen2.5-coder:7b)
+        // Debug: logita kaikki solmut ja niiden tilat
+        let all_nodes: Vec<String> = tasks.iter().map(|(id, task)| {
+            let ty = node_types.get(id).map(|s| s.as_str()).unwrap_or("?");
+            let b = if busy.contains(id) { " BUSY" } else { "" };
+            let p = if paused.contains(id) { " PAUSED" } else { "" };
+            format!("#{}({}:{}{}{}", id, ty, task, b, p)
+        }).collect();
+        tracing::info!("Reititys '{}'{} — solmut: [{}]", payload.model, if want_heavy { " (heavy)" } else { "" }, all_nodes.join(", "));
+        let matching: Vec<u64> = tasks.iter().filter(|(k, task)| {
+            if paused.contains(k) { return false; } // Ei tauotettuja
+            if busy.contains(k) { return false; }   // Ei varattuja
             let req_model = payload.model.to_lowercase();
             let node_task = task.to_lowercase();
             if req_model.starts_with("qwen") {
@@ -1118,11 +1338,32 @@ async fn api_chat_completions(
                 **task == payload.model
             }
         }).map(|(k, _)| *k).collect();
-        // Etsitään mikä tahansa matchaava solmu (natiivi priorisoidaan)
+
-        let native = matching.iter().find(|id| {
+        let any = if want_heavy {
-            node_types.get(id).map(|t| t == "native").unwrap_or(false)
+            // Heavy: priorisoi solmu jolla on suurin malli (B-parametrit)
-        }).copied();
+            let mut ranked: Vec<(u64, u32)> = matching.iter().map(|id| {
-        let any = native.or_else(|| matching.first().copied());
+                (*id, param_b_snapshot.get(id).copied().unwrap_or(0))
+            }).collect();
+            ranked.sort_by(|a, b| b.1.cmp(&a.1)); // suurin ensin
+            if let Some((best_id, best_b)) = ranked.first() {
+                tracing::info!("Heavy-reititys: solmu {} valittu ({}B parametria)", best_id, best_b);
+                Some(*best_id)
+            } else {
+                // Kaikki heavy-solmut busy — fallback mihin tahansa vapaaseen
+                let all_matching: Vec<u64> = tasks.iter().filter(|(k, task)| {
+                    if paused.contains(k) || busy.contains(k) { return false; }
+                    let req_model = payload.model.to_lowercase();
+                    task.to_lowercase().starts_with(&req_model.split('-').next().unwrap_or(""))
+                }).map(|(k, _)| *k).collect();
+                all_matching.first().copied()
+            }
+        } else {
+            // Oletus: vapaa natiivi ensin, sitten mikä tahansa vapaa
+            let native = matching.iter().find(|id| {
+                node_types.get(id).map(|t| t == "native").unwrap_or(false)
+            }).copied();
+            native.or_else(|| matching.first().copied())
+        };
         (any, matching.len())
     };
 
@@ -1149,6 +1390,7 @@ async fn api_chat_completions(
 
     // Merkitään solmu varatuksi ja task_id jaetuksi
     state.node_busy.lock().unwrap().insert(target_node_id);
+    state.node_active_task.lock().unwrap().insert(target_node_id, payload.task_id.clone());
     state.pending_task_ids.lock().unwrap().insert(payload.task_id.clone());
 
     let mut msg = serde_json::json!({
@@ -1157,12 +1399,17 @@ async fn api_chat_completions(
         "model": payload.model,
         "task_id": payload.task_id,
     });
-    if let Some(mt) = payload.max_tokens {
+    let obj = msg.as_object_mut().unwrap();
-        msg.as_object_mut().unwrap().insert("max_tokens".to_string(), serde_json::json!(mt));
+    if let Some(mt) = payload.max_tokens { obj.insert("max_tokens".to_string(), serde_json::json!(mt)); }
-    }
+    if let Some(ref sp) = payload.system_prompt { obj.insert("system_prompt".to_string(), serde_json::json!(sp)); }
+    if let Some(t) = payload.temperature { obj.insert("temperature".to_string(), serde_json::json!(t)); }
+    if let Some(k) = payload.top_k { obj.insert("top_k".to_string(), serde_json::json!(k)); }
+    if let Some(rp) = payload.repeat_penalty { obj.insert("repeat_penalty".to_string(), serde_json::json!(rp)); }
+    if let Some(ref s) = payload.stop { obj.insert("stop".to_string(), serde_json::json!(s)); }
 
-    // Odotuskanava valmiiksi (solmu palauttaa tuloksen stats_tx kautta)
+    // Oneshot-kanava: solmu palauttaa tuloksen suoraan tälle pyynnölle
-    let mut rx = state.stats_tx.subscribe();
+    let (resp_tx, resp_rx) = tokio::sync::oneshot::channel::<serde_json::Value>();
+    state.pending_responses.lock().unwrap().insert(payload.task_id.clone(), resp_tx);
 
     // Kohdennettu reititys: lähetetään AI-tehtävä suoraan VAIN valitulle solmulle
     {
@@ -1171,48 +1418,39 @@ async fn api_chat_completions(
             let _ = tx.send(msg.to_string());
             tracing::info!("Reititettiin API-pyyntö solmulle {} (Malli: {})", target_node_id, payload.model);
         } else {
+             state.pending_responses.lock().unwrap().remove(&payload.task_id);
              return (axum::http::StatusCode::SERVICE_UNAVAILABLE, "Verkkovirhe: solmun yhteys katkesi reitityksen aikana").into_response();
         }
     }
 
-    let timeout = tokio::time::timeout(std::time::Duration::from_secs(600), async move {
+    let timeout = tokio::time::timeout(std::time::Duration::from_secs(120), resp_rx).await;
-        loop {
-            let msg_str = match rx.recv().await {
-                Ok(msg) => msg,
-                Err(broadcast::error::RecvError::Lagged(n)) => {
-                    tracing::debug!("API-kanava lagged {} viestiä", n);
-                    continue;
-                }
-                Err(_) => return Ok(None), // Kanava suljettu
-            };
-            if let Ok(v) = serde_json::from_str::<serde_json::Value>(&msg_str) {
-                if v["type"].as_str() == Some("llm_done") {
-                    if let Some(tid) = v["task_id"].as_str() {
-                        if tid == payload.task_id {
-                            return Ok(Some(ChatCompletionResponse {
-                                response: v["response"].as_str().unwrap_or("").to_string(),
-                                model: v["model"].as_str().unwrap_or("").to_string(),
-                                tokens_generated: v["tokens_generated"].as_u64().unwrap_or(0),
-                            }));
-                        }
-                    }
-                } else if v["type"].as_str() == Some("llm_error") {
-                    if let Some(tid) = v["task_id"].as_str() {
-                        if tid == payload.task_id {
-                            return Err(v["error"].as_str().unwrap_or("Määrittelemätön virhe solmussa").to_string());
-                        }
-                    }
-                }
-            }
-        }
-        #[allow(unreachable_code)]
-        Ok(None)
-    }).await;
 
     match timeout {
-        Ok(Ok(Some(res))) => axum::Json(res).into_response(),
+        Ok(Ok(v)) => {
-        Ok(Ok(None)) => (axum::http::StatusCode::INTERNAL_SERVER_ERROR, "Verkkovirhe: yhteys katkesi").into_response(),
+            if v["type"].as_str() == Some("llm_error") {
-        Ok(Err(err)) => (axum::http::StatusCode::CONFLICT, err).into_response(),
+                let err = v["error"].as_str().unwrap_or("Määrittelemätön virhe solmussa").to_string();
-        Err(_) => (axum::http::StatusCode::GATEWAY_TIMEOUT, "Aikakatkaisu: solmu ei saanut tehtävää ajoissa valmiiksi").into_response(),
+                (axum::http::StatusCode::CONFLICT, err).into_response()
+            } else {
+                axum::Json(ChatCompletionResponse {
+                    response: v["response"].as_str().unwrap_or("").to_string(),
+                    model: v["model"].as_str().unwrap_or("").to_string(),
+                    tokens_generated: v["tokens_generated"].as_u64().unwrap_or(0),
+                }).into_response()
+            }
+        }
+        Ok(Err(_)) => {
+            // Oneshot-kanava sulkeutui (solmu katosi kesken laskennan)
+            state.pending_responses.lock().unwrap().remove(&payload.task_id);
+            state.node_busy.lock().unwrap().remove(&target_node_id);
+            state.node_active_task.lock().unwrap().remove(&target_node_id);
+            (axum::http::StatusCode::SERVICE_UNAVAILABLE, "Solmu katosi kesken laskennan — yritä uudelleen").into_response()
+        }
+        Err(_) => {
+            // Timeout — solmu ei vastannut ajoissa
+            state.pending_responses.lock().unwrap().remove(&payload.task_id);
+            state.node_busy.lock().unwrap().remove(&target_node_id);
+            state.node_active_task.lock().unwrap().remove(&target_node_id);
+            (axum::http::StatusCode::GATEWAY_TIMEOUT, "Aikakatkaisu: solmu ei saanut tehtävää ajoissa valmiiksi").into_response()
+        }
     }
 }

network-poc/install.sh

@@ -1,59 +0,0 @@
-#!/bin/bash
-# Kipinä Agentic Studio — asennusskripti (Debian/Ubuntu)
-set -e
-
-echo "=== Kipinä Agentic Studio — Asennus ==="
-echo ""
-
-# Tarkistetaan käyttöjärjestelmä
-if [ ! -f /etc/debian_version ]; then
-    echo "⚠ Tämä skripti on suunniteltu Debian/Ubuntu-järjestelmille."
-    echo "  Muilla jakeluilla voit asentaa riippuvuudet manuaalisesti."
-    read -p "  Jatketaanko? (k/e) " -n 1 -r; echo
-    [[ $REPLY =~ ^[Kk]$ ]] || exit 1
-fi
-
-echo "[1/6] Päivitetään pakettilistaus..."
-sudo apt-get update -qq
-
-echo "[2/6] Asennetaan peruspaketteja..."
-sudo apt-get install -y -qq curl git build-essential pkg-config libssl-dev
-
-# Rust
-if command -v rustc &>/dev/null; then
-    echo "[3/6] Rust löytyi: $(rustc --version)"
-else
-    echo "[3/6] Asennetaan Rust..."
-    curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
-    source "$HOME/.cargo/env"
-fi
-
-# Node.js (Astro-frontend vaatii)
-if command -v node &>/dev/null; then
-    echo "[4/6] Node.js löytyi: $(node --version)"
-else
-    echo "[4/6] Asennetaan Node.js 22..."
-    curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
-    sudo apt-get install -y -qq nodejs
-fi
-
-# Ollama
-if command -v ollama &>/dev/null; then
-    echo "[5/6] Ollama löytyi"
-else
-    echo "[5/6] Asennetaan Ollama..."
-    curl -fsSL https://ollama.ai/install.sh | sh
-fi
-
-# Malli
-echo "[6/6] Ladataan kielimalli (qwen2.5-coder:3b)..."
-ollama pull qwen2.5-coder:3b
-
-echo ""
-echo "=== Asennus valmis! ==="
-echo ""
-echo "Käynnistä:"
-echo "  cd $(pwd)"
-echo "  ./network-poc/local.sh"
-echo ""
-echo "Avaa selaimessa: http://localhost:3000"

network-poc/kipina-node

@@ -0,0 +1,135 @@
+#!/bin/bash
+# Kipinä Node — lataa oikea binääri ja käynnistä
+set -e
+
+BASE_URL="https://kipina.studio/download"
+HUB_URL="${KIPINA_HUB:-wss://kipina.studio/ws}"
+OLLAMA_URL="${OLLAMA_URL:-http://localhost:11434}"
+
+# Tunnista OS ja arkkitehtuuri
+OS=$(uname -s | tr '[:upper:]' '[:lower:]')
+ARCH=$(uname -m)
+
+case "$OS-$ARCH" in
+    darwin-arm64)  BINARY="kipina-node-macos-arm64" ;;
+    darwin-x86_64) BINARY="kipina-node-macos-arm64" ;; # Rosetta
+    linux-x86_64)  BINARY="kipina-node-linux-x86_64" ;;
+    linux-aarch64) BINARY="kipina-node-linux-arm64" ;;
+    *) echo "Ei tuettu: $OS-$ARCH"; exit 1 ;;
+esac
+
+echo ""
+echo "  ╔══════════════════════════════════════╗"
+echo "  ║       Kipinä Agentic Node            ║"
+echo "  ╚══════════════════════════════════════╝"
+echo ""
+echo "  OS: $OS ($ARCH)"
+echo ""
+
+# Etsi Ollama-instanssit
+CANDIDATES=(
+    "http://localhost:11434"
+    "http://127.0.0.1:11434"
+    "http://ollama:11434"
+    "http://host.docker.internal:11434"
+)
+
+# Lisää OLLAMA_URL listaan jos asetettu ja ei jo mukana
+if [ -n "$OLLAMA_URL" ]; then
+    ALREADY=false
+    for c in "${CANDIDATES[@]}"; do
+        [ "$c" = "$OLLAMA_URL" ] && ALREADY=true
+    done
+    $ALREADY || CANDIDATES=("$OLLAMA_URL" "${CANDIDATES[@]}")
+fi
+
+echo "  Etsitään Ollama-instansseja..."
+FOUND=()
+for url in "${CANDIDATES[@]}"; do
+    if curl -s --connect-timeout 1 "$url/api/tags" &>/dev/null; then
+        FOUND+=("$url")
+    fi
+done
+
+if [ ${#FOUND[@]} -eq 0 ]; then
+    # Ei löytynyt — yritä käynnistää lokaali
+    if command -v ollama &>/dev/null; then
+        echo "  Käynnistetään Ollama..."
+        ollama serve &>/dev/null &
+        sleep 3
+        if curl -s --connect-timeout 1 "http://localhost:11434/api/tags" &>/dev/null; then
+            OLLAMA_URL="http://localhost:11434"
+            echo "  ✓ Ollama käynnistetty ($OLLAMA_URL)"
+        else
+            echo "  ✗ Ollaman käynnistys epäonnistui."
+            exit 1
+        fi
+    else
+        echo ""
+        echo "  ✗ Ollamaa ei löytynyt."
+        echo "    Kontti/remote: OLLAMA_URL=http://HOST:11434 ./kipina-node"
+        echo "    Asenna:        curl -fsSL https://ollama.ai/install.sh | sh"
+        exit 1
+    fi
+elif [ ${#FOUND[@]} -eq 1 ]; then
+    OLLAMA_URL="${FOUND[0]}"
+    echo "  ✓ Ollama löytyi: $OLLAMA_URL"
+else
+    echo ""
+    echo "  Löytyi ${#FOUND[@]} Ollama-instanssia:"
+    echo ""
+    for i in "${!FOUND[@]}"; do
+        echo "    $((i+1))) ${FOUND[$i]}"
+    done
+    echo ""
+    read -p "  Valitse [1-${#FOUND[@]}]: " -r CHOICE
+    if [[ "$CHOICE" =~ ^[0-9]+$ ]] && [ "$CHOICE" -ge 1 ] && [ "$CHOICE" -le ${#FOUND[@]} ]; then
+        OLLAMA_URL="${FOUND[$((CHOICE-1))]}"
+    else
+        OLLAMA_URL="${FOUND[0]}"
+        echo "  Käytetään oletusta: $OLLAMA_URL"
+    fi
+    echo "  ✓ Valittu: $OLLAMA_URL"
+fi
+
+echo ""
+echo "  Hub: $HUB_URL"
+echo "  Ollama: $OLLAMA_URL"
+if [ -n "$KIPINA_MODEL" ]; then
+    echo "  Malli: $KIPINA_MODEL (Ympäristömuuttujasta)"
+fi
+
+# Binäärin automaattinen päivitys — vertaa build-hashia palvelimeen
+BIN_PATH="./kipina-node-bin"
+HASH_PATH="./kipina-node-bin.hash"
+
+REMOTE_HASH=$(curl -sSL "$BASE_URL/.build-hash?v=$(date +%s)" 2>/dev/null | tr -d '[:space:]')
+LOCAL_HASH=""
+[ -f "$HASH_PATH" ] && LOCAL_HASH=$(cat "$HASH_PATH" | tr -d '[:space:]')
+
+if [ -f "$BIN_PATH" ] && [ -n "$REMOTE_HASH" ] && [ "$REMOTE_HASH" = "$LOCAL_HASH" ]; then
+    echo "  ✓ Binääri ajan tasalla (versio: $LOCAL_HASH)"
+else
+    if [ -f "$BIN_PATH" ]; then
+        echo "  ↻ Uusi versio saatavilla ($LOCAL_HASH → $REMOTE_HASH)"
+    else
+        echo "  Ladataan $BINARY..."
+    fi
+    rm -f "$BIN_PATH"
+    curl -sSL "$BASE_URL/$BINARY?v=$(date +%s)" -o "$BIN_PATH"
+    chmod +x "$BIN_PATH"
+    echo "$REMOTE_HASH" > "$HASH_PATH"
+    echo "  ✓ Päivitetty versioon $REMOTE_HASH"
+fi
+
+echo ""
+echo "  ✓ Siirrytään Kipinä Noden hallintaan..."
+echo "    Ctrl+C pysäyttää"
+echo ""
+
+if [ -n "$KIPINA_MODEL" ]; then
+    export OLLAMA_MODEL="$KIPINA_MODEL"
+fi
+export HUB_URL="$HUB_URL"
+export OLLAMA_URL="$OLLAMA_URL"
+exec "$BIN_PATH"

network-poc/local.sh

@@ -1,37 +0,0 @@
-#!/bin/bash
-set -e
-
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-echo "=== Kipinä Studio Local Development ==="
-
-# Frontend
-echo "[1/3] Rakennetaan frontend..."
-cd "$SCRIPT_DIR/frontend"
-[ -d node_modules ] || npm install --silent
-npm run build --silent 2>&1 | tail -1
-
-# Hub
-echo "[2/3] Käynnistetään hub..."
-cd "$SCRIPT_DIR/hub"
-cargo run &
-HUB_PID=$!
-sleep 3
-
-# Native-node (jos Ollama on käynnissä)
-if curl -s http://localhost:11434/api/tags >/dev/null 2>&1; then
-    echo "[3/3] Ollama löytyi — käynnistetään native-node..."
-    cd "$SCRIPT_DIR/native-node"
-    HUB_URL=ws://localhost:3000/ws cargo run --no-default-features &
-    NODE_PID=$!
-    echo "      Native-node PID: $NODE_PID"
-else
-    echo "[3/3] Ollama ei käynnissä — käytetään selaimen Wasm-laskentaa"
-    echo "      Nopeampi: ollama serve & ollama pull qwen2.5-coder:7b && ./local.sh"
-fi
-
-echo ""
-echo "=== http://localhost:3000 ==="
-echo "    Ctrl+C pysäyttää"
-
-# Odotetaan hub-prosessia
-wait $HUB_PID

network-poc/native-node/Cargo.toml

@@ -19,3 +19,8 @@ wgpu = { version = "24", optional = true }
 reqwest = { version = "0.12", features = ["json"] }
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
+dialoguer = "0.12.0"
+ratatui = "0.29.0"
+crossterm = { version = "0.28.1", features = ["event-stream"] }
+tracing-appender = "0.2.4"
+chrono = "0.4"

network-poc/native-node/src/inference.rs

@@ -1,6 +1,15 @@
 use std::time::Instant;
 use std::cell::RefCell;
 
+pub struct GenerateOptions {
+    pub max_tokens: usize,
+    pub system_prompt: Option<String>,
+    pub temperature: Option<f64>,
+    pub top_k: Option<u64>,
+    pub repeat_penalty: Option<f64>,
+    pub stop: Option<Vec<String>>,
+}
+
 pub struct LlmEngine {
     ollama_url: String,
     model: RefCell<String>,
@@ -9,8 +18,6 @@ pub struct LlmEngine {
 
 impl LlmEngine {
     pub async fn load() -> Result<Self, String> {
-        let model = std::env::var("OLLAMA_MODEL").unwrap_or_else(|_| "qwen2.5-coder:3b".to_string());
-
         let client = reqwest::Client::builder()
             .timeout(std::time::Duration::from_secs(600))
             .connect_timeout(std::time::Duration::from_secs(3))
@@ -48,6 +55,12 @@ impl LlmEngine {
             })
         };
 
+        // Kysytään malli TUI:lla jos ei pakotettu ympäristöstä
+        let model = match std::env::var("OLLAMA_MODEL") {
+            Ok(m) if !m.is_empty() => m,
+            _ => crate::tui::select_model(&ollama_url, &client).await?
+        };
+
         tracing::info!("Ollama backend: {} | malli: {}", ollama_url, model);
         Ok(LlmEngine { ollama_url, model: RefCell::new(model), client })
     }
@@ -56,6 +69,10 @@ impl LlmEngine {
         self.model.borrow().clone()
     }
 
+    pub fn ollama_url(&self) -> &str {
+        &self.ollama_url
+    }
+
     pub fn set_model(&self, new_model: String) {
         *self.model.borrow_mut() = new_model;
     }
@@ -78,28 +95,82 @@ impl LlmEngine {
         }
     }
 
-    pub async fn generate(&self, prompt: &str, max_tokens: usize) -> Result<GenerateResult, String> {
+    /// Hakee käynnissä olevan mallin VRAM-tilan (ollama ps)
-        // System prompt tulee agentin konfiguraatiosta (frontend lähettää sen osana promptia).
+    pub async fn fetch_ps(&self) -> Result<Option<ModelVramStatus>, String> {
-        // Tässä ei yliajeta sitä — Ollama saa vain prompt-kentän.
+        let resp = self.client.get(format!("{}/api/ps", self.ollama_url))
-        let model = self.model.borrow().clone();
-
-        let start = Instant::now();
-        let resp = self.client.post(format!("{}/api/generate", self.ollama_url))
-            .json(&serde_json::json!({
-                "model": model,
-                "prompt": prompt,
-                "stream": false,
-                "options": {
-                    "num_predict": max_tokens,
-                    "temperature": 0.7,
-                    "top_k": 40,
-                    "repeat_penalty": 1.15,
-                    "stop": ["<|im_end|>", "\n###", "\nExplanation", "\nNote:", "\nPlease note", "\nThis is", "\n```\n\n", "\n// Example", "\n# Example"]
-                }
-            }))
             .send()
             .await
-            .map_err(|e| format!("Ollama generate: {}", e))?;
+            .map_err(|e| format!("Ollama ps: {}", e))?;
+
+        if !resp.status().is_success() {
+            return Err(format!("Ollama ps HTTP {}", resp.status()));
+        }
+
+        let body: serde_json::Value = resp.json().await
+            .map_err(|e| format!("Ollama ps json: {}", e))?;
+
+        let models = body["models"].as_array();
+        if let Some(arr) = models {
+            if let Some(m) = arr.first() {
+                let name = m["name"].as_str().unwrap_or("?").to_string();
+                let size = m["size"].as_u64().unwrap_or(0);
+                let size_vram = m["size_vram"].as_u64().unwrap_or(0);
+                return Ok(Some(ModelVramStatus { name, size, size_vram }));
+            }
+        }
+        Ok(None) // ei ladattua mallia
+    }
+
+    /// Hakee kaikki Ollamaan asennetut mallit
+    pub async fn fetch_models(&self) -> Result<serde_json::Value, String> {
+        let resp = self.client.get(format!("{}/api/tags", self.ollama_url))
+            .send()
+            .await
+            .map_err(|e| format!("Ollama tags fetch: {}", e))?;
+
+        if resp.status().is_success() {
+            resp.json().await.map_err(|e| format!("Ollama tags json: {}", e))
+        } else {
+            Err(format!("Ollama tags epäonnistui: {}", resp.status()))
+        }
+    }
+
+    pub async fn generate(&self, prompt: &str, opts: &GenerateOptions) -> Result<GenerateResult, String> {
+        let model = self.model.borrow().clone();
+
+        let default_stop: Vec<String> = vec![
+            "<|im_end|>".into(),
+        ];
+
+        // Rakennetaan messages-lista (chat API)
+        let mut messages = Vec::new();
+        if let Some(ref sp) = opts.system_prompt {
+            if !sp.is_empty() {
+                messages.push(serde_json::json!({"role": "system", "content": sp}));
+            }
+        }
+        messages.push(serde_json::json!({"role": "user", "content": prompt}));
+
+        let body = serde_json::json!({
+            "model": model,
+            "messages": messages,
+            "stream": false,
+            "options": {
+                "num_ctx": 16384,
+                "num_predict": opts.max_tokens,
+                "temperature": opts.temperature.unwrap_or(0.7),
+                "top_k": opts.top_k.unwrap_or(40),
+                "repeat_penalty": opts.repeat_penalty.unwrap_or(1.15),
+                "stop": opts.stop.as_ref().unwrap_or(&default_stop),
+            }
+        });
+
+        let start = Instant::now();
+        let resp = self.client.post(format!("{}/api/chat", self.ollama_url))
+            .json(&body)
+            .send()
+            .await
+            .map_err(|e| format!("Ollama chat: {}", e))?;
 
         if !resp.status().is_success() {
             return Err(format!("Ollama HTTP {}", resp.status()));
@@ -108,7 +179,7 @@ impl LlmEngine {
         let body: serde_json::Value = resp.json().await
             .map_err(|e| format!("Ollama JSON: {}", e))?;
 
-        let text = body["response"].as_str().unwrap_or("").to_string();
+        let text = body["message"]["content"].as_str().unwrap_or("").to_string();
         let _total_duration_ns = body["total_duration"].as_u64().unwrap_or(0);
         let eval_count = body["eval_count"].as_u64().unwrap_or(0) as usize;
         let eval_duration_ns = body["eval_duration"].as_u64().unwrap_or(1);
@@ -127,40 +198,15 @@ impl LlmEngine {
     }
 }
 
-/// Siivoa markdown-koodiblokki-merkit ja selitystekstit
+/// Siivoa markdown-koodiblokki-merkit vastauksesta
 fn strip_code_fences(text: &str) -> String {
-    // Poistetaan kaikki ```-rivit ja kielitunnisteet (```python, ```rust jne.)
     let lines: Vec<&str> = text.lines().collect();
     let filtered: Vec<&str> = lines.into_iter().filter(|line| {
         let trimmed = line.trim();
         // Poista rivit jotka ovat pelkkiä ``` tai ```kielitunniste
-        if trimmed.starts_with("```") {
+        trimmed != "```" && !(trimmed.starts_with("```") && !trimmed[3..].contains('`'))
-            return false;
-        }
-        true
     }).collect();
-    let mut result = filtered.join("\n").trim().to_string();
+    filtered.join("\n").trim().to_string()
-
-    // Poista selitysteksti lopusta (kaikki rivin "\nPlease note" jälkeen jne.)
-    let lower = result.to_lowercase();
-    for stop in &["\nplease note", "\nthis is a basic", "\nthis code", "\nnote that", "\nremember to", "\nyou can", "\nto run"] {
-        if let Some(pos) = lower.find(stop) {
-            result = result[..pos].trim_end().to_string();
-        }
-    }
-
-    // Poista johdantolauseet alusta
-    let lower = result.to_lowercase();
-    for prefix in &["sure!", "here is", "here's", "certainly!", "below is"] {
-        if lower.starts_with(prefix) {
-            if let Some(nl) = result.find('\n') {
-                result = result[nl + 1..].to_string();
-            }
-            break;
-        }
-    }
-
-    result.trim().to_string()
 }
 
 pub struct GenerateResult {
@@ -169,3 +215,32 @@ pub struct GenerateResult {
     pub duration_ms: f64,
     pub tokens_per_sec: f64,
 }
+
+pub struct ModelVramStatus {
+    pub name: String,
+    pub size: u64,      // kokonaiskoko (tavuina)
+    pub size_vram: u64, // VRAM:ssa oleva osuus (tavuina)
+}
+
+impl ModelVramStatus {
+    pub fn fully_in_vram(&self) -> bool {
+        self.size > 0 && self.size_vram >= self.size
+    }
+
+    pub fn vram_percent(&self) -> f64 {
+        if self.size == 0 { return 0.0; }
+        (self.size_vram as f64 / self.size as f64) * 100.0
+    }
+
+    pub fn display(&self) -> String {
+        let size_gb = self.size as f64 / 1_073_741_824.0;
+        let vram_gb = self.size_vram as f64 / 1_073_741_824.0;
+        if self.fully_in_vram() {
+            format!("✓ {} ({:.1} GB) — 100% GPU", self.name, size_gb)
+        } else if self.size_vram == 0 {
+            format!("✗ {} ({:.1} GB) — 100% CPU", self.name, size_gb)
+        } else {
+            format!("◐ {} ({:.1}/{:.1} GB VRAM, {:.0}% GPU)", self.name, vram_gb, size_gb, self.vram_percent())
+        }
+    }
+}

network-poc/native-node/src/main.rs

@@ -1,10 +1,13 @@
 use futures_util::{SinkExt, StreamExt};
 use serde_json::json;
+use std::io::IsTerminal;
 use sysinfo::System;
 use tokio_tungstenite::connect_async;
 use tokio_tungstenite::tungstenite::Message;
 
 mod inference;
+mod tui;
+mod tui_dashboard;
 
 /// GPU-tietorakenne — yhtenäinen kaikille valmistajille
 struct GpuInfo {
@@ -222,7 +225,7 @@ fn collect_system_info() -> serde_json::Value {
 }
 
 /// Koko auth-viesti hubille
-fn build_auth_message(allocated_gb: u32) -> String {
+fn build_auth_message(allocated_gb: u32, model_name: &str, models_data: Option<serde_json::Value>) -> String {
     let sys = collect_system_info();
     let gpus = collect_all_gpus();
 
@@ -239,7 +242,7 @@ fn build_auth_message(allocated_gb: u32) -> String {
         "status": "agent_ready",
         "node_type": "native",
         "allocated_gb": allocated_gb,
-        "selected_task": "qwen2.5-coder:7b",
+        "selected_task": model_name,
         "system": sys,
     });
 
@@ -251,6 +254,10 @@ fn build_auth_message(allocated_gb: u32) -> String {
         msg.as_object_mut().unwrap().insert("gpus".to_string(), json!(gpu_json));
     }
 
+    if let Some(models) = models_data {
+        msg.as_object_mut().unwrap().insert("models".to_string(), models);
+    }
+
     msg.to_string()
 }
 
@@ -263,10 +270,24 @@ fn format_optional<T: std::fmt::Display>(val: Option<T>, suffix: &str) -> String
 
 #[tokio::main]
 async fn main() {
+    let file_appender = tracing_appender::rolling::never(".", "native-node.log");
+    let (non_blocking, _guard) = tracing_appender::non_blocking(file_appender);
+    
     tracing_subscriber::fmt()
         .with_env_filter("native_node=debug")
+        .with_writer(non_blocking)
         .init();
 
+    // Hookataan paniikkitilanteet palauttamaan terminaalin raw-moodista
+    let original_hook = std::panic::take_hook();
+    std::panic::set_hook(Box::new(move |panic_info| {
+        tui_dashboard::restore_terminal();
+        original_hook(panic_info);
+    }));
+
+    let tui_state = std::sync::Arc::new(tokio::sync::RwLock::new(tui_dashboard::DashboardState::new()));
+    let (cmd_tx, mut cmd_rx) = tokio::sync::mpsc::unbounded_channel::<String>();
+
     let hub_url = std::env::var("HUB_URL").unwrap_or_else(|_| "ws://hub:3000/ws".to_string());
     let allocated_gb: u32 = std::env::var("ALLOCATED_GB")
         .ok()
@@ -283,6 +304,18 @@ async fn main() {
         sys["ram_total_mb"]
     );
     
+    {
+        let mut st = tui_state.write().await;
+        st.sys_info = format!("{} | {} | {} ydintä | {} MB RAM",
+            sys["hostname"].as_str().unwrap_or("?"),
+            sys["os"].as_str().unwrap_or("?"),
+            sys["cpu_cores"],
+            sys["ram_total_mb"]
+        );
+        let i = st.sys_info.clone();
+        st.push_log("System", format!("Järjestelmä: {}", i), None);
+    }
+
     let gpus = collect_all_gpus();
     if gpus.is_empty() {
         #[cfg(not(feature = "gpu-detect"))]
@@ -321,6 +354,82 @@ async fn main() {
         }
     };
 
+    let active_model = llm.as_ref().map(|e| e.model_name()).unwrap_or_else(|| "unknown".to_string());
+    tracing::info!("Käytettävä kielimalli konfiguroitu (selected_task): {}", active_model);
+    
+    {
+        let mut st = tui_state.write().await;
+        st.model_name = active_model.clone();
+        st.push_log("System", format!("Malli valmis: {}", active_model), None);
+    }
+
+    // Lämmittelykutsu: ladataan malli VRAM:iin ja haetaan VRAM-tila
+    if let Some(ref engine) = llm {
+        {
+            let mut st = tui_state.write().await;
+            st.vram_status = "Ladataan VRAM:iin...".to_string();
+            st.push_log("System", "Ladataan mallia VRAM:iin...".to_string(), None);
+        }
+        // Lyhyt generate-kutsu pakottaa Ollaman lataamaan mallin GPU:lle
+        let _ = engine.generate("hi", &inference::GenerateOptions {
+            max_tokens: 1, system_prompt: None, temperature: Some(0.0),
+            top_k: Some(1), repeat_penalty: None, stop: None,
+        }).await;
+        if let Ok(Some(ps)) = engine.fetch_ps().await {
+            let mut st = tui_state.write().await;
+            st.vram_status = ps.display();
+            st.push_log("System", format!("VRAM: {}", ps.display()), None);
+        }
+        let vram_engine_url = engine.ollama_url().to_string();
+        let vram_state = tui_state.clone();
+        tokio::spawn(async move {
+            let client = reqwest::Client::new();
+            loop {
+                tokio::time::sleep(std::time::Duration::from_secs(30)).await;
+                if let Ok(resp) = client.get(format!("{}/api/ps", vram_engine_url)).send().await {
+                    if let Ok(body) = resp.json::<serde_json::Value>().await {
+                        if let Some(arr) = body["models"].as_array() {
+                            if let Some(m) = arr.first() {
+                                let name = m["name"].as_str().unwrap_or("?").to_string();
+                                let size = m["size"].as_u64().unwrap_or(0);
+                                let size_vram = m["size_vram"].as_u64().unwrap_or(0);
+                                let status = inference::ModelVramStatus { name, size, size_vram };
+                                vram_state.write().await.vram_status = status.display();
+                            } else {
+                                vram_state.write().await.vram_status = "Ei ladattua mallia".to_string();
+                            }
+                        }
+                    }
+                }
+            }
+        });
+    }
+
+    // Käynnistetään graafinen TUI vain jos stdin on terminaali (ei taustaprosessina)
+    let ui_state = tui_state.clone();
+    if std::io::stdin().is_terminal() {
+        tokio::spawn(async move {
+            if let Err(e) = tui_dashboard::run_dashboard(ui_state, cmd_tx).await {
+                tracing::error!("Pääluupin TUI kaatui: {}", e);
+            }
+        });
+    } else {
+        tracing::info!("Ei terminaalia — TUI ohitettu, lokitetaan stdoutiin");
+    };
+
+    // Haetaan paikalliset mallit hubille lähetettäväksi
+    let mut available_models = None;
+    if let Some(ref engine) = llm {
+        match engine.fetch_models().await {
+            Ok(models) => {
+                available_models = Some(models);
+            }
+            Err(e) => {
+                tracing::warn!("Mallilistauksen haku epäonnistui: {}", e);
+            }
+        }
+    }
+
     // Yhdistetään hubiin
     loop {
         match connect_async(&hub_url).await {
@@ -328,79 +437,266 @@ async fn main() {
                 tracing::info!("Yhdistetty hubiin!");
                 let (mut write, mut read) = ws_stream.split();
 
-                let auth = build_auth_message(allocated_gb);
+                let auth = build_auth_message(allocated_gb, &active_model, available_models.clone());
                 if write.send(Message::Text(auth)).await.is_err() {
                     tracing::error!("Auth-viestin lähetys epäonnistui");
                     continue;
                 }
 
-                while let Some(Ok(msg)) = read.next().await {
+                // Merkitään yhdistetyksi TUI:ssa
-                    if let Message::Text(text) = msg {
+                {
-                        // LLM-promptit
+                    let mut st = tui_state.write().await;
-                        if text.contains("llm_prompt") {
+                    st.status = "ACTIVE".to_string();
-                            if let Ok(task) = serde_json::from_str::<serde_json::Value>(&text) {
+                    st.push_log("Network", "Yhdistetty hubiin".to_string(), None);
-                                let prompt = task.get("prompt").and_then(|v| v.as_str()).unwrap_or("");
+                }
-                                let task_id = task.get("task_id").and_then(|v| v.as_str()).unwrap_or("?");
-                                let msg_model = task.get("model").and_then(|v| v.as_str()).unwrap_or("");
-                                
-                                if !prompt.is_empty() && (msg_model.starts_with("qwen-coder") || msg_model.starts_with("qwen2.5-coder")) {
 
+                loop {
+                    tokio::select! {
+                        cmd = cmd_rx.recv() => {
+                            if let Some(cmd_str) = cmd {
+                                if cmd_str == "pause" {
+                                    tracing::info!("Tauotetaan solmun suoritus (Hub ei lähetä tehtäviä)...");
+                                    let req = json!({"type": "status_update", "status": "paused"});
+                                    let _ = write.send(Message::Text(req.to_string())).await;
+                                    {
+                                        let mut st = tui_state.write().await;
+                                        st.status = "PAUSED".to_string();
+                                        st.push_log("Network", "Solmu siirretty taukotilaan".to_string(), None);
+                                    }
+                                } else if cmd_str == "resume" {
+                                    tracing::info!("Jatketaan solmun suoritusta...");
+                                    let req = json!({"type": "status_update", "status": "active"});
+                                    let _ = write.send(Message::Text(req.to_string())).await;
+                                    {
+                                        let mut st = tui_state.write().await;
+                                        st.status = "ACTIVE".to_string();
+                                        st.push_log("System", "Suoritus jatkuu...".to_string(), None);
+                                    }
+                                } else if cmd_str == "fetch_models" {
+                                    // Haetaan mallit Ollamasta ja avataan valikkö
                                     if let Some(ref engine) = llm {
-                                        let max_tokens = task.get("max_tokens").and_then(|v| v.as_u64()).unwrap_or(1024) as usize;
+                                        match engine.fetch_models().await {
-                                        tracing::info!("Generoidaan (task_id: {}, max_tokens: {}): \"{}\"", task_id, max_tokens, &prompt[..prompt.len().min(100)]);
+                                            Ok(tags) => {
-
+                                                let models: Vec<String> = tags.get("models")
-                                        let model_name = engine.model_name();
+                                                    .and_then(|v| v.as_array())
-                                        match engine.generate(prompt, max_tokens).await {
+                                                    .map(|arr| arr.iter()
-                                            Ok(result) => {
+                                                        .filter_map(|m| m.get("name").and_then(|n| n.as_str()).map(|s| s.to_string()))
-                                                tracing::info!(
+                                                        .collect())
-                                                    "Tulos: {} tokenia | {:.0}ms | {:.1} tok/s | \"{}\"",
+                                                    .unwrap_or_default();
-                                                    result.tokens_generated,
+                                                let mut st = tui_state.write().await;
-                                                    result.duration_ms,
+                                                st.model_picker_items = models;
-                                                    result.tokens_per_sec,
+                                                st.model_picker_idx = 0;
-                                                    &result.text[..result.text.len().min(80)]
+                                                st.model_picker_open = true;
-                                                );
-
-                                                let done = json!({
-                                                    "type": "llm_done",
-                                                    "prompt": prompt,
-                                                    "model": format!("{} (Ollama)", model_name),
-                                                    "response": result.text,
-                                                    "tokens_generated": result.tokens_generated,
-                                                    "duration_ms": result.duration_ms,
-                                                    "tokens_per_sec": (result.tokens_per_sec * 10.0).round() / 10.0,
-                                                    "load_time_ms": 0,
-                                                    "task_id": task_id,
-                                                });
-                                                let _ = write.send(Message::Text(done.to_string())).await;
                                             }
                                             Err(e) => {
-                                                tracing::error!("Inferenssivirhe: {}", e);
+                                                let mut st = tui_state.write().await;
+                                                st.push_log("System", format!("Mallilistan haku epäonnistui: {}", e), None);
+                                            }
+                                        }
+                                    }
+                                } else if let Some(model) = cmd_str.strip_prefix("change_model:") {
+                                    // TUI:sta valittu malli — vaihdetaan
+                                    if let Some(ref engine) = llm {
+                                        engine.set_model(model.to_string());
+                                        match engine.ensure_model().await {
+                                            Ok(()) => {
+                                                tracing::info!("Malli vaihdettu: {}", model);
+                                                let mut st = tui_state.write().await;
+                                                st.model_name = model.to_string();
+                                                st.push_log("System", format!("Malli vaihdettu: {}", model), None);
+                                                // Ilmoitetaan hubille
+                                                let auth = build_auth_message(allocated_gb, model, available_models.clone());
+                                                let _ = write.send(Message::Text(auth)).await;
+                                            }
+                                            Err(e) => {
+                                                let mut st = tui_state.write().await;
+                                                st.push_log("System", format!("Mallin vaihto epäonnistui: {}", e), None);
                                             }
                                         }
                                     }
                                 }
                             }
                         }
-                        // Mallin vaihto lennossa
+                        ws_msg = read.next() => {
-                        if text.contains("change_model") {
+                            match ws_msg {
-                            if let Ok(task) = serde_json::from_str::<serde_json::Value>(&text) {
+                                Some(Ok(Message::Text(text))) => {
-                                if let Some(new_model) = task.get("model").and_then(|v| v.as_str()) {
+                                    // Hubin control-viestit
-                                    if let Some(ref engine) = llm {
+                                    if text.contains(r#""type":"control""#) {
-                                        tracing::info!("Vaihdetaan malli: {}", new_model);
+                                        if let Ok(task) = serde_json::from_str::<serde_json::Value>(&text) {
-                                        engine.set_model(new_model.to_string());
+                                            if let Some(action) = task.get("action").and_then(|v| v.as_str()) {
-                                        match engine.ensure_model().await {
+                                                if action == "pause" {
-                                            Ok(()) => tracing::info!("Malli {} valmis!", new_model),
+                                                    tracing::info!("Hub pakotti solmun tauolle (Pause)");
-                                            Err(e) => tracing::error!("Mallin lataus epäonnistui: {}", e),
+                                                    let req = json!({"type": "status_update", "status": "paused"});
+                                                    let _ = write.send(Message::Text(req.to_string())).await;
+                                                    {
+                                                        let mut st = tui_state.write().await;
+                                                        st.status = "PAUSED".to_string();
+                                                        st.push_log("Network", "Hub kytki solmun tauolle".to_string(), None);
+                                                    }
+                                                } else if action == "resume" {
+                                                    tracing::info!("Hub aktivoi solmun suorituksen (Resume)");
+                                                    let req = json!({"type": "status_update", "status": "active"});
+                                                    let _ = write.send(Message::Text(req.to_string())).await;
+                                                    {
+                                                        let mut st = tui_state.write().await;
+                                                        st.status = "ACTIVE".to_string();
+                                                        st.push_log("Network", "Hub palautti solmun töihin".to_string(), None);
+                                                    }
+                                                }
+                                            }
+                                        }
+                                    }
+                                    // Node joined → oma node_id
+                                    if text.contains(r#""type":"node_joined""#) {
+                                        if let Ok(msg) = serde_json::from_str::<serde_json::Value>(&text) {
+                                            if let Some(nid) = msg.get("node_id").and_then(|v| v.as_u64()) {
+                                                let mut st = tui_state.write().await;
+                                                if st.node_id.is_none() {
+                                                    st.node_id = Some(nid);
+                                                    st.push_log("Network", format!("Node ID: #{}", nid), None);
+                                                }
+                                            }
+                                        }
+                                    }
+                                    // Verkon globaali tila
+                                    if text.contains(r#""type":"network_status""#) {
+                                        if let Ok(status) = serde_json::from_str::<serde_json::Value>(&text) {
+                                            if let Some(nodes) = status.get("active_nodes").and_then(|v| v.as_u64()) {
+                                                if let Some(tasks) = status.get("tasks").and_then(|v| v.as_u64()) {
+                                                    let mut st = tui_state.write().await;
+                                                    st.network_active_nodes = nodes as usize;
+                                                    st.network_total_tasks = tasks;
+                                                }
+                                            }
+                                        }
+                                    }
+
+                                    // LLM-promptit
+                                    if text.contains("llm_prompt") {
+                                        if let Ok(task) = serde_json::from_str::<serde_json::Value>(&text) {
+                                            let prompt = task.get("prompt").and_then(|v| v.as_str()).unwrap_or("");
+                                            let task_id = task.get("task_id").and_then(|v| v.as_str()).unwrap_or("?");
+                                            let msg_model = task.get("model").and_then(|v| v.as_str()).unwrap_or("");
+
+                                            if !prompt.is_empty() && (msg_model.starts_with("qwen-coder") || msg_model.starts_with("qwen2.5-coder") || msg_model.starts_with("phi")) {
+                                                if let Some(ref engine) = llm {
+                                                    let gen_opts = inference::GenerateOptions {
+                                                        max_tokens: task.get("max_tokens").and_then(|v| v.as_u64()).unwrap_or(1024) as usize,
+                                                        system_prompt: task.get("system_prompt").and_then(|v| v.as_str()).map(|s| s.to_string()),
+                                                        temperature: task.get("temperature").and_then(|v| v.as_f64()),
+                                                        top_k: task.get("top_k").and_then(|v| v.as_u64()),
+                                                        repeat_penalty: task.get("repeat_penalty").and_then(|v| v.as_f64()),
+                                                        stop: task.get("stop").and_then(|v| v.as_array()).map(|a| a.iter().filter_map(|s| s.as_str().map(|s| s.to_string())).collect()),
+                                                    };
+                                                    let prompt_lines = prompt.lines().count();
+                                                    let prompt_last: String = prompt.lines().last().unwrap_or("").chars().take(60).collect();
+                                                    tracing::info!("→ task_id:{} | {}r prompti | \"{}...\"", task_id, prompt_lines, prompt_last);
+                                                    {
+                                                        let mut st = tui_state.write().await;
+                                                        st.cur_task_id = Some(task_id.to_string());
+                                                        st.cur_prompt = Some(format!("→ {} riviä | \"{}...\"", prompt_lines, prompt_last));
+                                                    }
+
+                                                    let model_name = engine.model_name();
+                                                    match engine.generate(prompt, &gen_opts).await {
+                                                        Ok(result) => {
+                                                            let tokens_sec = (result.tokens_per_sec * 10.0).round() / 10.0;
+                                                            tracing::info!(
+                                                                "✓ {} | {} tok | {:.0}ms | {:.1} tok/s",
+                                                                model_name,
+                                                                result.tokens_generated,
+                                                                result.duration_ms,
+                                                                tokens_sec,
+                                                            );
+                                                            {
+                                                                let mut st = tui_state.write().await;
+                                                                st.tasks_completed += 1;
+                                                                st.last_tokens_sec = tokens_sec as f64;
+                                                                st.cur_task_id = None;
+                                                                st.cur_prompt = None;
+                                                                
+                                                                let msg_type = if task_id == "status-check" { "Ping" } else { "Task" };
+                                                                let msg_text = format!("{}  ({} tok)", task_id, result.tokens_generated);
+                                                                st.push_log(msg_type, msg_text, Some(tokens_sec as f64));
+                                                            }
+                                                            let prompt_short: String = prompt.lines().last().unwrap_or("").chars().take(100).collect();
+                                                            let done = json!({
+                                                                "type": "llm_done",
+                                                                "prompt": prompt_short,
+                                                                "model": format!("{} (Ollama)", model_name),
+                                                                "response": result.text,
+                                                                "tokens_generated": result.tokens_generated,
+                                                                "duration_ms": result.duration_ms,
+                                                                "tokens_per_sec": tokens_sec,
+                                                                "load_time_ms": 0,
+                                                                "task_id": task_id,
+                                                            });
+                                                            let _ = write.send(Message::Text(done.to_string())).await;
+                                                        }
+                                                        Err(e) => {
+                                                            tracing::error!("Inferenssivirhe: {}", e);
+                                                            {
+                                                                let mut st = tui_state.write().await;
+                                                                st.cur_task_id = None;
+                                                                st.cur_prompt = None;
+                                                                st.push_log("System", format!("Virhe inferenssissä: {}", e), None);
+                                                            }
+                                                        }
+                                                    }
+                                                }
+                                            }
+                                        }
+                                    }
+
+                                    // Mallin vaihto lennossa
+                                    if text.contains("change_model") {
+                                        if let Ok(task) = serde_json::from_str::<serde_json::Value>(&text) {
+                                            if let Some(new_model) = task.get("model").and_then(|v| v.as_str()) {
+                                                if let Some(ref engine) = llm {
+                                                    tracing::info!("Vaihdetaan malli: {}", new_model);
+                                                    engine.set_model(new_model.to_string());
+                                                    match engine.ensure_model().await {
+                                                        Ok(()) => {
+                                                            tracing::info!("Malli {} valmis!", new_model);
+                                                            let mut st = tui_state.write().await;
+                                                            st.model_name = new_model.to_string();
+                                                            st.push_log("System", format!("Malli {} ladattu & valmis!", new_model), None);
+                                                        }
+                                                        Err(e) => tracing::error!("Mallin lataus epäonnistui: {}", e),
+                                                    }
+                                                }
+                                            }
                                         }
                                     }
                                 }
+                                Some(Ok(_)) => {} // Muut viestityypit (binary/ping)
+                                Some(Err(_)) | None => break, // Yhteys poikki
                             }
                         }
                     }
                 }
+                
+                // Yhteys katkesi — nollataan TUI:n busy-tila
+                {
+                    let mut st = tui_state.write().await;
+                    let lost_task = st.cur_task_id.clone();
+                    if let Some(tid) = lost_task {
+                        st.push_log("Network", format!("Tehtävä {} keskeytyi yhteyden katketessa", tid), None);
+                    }
+                    st.cur_task_id = None;
+                    st.cur_prompt = None;
+                    st.node_id = None;
+                    st.status = "RECONNECTING".to_string();
+                    st.push_log("Network", "Yhteys hubiin katkesi — yhdistetään uudelleen 5s...".to_string(), None);
+                }
                 tracing::warn!("Yhteys hubiin katkesi — yritetään uudelleen 5s...");
             }
             Err(e) => {
+                {
+                    let mut st = tui_state.write().await;
+                    st.status = "RECONNECTING".to_string();
+                    st.push_log("Network", format!("Yhdistäminen epäonnistui: {} — yritetään 5s...", e), None);
+                }
                 tracing::warn!("Hubiin yhdistäminen epäonnistui: {} — yritetään uudelleen 5s...", e);
             }
         }

network-poc/native-node/src/tui.rs

@@ -0,0 +1,67 @@
+use dialoguer::{Select, Input, theme::ColorfulTheme};
+use reqwest::Client;
+
+pub async fn select_model(ollama_url: &str, client: &Client) -> Result<String, String> {
+    // 1. Hae tagit
+    let mut models = vec![];
+    println!("  Haetaan asennettuja malleja osoitteesta {}...", ollama_url);
+    if let Ok(resp) = client.get(&format!("{}/api/tags", ollama_url)).send().await {
+        if resp.status().is_success() {
+            if let Ok(json) = resp.json::<serde_json::Value>().await {
+                if let Some(arr) = json.get("models").and_then(|v| v.as_array()) {
+                    for m in arr {
+                        if let Some(name) = m.get("name").and_then(|v| v.as_str()) {
+                            models.push(name.to_string());
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    let download_opt = "[➕ Lataa uusi malli internetistä]";
+    let mut options = vec![download_opt.to_string()];
+    options.extend(models);
+
+    // 2. Kysy käyttäjältä Selectillä
+    let theme = ColorfulTheme::default();
+    let selection = Select::with_theme(&theme)
+        .with_prompt("Valitse Ollama-malli Kipinä-verkkoa varten:")
+        .default(if options.len() > 1 { 1 } else { 0 })
+        .items(&options)
+        .interact()
+        .map_err(|e| format!("TUI virhe: {}", e))?;
+
+    let selected = &options[selection];
+
+    // 3. Jos käyttäjä haluaa uuden, kysy nimeä
+    if selected == download_opt {
+        let new_model: String = Input::with_theme(&theme)
+            .with_prompt("Syötä ladattavan mallin nimi (esim. llama3 tai qwen2.5-coder:3b)")
+            .interact_text()
+            .map_err(|e| format!("TUI virhe: {}", e))?;
+        
+        let new_model = new_model.trim().to_string();
+        if new_model.is_empty() {
+            return Err("Mallin nimi ei voi olla tyhjä".to_string());
+        }
+        
+        println!("  Ladataan malleja taustalla... Tämä voi kestää hetken ({})", new_model);
+        // Odotetaan että pull on valmis
+        let pull_body = serde_json::json!({ "name": &new_model });
+        let resp = client.post(&format!("{}/api/pull", ollama_url))
+            .json(&pull_body)
+            .send()
+            .await
+            .map_err(|e| format!("Pull req virhe: {}", e))?;
+            
+        if resp.status().is_success() {
+            println!("  ✓ Malli {} ladattu onnistuneesti!", new_model);
+            return Ok(new_model);
+        } else {
+            return Err(format!("Ollama pull epäonnistui: {}", resp.status()));
+        }
+    }
+
+    Ok(selected.clone())
+}

network-poc/native-node/src/tui_dashboard.rs

@@ -0,0 +1,330 @@
+use crossterm::{
+    event::{Event, EventStream, KeyCode},
+    execute,
+    terminal::{disable_raw_mode, enable_raw_mode, EnterAlternateScreen, LeaveAlternateScreen},
+};
+use ratatui::{
+    backend::CrosstermBackend,
+    layout::{Constraint, Direction, Layout, Alignment},
+    style::{Color, Modifier, Style},
+    widgets::{Block, Borders, Paragraph, Wrap},
+    Terminal,
+};
+use std::io;
+use tokio::sync::RwLock;
+use std::sync::Arc;
+use futures_util::StreamExt;
+use std::time::Duration;
+
+#[derive(Clone)]
+pub struct LogEntry {
+    pub ty: String,
+    pub msg: String,
+    pub speed: Option<f64>,
+    pub timestamp: String,
+}
+
+pub struct DashboardState {
+    pub logs: Vec<LogEntry>,
+    pub status: String,
+    pub node_id: Option<u64>,
+    pub sys_info: String,
+    pub model_name: String,
+    pub cur_task_id: Option<String>,
+    pub cur_prompt: Option<String>,
+    pub tasks_completed: u32,
+    pub last_tokens_sec: f64,
+    pub network_active_nodes: usize,
+    pub network_total_tasks: u64,
+    // VRAM-tila (ollama ps)
+    pub vram_status: String,
+    // Mallivalikko
+    pub model_picker_open: bool,
+    pub model_picker_items: Vec<String>,
+    pub model_picker_idx: usize,
+}
+
+impl DashboardState {
+    pub fn new() -> Self {
+        Self {
+            logs: Vec::new(),
+            status: "ACTIVE".to_string(),
+            node_id: None,
+            sys_info: "".to_string(),
+            model_name: "Yhdistetään...".to_string(),
+            cur_task_id: None,
+            cur_prompt: None,
+            tasks_completed: 0,
+            last_tokens_sec: 0.0,
+            network_active_nodes: 1, // oletetaan itsemme
+            network_total_tasks: 0,
+            vram_status: "Haetaan...".to_string(),
+            model_picker_open: false,
+            model_picker_items: Vec::new(),
+            model_picker_idx: 0,
+        }
+    }
+
+    pub fn push_log(&mut self, ty: &str, msg: String, speed: Option<f64>) {
+        let now = chrono::Local::now().format("%H:%M:%S").to_string();
+        self.logs.push(LogEntry {
+            timestamp: now,
+            ty: ty.to_string(),
+            msg,
+            speed,
+        });
+        if self.logs.len() > 100 {
+            self.logs.remove(0);
+        }
+    }
+}
+
+pub async fn run_dashboard(
+    state: Arc<RwLock<DashboardState>>,
+    cmd_tx: tokio::sync::mpsc::UnboundedSender<String>,
+) -> Result<(), io::Error> {
+    enable_raw_mode()?;
+    let mut stdout = io::stdout();
+    execute!(stdout, EnterAlternateScreen)?;
+    let backend = CrosstermBackend::new(stdout);
+    let mut terminal = Terminal::new(backend)?;
+    terminal.clear()?;
+
+    let mut reader = EventStream::new();
+    let mut interval = tokio::time::interval(Duration::from_millis(100));
+
+    loop {
+        tokio::select! {
+            _ = interval.tick() => {
+                let st = state.read().await;
+                terminal.draw(|f| ui(f, &st))?;
+            }
+            ev = reader.next() => {
+                if let Some(Ok(Event::Key(key))) = ev {
+                    let picker_open = state.read().await.model_picker_open;
+
+                    if picker_open {
+                        // Mallivalikko auki — navigointi
+                        match key.code {
+                            KeyCode::Up | KeyCode::Char('k') => {
+                                let mut st = state.write().await;
+                                if st.model_picker_idx > 0 { st.model_picker_idx -= 1; }
+                            }
+                            KeyCode::Down | KeyCode::Char('j') => {
+                                let mut st = state.write().await;
+                                let max = st.model_picker_items.len().saturating_sub(1);
+                                if st.model_picker_idx < max { st.model_picker_idx += 1; }
+                            }
+                            KeyCode::Enter => {
+                                let mut st = state.write().await;
+                                let idx = st.model_picker_idx;
+                                if let Some(model) = st.model_picker_items.get(idx).cloned() {
+                                    st.model_picker_open = false;
+                                    st.push_log("System", format!("Vaihdetaan malliin: {}...", model), None);
+                                    let _ = cmd_tx.send(format!("change_model:{}", model));
+                                }
+                            }
+                            KeyCode::Esc | KeyCode::Char('m') | KeyCode::Char('M') => {
+                                state.write().await.model_picker_open = false;
+                            }
+                            _ => {}
+                        }
+                    } else {
+                        // Normaali tila
+                        match key.code {
+                            KeyCode::Char('q') | KeyCode::Esc => {
+                                disable_raw_mode()?;
+                                execute!(terminal.backend_mut(), LeaveAlternateScreen)?;
+                                std::process::exit(0);
+                            }
+                            KeyCode::Char('p') | KeyCode::Char('P') => {
+                                let _ = cmd_tx.send("pause".to_string());
+                            }
+                            KeyCode::Char('r') | KeyCode::Char('R') | KeyCode::Char('s') => {
+                                let _ = cmd_tx.send("resume".to_string());
+                            }
+                            KeyCode::Char('m') | KeyCode::Char('M') => {
+                                let _ = cmd_tx.send("fetch_models".to_string());
+                            }
+                            _ => {}
+                        }
+                    }
+                }
+            }
+        }
+    }
+}
+
+pub fn restore_terminal() {
+    let _ = disable_raw_mode();
+    let _ = execute!(io::stdout(), LeaveAlternateScreen);
+}
+
+fn ui(f: &mut ratatui::Frame, st: &DashboardState) {
+    let chunks = Layout::default()
+        .direction(Direction::Vertical)
+        .constraints([
+            Constraint::Length(3),  // Header
+            Constraint::Min(0),     // Body
+            Constraint::Length(3),  // Footer / Status
+        ].as_ref())
+        .split(f.area());
+
+    // --- Header ---
+    let header_text = match st.node_id {
+        Some(id) => format!(" Kipinä Agentic Node #{} ", id),
+        None => " Kipinä Agentic Node (Yhdistää...) ".to_string(),
+    };
+    let header = Paragraph::new(header_text)
+        .style(Style::default().fg(Color::Cyan).add_modifier(Modifier::BOLD))
+        .alignment(Alignment::Center)
+        .block(Block::default().borders(Borders::ALL).style(Style::default().fg(Color::DarkGray)));
+    f.render_widget(header, chunks[0]);
+
+    // --- Body ---
+    let body_chunks = Layout::default()
+        .direction(Direction::Vertical)
+        .constraints([
+            Constraint::Length(8), // Yläosan info ja tehtävä
+            Constraint::Min(0),    // Lokit / Chat alas
+        ].as_ref())
+        .split(chunks[1]);
+
+    let top_panels = Layout::default()
+        .direction(Direction::Horizontal)
+        .constraints([
+            Constraint::Percentage(40), // Vasen paneeli (Info)
+            Constraint::Percentage(60), // Oikea paneeli (Tehtävä)
+        ].as_ref())
+        .split(body_chunks[0]);
+
+    // Vasen paneeli: Laitteisto, Malli & Verkosto — VRAM-rivi värikoodattu
+    let vram_color = if st.vram_status.starts_with('✓') {
+        Color::Green
+    } else if st.vram_status.starts_with('◐') {
+        Color::Yellow
+    } else if st.vram_status.starts_with('✗') {
+        Color::Red
+    } else {
+        Color::DarkGray
+    };
+
+    let info_lines = vec![
+        ratatui::text::Line::from(vec![
+            ratatui::text::Span::raw("🚀 Malli: "),
+            ratatui::text::Span::styled(&st.model_name, Style::default().fg(Color::Cyan).add_modifier(Modifier::BOLD)),
+        ]),
+        ratatui::text::Line::from(vec![
+            ratatui::text::Span::raw("🎮 VRAM: "),
+            ratatui::text::Span::styled(&st.vram_status, Style::default().fg(vram_color)),
+        ]),
+        ratatui::text::Line::from(vec![
+            ratatui::text::Span::raw("💻 Järjestelmä: "),
+            ratatui::text::Span::styled(&st.sys_info, Style::default().fg(Color::White)),
+        ]),
+        ratatui::text::Line::from(format!(
+            "📊 Tehdyt: {} | Nopeus: {:.1} t/s", st.tasks_completed, st.last_tokens_sec
+        )),
+        ratatui::text::Line::from(format!(
+            "🌐 Verkosto: {} solmua | {} tehtävää", st.network_active_nodes, st.network_total_tasks
+        )),
+    ];
+    let left_panel = Paragraph::new(info_lines)
+        .block(Block::default().title(" Laitteisto ja AI ").borders(Borders::ALL))
+        .style(Style::default().fg(Color::White))
+        .wrap(Wrap { trim: true });
+    f.render_widget(left_panel, top_panels[0]);
+
+    // Oikea paneeli: Käynnissä oleva tehtävä
+    let task_title = match &st.cur_task_id {
+        Some(id) => format!(" Työn alla: {} ", id),
+        None => " Vapaana ".to_string(),
+    };
+    let task_content = st.cur_prompt.clone().unwrap_or_else(|| "Odotetaan tehtäviä Hubilta...".to_string());
+    
+    let task_style = if st.cur_task_id.is_some() {
+        Style::default().fg(Color::Magenta)
+    } else {
+        Style::default().fg(Color::DarkGray)
+    };
+    
+    let task_panel = Paragraph::new(task_content)
+        .wrap(Wrap { trim: true })
+        .block(Block::default().title(task_title).borders(Borders::ALL).style(task_style));
+    f.render_widget(task_panel, top_panels[1]);
+
+    // Alaosan paneeli: Tapahtumaloki koko leveydeltä
+    let area_height = body_chunks[1].height.saturating_sub(2) as usize;
+    let skip_count = if st.logs.len() > area_height { st.logs.len() - area_height } else { 0 };
+    
+    let visible_logs: Vec<ratatui::text::Line> = st.logs.iter().skip(skip_count).map(|log| {
+        let ty_color = match log.ty.as_str() {
+            "System" => Color::Yellow,
+            "Network" => Color::Blue,
+            "Task" => Color::Magenta,
+            "Ping" => Color::DarkGray,
+            _ => Color::White,
+        };
+        
+        let speed_str = if let Some(s) = log.speed {
+            format!(" | {:.1} tok/s", s)
+        } else {
+            "".to_string()
+        };
+        
+        ratatui::text::Line::from(vec![
+            ratatui::text::Span::styled(&log.timestamp, Style::default().fg(Color::DarkGray)),
+            ratatui::text::Span::raw(" "),
+            ratatui::text::Span::styled(format!("{: <8}", log.ty), Style::default().fg(ty_color).add_modifier(Modifier::BOLD)),
+            ratatui::text::Span::raw(" | "),
+            ratatui::text::Span::styled(log.msg.clone(), Style::default().fg(Color::White)),
+            ratatui::text::Span::styled(speed_str, Style::default().fg(ty_color)),
+        ])
+    }).collect();
+
+    let logs_panel = Paragraph::new(visible_logs)
+        .block(Block::default().title(" Tapahtumaloki ").borders(Borders::ALL).style(Style::default().fg(Color::Cyan)));
+    f.render_widget(logs_panel, body_chunks[1]);
+
+    // --- Footer / Status ---
+    let status_color = if st.status == "ACTIVE" { Color::Green } else { Color::Yellow };
+    let status_text = format!(" Tila: {} | [P] Pause [R] Työhön [M] Malli [Q] Sulje ", st.status);
+    let footer = Paragraph::new(status_text)
+        .style(Style::default().fg(status_color).add_modifier(Modifier::BOLD))
+        .alignment(Alignment::Center)
+        .block(Block::default().borders(Borders::ALL));
+    f.render_widget(footer, chunks[2]);
+
+    // --- Mallivalikko-overlay ---
+    if st.model_picker_open && !st.model_picker_items.is_empty() {
+        let area = f.area();
+        let popup_h = (st.model_picker_items.len() as u16 + 4).min(area.height - 4);
+        let popup_w = 50.min(area.width - 4);
+        let popup = ratatui::layout::Rect::new(
+            (area.width - popup_w) / 2,
+            (area.height - popup_h) / 2,
+            popup_w,
+            popup_h,
+        );
+
+        // Tausta
+        f.render_widget(ratatui::widgets::Clear, popup);
+
+        let items: Vec<ratatui::text::Line> = st.model_picker_items.iter().enumerate().map(|(i, name)| {
+            if i == st.model_picker_idx {
+                ratatui::text::Line::from(format!(" ▸ {} ", name))
+                    .style(Style::default().fg(Color::Cyan).add_modifier(Modifier::BOLD))
+            } else {
+                ratatui::text::Line::from(format!("   {} ", name))
+                    .style(Style::default().fg(Color::White))
+            }
+        }).collect();
+
+        let picker = Paragraph::new(items)
+            .block(Block::default()
+                .title(" Vaihda malli [↑↓] Enter=valitse Esc=peruuta ")
+                .borders(Borders::ALL)
+                .style(Style::default().fg(Color::Cyan)));
+        f.render_widget(picker, popup);
+    }
+}

network-poc/node/Cargo.toml

@@ -10,32 +10,22 @@ crate-type = ["cdylib"]
 wasm-bindgen = "0.2.91"
 js-sys = "0.3.68"
 web-sys = { version = "0.3.68", features = [
-    "Window",
-    "Document",
-    "HtmlElement",
     "WebSocket",
     "MessageEvent",
     "Performance",
     "console",
-    "Request",
-    "RequestInit",
     "Response",
-    "Headers",
     "ReadableStream",
     "ReadableStreamDefaultReader",
 ] }
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
-burn = { version = "0.14.0", features = ["wgpu", "ndarray"] }
-burn-wgpu = "0.14.0"
-burn-ndarray = "0.14.0"
 wasm-bindgen-futures = "0.4"
 console_error_panic_hook = "0.1.7"
 reqwest = { version = "0.12", default-features = false, features = ["json"] }
 tokenizers = { version = "0.19.1", default-features = false, features = ["unstable_wasm"] }
 rexie = "0.6"
-log = "0.4"
+candle-core = "0.8"
-candle-core = { version = "0.8" }
 candle-nn = "0.8"
 candle-transformers = "0.8"
 getrandom = { version = "0.3", features = ["wasm_js"] }

network-poc/node/src/burn_smollm/attention.rs

@@ -1,118 +0,0 @@
-use burn::module::{Module, Param};
-use burn::tensor::{backend::Backend, Tensor};
-use super::rope::RoPE;
-use super::config::SmolLMConfig;
-
-#[derive(Clone, Debug)]
-pub struct KVCache<B: Backend> {
-    pub k: Tensor<B, 4>,
-    pub v: Tensor<B, 4>,
-}
-
-#[derive(Module, Debug)]
-pub struct Attention<B: Backend> {
-    pub q_proj: Param<Tensor<B, 2>>,  // [hidden, num_heads * head_dim]
-    pub k_proj: Param<Tensor<B, 2>>,  // [hidden, num_kv_heads * head_dim]
-    pub v_proj: Param<Tensor<B, 2>>,  // [hidden, num_kv_heads * head_dim]
-    pub o_proj: Param<Tensor<B, 2>>,  // [num_heads * head_dim, hidden]
-
-    num_heads: usize,
-    num_kv_heads: usize,
-    head_dim: usize,
-
-    rope: RoPE<B>,
-}
-
-impl<B: Backend> Attention<B> {
-    pub fn new(config: &SmolLMConfig, device: &B::Device) -> Self {
-        let head_dim = config.hidden_size / config.num_attention_heads;
-
-        Self {
-            q_proj: Param::from_tensor(Tensor::zeros([config.hidden_size, config.num_attention_heads * head_dim], device)),
-            k_proj: Param::from_tensor(Tensor::zeros([config.hidden_size, config.num_key_value_heads * head_dim], device)),
-            v_proj: Param::from_tensor(Tensor::zeros([config.hidden_size, config.num_key_value_heads * head_dim], device)),
-            o_proj: Param::from_tensor(Tensor::zeros([config.num_attention_heads * head_dim, config.hidden_size], device)),
-
-            num_heads: config.num_attention_heads,
-            num_kv_heads: config.num_key_value_heads,
-            head_dim,
-
-            rope: RoPE::new(head_dim, config.max_position_embeddings, config.rope_theta, device),
-        }
-    }
-
-    pub fn forward(
-        &self,
-        x: Tensor<B, 3>,
-        offset: usize,
-        cache: Option<KVCache<B>>
-    ) -> (Tensor<B, 3>, KVCache<B>) {
-        let [batch, seq_len, hidden_dim] = x.dims();
-
-        // Project Q, K, V: x @ W -> [batch, seq, proj_dim]
-        let q = x.clone().matmul(self.q_proj.val().unsqueeze());
-        let k = x.clone().matmul(self.k_proj.val().unsqueeze());
-        let v = x.matmul(self.v_proj.val().unsqueeze());
-
-        // Reshape: [batch, seq, heads, head_dim] -> [batch, heads, seq, head_dim]
-        let q = q.reshape([batch, seq_len, self.num_heads, self.head_dim]).swap_dims(1, 2);
-        let k = k.reshape([batch, seq_len, self.num_kv_heads, self.head_dim]).swap_dims(1, 2);
-        let v = v.reshape([batch, seq_len, self.num_kv_heads, self.head_dim]).swap_dims(1, 2);
-
-        // Apply RoPE
-        let q = self.rope.forward(q, offset);
-        let k = self.rope.forward(k, offset);
-
-        // KV cache
-        let (k, v) = if let Some(c) = cache {
-            (Tensor::cat(vec![c.k, k], 2), Tensor::cat(vec![c.v, v], 2))
-        } else {
-            (k, v)
-        };
-
-        let new_cache = KVCache { k: k.clone(), v: v.clone() };
-        let kv_len = k.dims()[2];
-
-        // GQA: repeat K,V heads — [batch, kv_heads, kv_len, hd] -> [batch, num_heads, kv_len, hd]
-        let num_reps = self.num_heads / self.num_kv_heads;
-        let k = if num_reps > 1 {
-            let [b, kv_h, s, hd] = k.dims();
-            k.reshape([b, kv_h, 1, s, hd]).repeat_dim(2, num_reps).reshape([b, self.num_heads, s, hd])
-        } else { k };
-        let v = if num_reps > 1 {
-            let [b, kv_h, s, hd] = v.dims();
-            v.reshape([b, kv_h, 1, s, hd]).repeat_dim(2, num_reps).reshape([b, self.num_heads, s, hd])
-        } else { v };
-
-        // Attention: Q @ K^T / sqrt(d)
-        let scale = 1.0 / (self.head_dim as f64).sqrt();
-        let scores = q.matmul(k.swap_dims(2, 3)).mul_scalar(scale);
-        // scores: [batch, heads, seq_len, kv_len]
-
-        // Causal mask for prefill (seq_len > 1)
-        let scores = if seq_len > 1 {
-            let mask_data: Vec<f32> = (0..seq_len).flat_map(|i| {
-                (0..kv_len).map(move |j| {
-                    if j > offset + i { f32::NEG_INFINITY } else { 0.0 }
-                })
-            }).collect();
-            let mask = Tensor::<B, 2>::from_data(
-                burn::tensor::TensorData::new(mask_data, [seq_len, kv_len]),
-                &scores.device()
-            ).reshape([1, 1, seq_len, kv_len]);
-            scores + mask
-        } else {
-            scores
-        };
-
-        let attn_weights = burn::tensor::activation::softmax(scores, 3);
-
-        let context = attn_weights.matmul(v);
-        // [batch, heads, seq, hd] -> [batch, seq, heads*hd]
-        let context = context.swap_dims(1, 2).reshape([batch, seq_len, self.num_heads * self.head_dim]);
-
-        let output = context.matmul(self.o_proj.val().unsqueeze());
-
-        (output, new_cache)
-    }
-}

network-poc/node/src/burn_smollm/config.rs

@@ -1,28 +0,0 @@
-#[derive(Clone, Debug)]
-pub struct SmolLMConfig {
-    pub hidden_size: usize,
-    pub intermediate_size: usize,
-    pub vocab_size: usize,
-    pub num_hidden_layers: usize,
-    pub num_attention_heads: usize,
-    pub num_key_value_heads: usize,
-    pub rms_norm_eps: f64,
-    pub rope_theta: f32,
-    pub max_position_embeddings: usize,
-}
-
-impl Default for SmolLMConfig {
-    fn default() -> Self {
-        Self {
-            hidden_size: 576,
-            intermediate_size: 1536,
-            vocab_size: 49152,
-            num_hidden_layers: 30,
-            num_attention_heads: 9,
-            num_key_value_heads: 3,
-            rms_norm_eps: 1e-5,
-            rope_theta: 10000.0,
-            max_position_embeddings: 2048,
-        }
-    }
-}

network-poc/node/src/burn_smollm/loader.rs

@@ -1,90 +0,0 @@
-use burn::tensor::{backend::Backend, Tensor, TensorData};
-use candle_core::safetensors;
-use candle_core::Device as CandleDevice;
-use burn::module::Param;
-use super::model::LlamaModel;
-use super::config::SmolLMConfig;
-
-fn load_tensor_2d<B: Backend>(
-    tensors_map: &std::collections::HashMap<String, candle_core::Tensor>,
-    name: &str,
-    device: &B::Device,
-    shape_out_in: [usize; 2]
-) -> Result<Param<Tensor<B, 2>>, String> {
-    let t = tensors_map.get(name).ok_or_else(|| format!("Puuttuu: {}", name))?;
-    let t = t.to_dtype(candle_core::DType::F32).unwrap();
-    let vec = t.flatten_all().unwrap().to_vec1::<f32>().unwrap();
-    let t_burn = Tensor::<B, 2>::from_data(burn::tensor::TensorData::new(vec, shape_out_in), device);
-    // transpose from [out, in] to [in, out]
-    Ok(Param::from_tensor(t_burn.transpose()))
-}
-
-fn load_tensor_1d<B: Backend>(
-    tensors_map: &std::collections::HashMap<String, candle_core::Tensor>,
-    name: &str,
-    device: &B::Device,
-    _shape: [usize; 1]
-) -> Result<Param<Tensor<B, 1>>, String> {
-    let t = tensors_map.get(name).ok_or_else(|| format!("Puuttuu: {}", name))?;
-    let t = t.to_dtype(candle_core::DType::F32).unwrap();
-    let vec = t.flatten_all().unwrap().to_vec1::<f32>().unwrap();
-    Ok(Param::from_tensor(Tensor::<B, 1>::from_floats(vec.as_slice(), device)))
-}
-
-fn load_embed<B: Backend>(
-    tensors_map: &std::collections::HashMap<String, candle_core::Tensor>,
-    name: &str,
-    device: &B::Device,
-    shape: [usize; 2]
-) -> Result<Param<Tensor<B, 2>>, String> {
-    let t = tensors_map.get(name).ok_or_else(|| format!("Puuttuu: {}", name))?;
-    let t = t.to_dtype(candle_core::DType::F32).unwrap();
-    let vec = t.flatten_all().unwrap().to_vec1::<f32>().unwrap();
-    // Embed ei transponoi samalla tavalla, se pysyy [vocab, hidden]
-    Ok(Param::from_tensor(Tensor::<B, 2>::from_data(burn::tensor::TensorData::new(vec, shape), device)))
-}
-
-pub fn load_safetensors_to_model<B: Backend>(
-    buffer: &[u8], 
-    config: &SmolLMConfig, 
-    device: &B::Device
-) -> Result<LlamaModel<B>, String> {
-    
-    let mut model = LlamaModel::new(config, device);
-    let tensors_map = safetensors::load_buffer(buffer, &CandleDevice::Cpu)
-        .map_err(|e| format!("Virhe Safetensors luennassa: {}", e))?;
-    
-    // Embeddings
-    model.embed_tokens = load_embed(&tensors_map, "model.embed_tokens.weight", device, [config.vocab_size, config.hidden_size])?;
-    model.norm.weight = load_tensor_1d(&tensors_map, "model.norm.weight", device, [config.hidden_size])?;
-    model.lm_head = load_embed(&tensors_map, "lm_head.weight", device, [config.vocab_size, config.hidden_size]).or_else(|_| {
-        load_embed(&tensors_map, "model.embed_tokens.weight", device, [config.vocab_size, config.hidden_size])
-    })?;
-    
-    let head_dim = config.hidden_size / config.num_attention_heads;
-
-    for i in 0..config.num_hidden_layers {
-        let prefix = format!("model.layers.{}", i);
-        
-        let layer = &mut model.layers[i];
-        
-        // Norms
-        layer.input_layernorm.weight = load_tensor_1d(&tensors_map, &format!("{}.input_layernorm.weight", prefix), device, [config.hidden_size])?;
-        layer.post_attention_layernorm.weight = load_tensor_1d(&tensors_map, &format!("{}.post_attention_layernorm.weight", prefix), device, [config.hidden_size])?;
-        
-        // Attention
-        let num_heads = config.num_attention_heads;
-        let num_kv_heads = config.num_key_value_heads;
-        layer.self_attn.q_proj = load_tensor_2d(&tensors_map, &format!("{}.self_attn.q_proj.weight", prefix), device, [num_heads * head_dim, config.hidden_size])?;
-        layer.self_attn.k_proj = load_tensor_2d(&tensors_map, &format!("{}.self_attn.k_proj.weight", prefix), device, [num_kv_heads * head_dim, config.hidden_size])?;
-        layer.self_attn.v_proj = load_tensor_2d(&tensors_map, &format!("{}.self_attn.v_proj.weight", prefix), device, [num_kv_heads * head_dim, config.hidden_size])?;
-        layer.self_attn.o_proj = load_tensor_2d(&tensors_map, &format!("{}.self_attn.o_proj.weight", prefix), device, [config.hidden_size, num_heads * head_dim])?;
-        
-        // MLP
-        layer.mlp.gate_proj = load_tensor_2d(&tensors_map, &format!("{}.mlp.gate_proj.weight", prefix), device, [config.intermediate_size, config.hidden_size])?;
-        layer.mlp.up_proj = load_tensor_2d(&tensors_map, &format!("{}.mlp.up_proj.weight", prefix), device, [config.intermediate_size, config.hidden_size])?;
-        layer.mlp.down_proj = load_tensor_2d(&tensors_map, &format!("{}.mlp.down_proj.weight", prefix), device, [config.hidden_size, config.intermediate_size])?;
-    }
-    
-    Ok(model)
-}

network-poc/node/src/burn_smollm/mod.rs

@@ -1,6 +0,0 @@
-pub mod attention;
-pub mod config;
-pub mod loader;
-pub mod model;
-pub mod modules;
-pub mod rope;

network-poc/node/src/burn_smollm/model.rs

@@ -1,96 +0,0 @@
-use burn::module::{Module, Param};
-use burn::tensor::{backend::Backend, Tensor, Int};
-use super::modules::{RmsNorm, Mlp};
-use super::attention::{Attention, KVCache};
-use super::config::SmolLMConfig;
-
-#[derive(Module, Debug)]
-pub struct LlamaBlock<B: Backend> {
-    pub self_attn: Attention<B>,
-    pub mlp: Mlp<B>,
-    pub input_layernorm: RmsNorm<B>,
-    pub post_attention_layernorm: RmsNorm<B>,
-}
-
-impl<B: Backend> LlamaBlock<B> {
-    pub fn new(config: &SmolLMConfig, device: &B::Device) -> Self {
-        Self {
-            self_attn: Attention::new(config, device),
-            mlp: Mlp::new(config.hidden_size, config.intermediate_size, device),
-            input_layernorm: RmsNorm::new(config.hidden_size, config.rms_norm_eps, device),
-            post_attention_layernorm: RmsNorm::new(config.hidden_size, config.rms_norm_eps, device),
-        }
-    }
-
-    pub fn forward(
-        &self,
-        x: Tensor<B, 3>,
-        offset: usize,
-        cache: Option<KVCache<B>>
-    ) -> (Tensor<B, 3>, KVCache<B>) {
-        let residual = x.clone();
-        let x_norm = self.input_layernorm.forward(x);
-        
-        let (attn_out, new_cache) = self.self_attn.forward(x_norm, offset, cache);
-        
-        let x = residual + attn_out;
-        
-        let residual = x.clone();
-        let x_norm = self.post_attention_layernorm.forward(x);
-        let mlp_out = self.mlp.forward(x_norm);
-        
-        let x = residual + mlp_out;
-        (x, new_cache)
-    }
-}
-
-#[derive(Module, Debug)]
-pub struct LlamaModel<B: Backend> {
-    pub embed_tokens: Param<Tensor<B, 2>>,
-    pub layers: Vec<LlamaBlock<B>>,
-    pub norm: RmsNorm<B>,
-    pub lm_head: Param<Tensor<B, 2>>, // For tie_word_embeddings this can point to embed_tokens
-}
-
-impl<B: Backend> LlamaModel<B> {
-    pub fn new(config: &SmolLMConfig, device: &B::Device) -> Self {
-        let embed = Tensor::zeros([config.vocab_size, config.hidden_size], device);
-        let lm_head = Tensor::zeros([config.vocab_size, config.hidden_size], device);
-        
-        let mut layers = Vec::new();
-        for _ in 0..config.num_hidden_layers {
-            layers.push(LlamaBlock::new(config, device));
-        }
-
-        Self {
-            embed_tokens: Param::from_tensor(embed),
-            layers,
-            norm: RmsNorm::new(config.hidden_size, config.rms_norm_eps, device),
-            lm_head: Param::from_tensor(lm_head),
-        }
-    }
-
-    pub fn forward(
-        &self,
-        input_ids: Tensor<B, 2, Int>,
-        offset: usize,
-        caches: &mut Vec<Option<KVCache<B>>>
-    ) -> Tensor<B, 3> {
-        let [_batch, _seq_len] = input_ids.dims();
-        
-        let mut x = burn::tensor::module::embedding(self.embed_tokens.val(), input_ids);
-        
-        for (i, layer) in self.layers.iter().enumerate() {
-            let cache = caches[i].take();
-            let (out, new_cache) = layer.forward(x, offset, cache);
-            x = out;
-            caches[i] = Some(new_cache);
-        }
-        
-        x = self.norm.forward(x);
-        
-        // Matmul with lm_head (or embed_tokens if tied) to get logits
-        // Notice: lm_head is typically [vocab_size, hidden_size] in HF, so we swap dims
-        x.matmul(self.lm_head.val().swap_dims(0, 1).unsqueeze())
-    }
-}

network-poc/node/src/burn_smollm/modules.rs

@@ -1,59 +0,0 @@
-use burn::module::{Module, Param};
-use burn::tensor::{backend::Backend, Tensor};
-
-#[derive(Module, Debug)]
-pub struct RmsNorm<B: Backend> {
-    pub weight: Param<Tensor<B, 1>>,
-    epsilon: f64,
-}
-
-impl<B: Backend> RmsNorm<B> {
-    pub fn new(size: usize, epsilon: f64, device: &B::Device) -> Self {
-        let weight = Param::from_tensor(Tensor::ones([size], device));
-        Self { weight, epsilon }
-    }
-
-    pub fn forward(&self, x: Tensor<B, 3>) -> Tensor<B, 3> {
-        // x: [batch, seq_len, dim]
-        // RMSNorm: x * weight / sqrt(mean(x^2) + eps)
-        let x_sq = x.clone().powf_scalar(2.0);
-        // mean over last dim, keeping dims for broadcast
-        let [b, s, d] = x_sq.dims();
-        let variance = x_sq.sum_dim(2).div_scalar(d as f32);
-        let norm = x.div(variance.add_scalar(self.epsilon).sqrt());
-
-        let w = self.weight.val().unsqueeze::<2>().unsqueeze::<3>().reshape([1, 1, d]);
-        norm * w
-    }
-}
-
-#[derive(Module, Debug)]
-pub struct Mlp<B: Backend> {
-    pub gate_proj: Param<Tensor<B, 2>>, // [in, intermediate]
-    pub up_proj: Param<Tensor<B, 2>>,   // [in, intermediate]
-    pub down_proj: Param<Tensor<B, 2>>, // [intermediate, out]
-}
-
-impl<B: Backend> Mlp<B> {
-    pub fn new(hidden_size: usize, intermediate_size: usize, device: &B::Device) -> Self {
-        Self {
-            gate_proj: Param::from_tensor(Tensor::zeros([hidden_size, intermediate_size], device)),
-            up_proj: Param::from_tensor(Tensor::zeros([hidden_size, intermediate_size], device)),
-            down_proj: Param::from_tensor(Tensor::zeros([intermediate_size, hidden_size], device)),
-        }
-    }
-
-    pub fn forward(&self, x: Tensor<B, 3>) -> Tensor<B, 3> {
-        // x: [batch, seq, hidden]
-        // gate = x @ gate_proj  -> [batch, seq, intermediate]
-        let gate = x.clone().matmul(self.gate_proj.val().unsqueeze());
-        let up = x.matmul(self.up_proj.val().unsqueeze());
-
-        // SiLU(gate) * up
-        let silu = gate.clone() * burn::tensor::activation::sigmoid(gate);
-        let intermediate = silu * up;
-
-        // intermediate @ down_proj -> [batch, seq, hidden]
-        intermediate.matmul(self.down_proj.val().unsqueeze())
-    }
-}

network-poc/node/src/burn_smollm/rope.rs

@@ -1,59 +0,0 @@
-use burn::module::Module;
-use burn::tensor::{backend::Backend, Tensor};
-
-#[derive(Module, Debug)]
-pub struct RoPE<B: Backend> {
-    cos_cache: Tensor<B, 2>,
-    sin_cache: Tensor<B, 2>,
-}
-
-impl<B: Backend> RoPE<B> {
-    pub fn new(head_dim: usize, max_seq_len: usize, theta: f32, device: &B::Device) -> Self {
-        // (head_dim / 2) values
-        let half_dim = head_dim / 2;
-        let inv_freq: Vec<f32> = (0..half_dim)
-            .map(|i| 1.0 / theta.powf((2 * i) as f32 / head_dim as f32))
-            .collect();
-        
-        let inv_freq = Tensor::<B, 1>::from_floats(inv_freq.as_slice(), device).unsqueeze::<2>();
-        let t_floats: Vec<f32> = (0..max_seq_len).map(|v| v as f32).collect();
-        let t = Tensor::<B, 1>::from_floats(t_floats.as_slice(), device).unsqueeze::<2>().transpose();
-        // t shape: [max_seq_len, 1]
-        // inv_freq shape: [1, half_dim]
-        
-        // freqs shape: [max_seq_len, half_dim]
-        let freqs = t.matmul(inv_freq);
-        
-        let cos_cache = freqs.clone().cos();
-        let sin_cache = freqs.sin();
-
-        Self {
-            cos_cache,
-            sin_cache,
-        }
-    }
-
-    pub fn forward(&self, x: Tensor<B, 4>, offset: usize) -> Tensor<B, 4> {
-        let [batch, heads, seq_len, head_dim] = x.dims();
-        let half_dim = head_dim / 2;
-        
-        // x shape: [batch, heads, seq_len, head_dim]
-        // valitaan viipaleet (x1 ja x2) jotta saadaan pyöritettyä rotaatiot
-        let x1 = x.clone().slice([0..batch, 0..heads, 0..seq_len, 0..half_dim]);
-        let x2 = x.clone().slice([0..batch, 0..heads, 0..seq_len, half_dim..head_dim]);
-        
-        // haetaan vastaava seq offsetista alkaen
-        let cos = self.cos_cache.clone().slice([offset..offset+seq_len, 0..half_dim])
-            .unsqueeze::<4>() // [seq, half_dim, 1]
-            .reshape([1, 1, seq_len, half_dim]);
-        let sin = self.sin_cache.clone().slice([offset..offset+seq_len, 0..half_dim])
-            .reshape([1, 1, seq_len, half_dim]);
-
-        // x1 * cos - x2 * sin
-        let o1 = x1.clone().mul(cos.clone()) - x2.clone().mul(sin.clone());
-        // x2 * cos + x1 * sin
-        let o2 = x2.mul(cos) + x1.mul(sin);
-
-        Tensor::cat(vec![o1, o2], 3)
-    }
-}

network-poc/node/src/lib.rs

@@ -3,16 +3,11 @@ use web_sys::{WebSocket, MessageEvent};
 use std::cell::RefCell;
 use std::rc::Rc;
 use std::sync::atomic::{AtomicU32, AtomicBool, Ordering};
-use burn::tensor::Tensor;
-use burn::backend::{Wgpu, NdArray};
 
 pub mod storage;
 pub mod sampling;
-pub mod smollm;
 pub mod qwen;
 pub mod qwen_coder;
-pub mod phi3;
-pub mod burn_smollm;
 
 #[macro_export]
 macro_rules! console_log {
@@ -82,41 +77,6 @@ pub async fn worker_fetch(url: &str) -> Result<web_sys::Response, String> {
         .map_err(|_| "ei Response".to_string())
 }
 
-// Geneerinen tensorilaskenta — toimii millä tahansa Burn-backendillä
-fn run_matmul<B: burn::tensor::backend::Backend>(size: usize) -> String {
-    let device = Default::default();
-    let dist = burn::tensor::Distribution::Default;
-    let t1: Tensor<B, 2> = Tensor::random([size, size], dist, &device);
-    let t2: Tensor<B, 2> = Tensor::random([size, size], dist, &device);
-    let sum = t1.matmul(t2).sum();
-    format!("{:?}", sum)
-}
-
-// Päättelyfunktio — valitsee backendin automaattisesti
-async fn run_ai_tensor_inference(difficulty: usize) -> String {
-    let load_pct = GPU_LOAD_PERCENT.load(Ordering::SeqCst);
-
-    if load_pct == 0 {
-        sleep_ms(2000).await;
-        return format!("Paused (0%). Lepäillään zZz..");
-    }
-
-    let active_workload_size = (difficulty as f32 * (load_pct as f32 / 100.0)) as usize;
-
-    let sleep_delay = (100 - load_pct) * 10;
-    if sleep_delay > 0 {
-        sleep_ms(sleep_delay as i32).await;
-    }
-
-    let use_gpu = HAS_WEBGPU.load(Ordering::SeqCst);
-    let (backend_name, result) = if use_gpu {
-        ("WebGPU", run_matmul::<Wgpu>(active_workload_size))
-    } else {
-        ("CPU/NdArray", run_matmul::<NdArray>(active_workload_size))
-    };
-
-    format!("PoC {} Matmul ({}x{}) >> {}", backend_name, active_workload_size, active_workload_size, result)
-}
 
 /// JS-exportti: tokenisoi tekstin ja palauttaa JSON-merkkijonon
 /// Tokenizer ladataan IndexedDB:stä (täytyy olla ladattu aiemmin)
@@ -246,7 +206,7 @@ pub async fn start_agent_node(hub_url: String, has_webgpu: bool, device_info_jso
     HAS_WEBGPU.store(has_webgpu, Ordering::SeqCst);
     SELECTED_TASK.store(task_id, Ordering::SeqCst);
     let backend_name = if has_webgpu { "WebGPU" } else { "CPU (NdArray)" };
-    let task_names = ["tokenize", "smollm-135m", "qwen-05b", "phi3-mini", "qwen-coder-05b", "qwen-coder-3b"];
+    let task_names = ["tokenize", "qwen-05b", "qwen-coder-05b", "qwen-coder-3b"];
     let task_name = task_names.get(task_id as usize).unwrap_or(&"tokenize");
     console_log!("Kipinä Agent Node käynnistyy — backend: {} | tehtävä: {}", backend_name, task_name);
 
@@ -303,22 +263,6 @@ pub async fn start_agent_node(hub_url: String, has_webgpu: bool, device_info_jso
                     }
                 }
             } else if msg.contains("llm_prompt") && current_task == 1 && auto_on {
-                // Vain SmolLM-solmut, ja vain yksi inferenssi kerrallaan
-                if LLM_BUSY.load(Ordering::SeqCst) {
-                    // Ohitetaan — edellinen inferenssi vielä käynnissä
-                } else if let Ok(task) = serde_json::from_str::<serde_json::Value>(&msg) {
-                    let prompt = task.get("prompt").and_then(|v| v.as_str()).unwrap_or("").to_string();
-                    let model = task.get("model").and_then(|v| v.as_str()).unwrap_or("").to_string();
-                    if !prompt.is_empty() && model == "smollm-135m" {
-                        LLM_BUSY.store(true, Ordering::SeqCst);
-                        let ws_for_async = ws_clone.clone();
-                        wasm_bindgen_futures::spawn_local(async move {
-                            smollm::run_smollm_inference(prompt, ws_for_async).await;
-                            LLM_BUSY.store(false, Ordering::SeqCst);
-                        });
-                    }
-                }
-            } else if msg.contains("llm_prompt") && current_task == 2 && auto_on {
                 // Qwen2.5-0.5B
                 if LLM_BUSY.load(Ordering::SeqCst) {
                 } else if let Ok(task) = serde_json::from_str::<serde_json::Value>(&msg) {
@@ -333,21 +277,6 @@ pub async fn start_agent_node(hub_url: String, has_webgpu: bool, device_info_jso
                         });
                     }
                 }
-            } else if msg.contains("llm_prompt") && current_task == 3 && auto_on {
-                // Phi-3 Mini
-                if LLM_BUSY.load(Ordering::SeqCst) {
-                } else if let Ok(task) = serde_json::from_str::<serde_json::Value>(&msg) {
-                    let prompt = task.get("prompt").and_then(|v| v.as_str()).unwrap_or("").to_string();
-                    let model = task.get("model").and_then(|v| v.as_str()).unwrap_or("").to_string();
-                    if !prompt.is_empty() && model.starts_with("phi3-mini") {
-                        LLM_BUSY.store(true, Ordering::SeqCst);
-                        let ws_for_async = ws_clone.clone();
-                        wasm_bindgen_futures::spawn_local(async move {
-                            phi3::run_phi3_inference(prompt, ws_for_async).await;
-                            LLM_BUSY.store(false, Ordering::SeqCst);
-                        });
-                    }
-                }
             } else if msg.contains("llm_prompt") {
                 console_log!("[DEBUG] llm_prompt vastaanotettu! current_task={}, busy={}", current_task, LLM_BUSY.load(Ordering::SeqCst));
                 if current_task == 4 || current_task == 5 {
@@ -368,28 +297,23 @@ pub async fn start_agent_node(hub_url: String, has_webgpu: bool, device_info_jso
                                 let _ = ws_clone.borrow().send_with_str(&err_msg.to_string());
                             }
                         } else {
+                            // Välitetään parametrit JSON-promptina coderille
+                            let coder_prompt = serde_json::json!({
+                                "prompt": prompt,
+                                "system": task.get("system_prompt").and_then(|v| v.as_str()).unwrap_or(""),
+                                "max_tokens": task.get("max_tokens").and_then(|v| v.as_u64()).unwrap_or(512),
+                            }).to_string();
                             let use_3b = current_task == 5;
                             LLM_BUSY.store(true, Ordering::SeqCst);
                             let ws_for_async = ws_clone.clone();
                             wasm_bindgen_futures::spawn_local(async move {
-                                qwen_coder::run_coder_inference(prompt, ws_for_async, use_3b, task_id).await;
+                                qwen_coder::run_coder_inference(coder_prompt, ws_for_async, use_3b, task_id).await;
                                 LLM_BUSY.store(false, Ordering::SeqCst);
                             });
                         }
                     }
                 }
                 } // current_task == 4 || 5
-            } else if msg.contains("ai_task") {
-                console_log!("Hub task vastaanotettu, ajetaan GPU:lla...");
-                let ws_for_async = ws_clone.clone();
-                let diff = if msg.contains(r#""difficulty":1024"#) { 1024 } else { 512 };
-                
-                // Suoritetaan inference asynkronisesti erillisessä taaskissa välttääksemme UI-jäätymisen kokonaan
-                wasm_bindgen_futures::spawn_local(async move {
-                    let result = run_ai_tensor_inference(diff).await;
-                    let reply = format!("{{\"type\":\"result\", \"status\":\"success\", \"data\":\"{}\"}}", result);
-                    let _ = ws_for_async.borrow().send_with_str(&reply);
-                });
             } else if msg.contains("stats") {
                 // Sivuutetaan statsit täällä, UI hallitsee ne aivan itse HTML:n puolella
             }