Checkpoint: Phase 10: LLM Provider Configuration — Ollama Cloud как дефолт, локальный Ollama закомментирован (GPU only), docker-compose/stack обновлены, .env.example с 4 провайдерами

2026-03-20 19:16:44 -04:00
parent 02742f836c
commit dda99d7056
6 changed files with 259 additions and 71 deletions
--- a/docker/.env.example
+++ b/docker/.env.example
@@ -0,0 +1,53 @@
+##############################################################################
+# GoClaw Control Center — Environment Variables
+# Copy this file to docker/.env and fill in your values.
+#
+#   cp docker/.env.example docker/.env
+#   docker compose -f docker/docker-compose.yml up -d
+##############################################################################
+
+# ── LLM Provider ─────────────────────────────────────────────────────────────
+#
+# Choose ONE of the following providers by setting LLM_BASE_URL + LLM_API_KEY.
+#
+# Option 1: Ollama Cloud (default)
+#   Sign up at https://ollama.com  → Settings → API Keys
+#   Supports all public Ollama models (qwen2.5, llama3, mistral, etc.)
+LLM_BASE_URL=https://ollama.com/v1
+LLM_API_KEY=your-ollama-cloud-api-key-here
+
+# Option 2: OpenAI
+# LLM_BASE_URL=https://api.openai.com/v1
+# LLM_API_KEY=sk-...
+
+# Option 3: Any OpenAI-compatible API (Groq, Together, Mistral, etc.)
+# LLM_BASE_URL=https://api.groq.com/openai/v1
+# LLM_API_KEY=gsk_...
+
+# Option 4: Local Ollama on a GPU machine (no API key needed)
+#   Also uncomment the "ollama" service in docker-compose.yml if running locally.
+# LLM_BASE_URL=http://localhost:11434
+# LLM_API_KEY=
+
+# Default model to use when agent config is not available
+DEFAULT_MODEL=qwen2.5:7b
+
+# ── Database ──────────────────────────────────────────────────────────────────
+MYSQL_ROOT_PASSWORD=goClawRoot123
+MYSQL_DATABASE=goclaw
+MYSQL_USER=goclaw
+MYSQL_PASSWORD=goClawPass123
+
+# ── Authentication ────────────────────────────────────────────────────────────
+JWT_SECRET=change-me-to-a-random-64-char-string
+
+# ── Manus OAuth (optional, for Manus-hosted deployment) ──────────────────────
+VITE_APP_ID=
+OAUTH_SERVER_URL=
+VITE_OAUTH_PORTAL_URL=
+
+# ── Manus Built-in APIs (optional) ───────────────────────────────────────────
+BUILT_IN_FORGE_API_URL=
+BUILT_IN_FORGE_API_KEY=
+VITE_FRONTEND_FORGE_API_KEY=
+VITE_FRONTEND_FORGE_API_URL=
--- a/docker/Dockerfile.control-center
+++ b/docker/Dockerfile.control-center
@@ -6,8 +6,9 @@ RUN npm install -g pnpm@latest

 WORKDIR /app

-# Copy package files for layer caching
+# Copy package files and patches for layer caching
 COPY package.json pnpm-lock.yaml ./
+COPY patches/ ./patches/
 RUN pnpm install --frozen-lockfile

 # Copy source code (exclude gateway/ and docker/)
@@ -40,7 +41,9 @@ RUN npm install -g pnpm@latest

 # Copy package files and install production deps only
 COPY package.json pnpm-lock.yaml ./
-RUN pnpm install --frozen-lockfile --prod
+COPY patches/ ./patches/
+# Install all deps (vite is needed at runtime for SSR/proxy)
+RUN pnpm install --frozen-lockfile --ignore-scripts

 # Copy built artifacts
 COPY --from=builder /app/dist ./dist
--- a/docker/docker-compose.yml
+++ b/docker/docker-compose.yml
@@ -4,23 +4,34 @@
 # Services:
 #   control-center  — React + Node.js tRPC frontend/backend (:3000)
 #   gateway         — Go Orchestrator + Tool Executor (:18789)
-#   ollama          — Local LLM server (:11434)
 #   db              — MySQL 8 (:3306)
 #
+# LLM Provider (set in .env or environment):
+#   Cloud (default): LLM_BASE_URL=https://ollama.com/v1  + LLM_API_KEY=<key>
+#   OpenAI-compat:   LLM_BASE_URL=https://api.openai.com/v1 + LLM_API_KEY=<key>
+#   Local GPU node:  LLM_BASE_URL=http://<gpu-host>:11434   (no key needed)
+#
+# Local Ollama (GPU only):
+#   The ollama service below is commented out by default.
+#   Uncomment it only on machines with a compatible GPU.
+#   Then set: LLM_BASE_URL=http://ollama:11434
+#
 # Usage:
+#   cp docker/.env.example docker/.env   # fill in LLM_API_KEY etc.
 #   docker compose -f docker/docker-compose.yml up -d
 #   docker compose -f docker/docker-compose.yml logs -f gateway
+#   docker compose -f docker/docker-compose.yml down -v
 ##############################################################################

-version: "3.9"
+name: goclaw

 networks:
  goclaw-net:
    driver: bridge

 volumes:
-  ollama-data:
  mysql-data:
+  # ollama-data:   # Uncomment when using local Ollama service below

 services:

@@ -47,27 +58,32 @@ services:
      retries: 5
      start_period: 30s

-  # ── Ollama LLM Server ─────────────────────────────────────────────────────
-  ollama:
-    image: ollama/ollama:latest
-    container_name: goclaw-ollama
-    restart: unless-stopped
-    ports:
-      - "11434:11434"
-    volumes:
-      - ollama-data:/root/.ollama
-    networks:
-      - goclaw-net
-    deploy:
-      resources:
-        reservations:
-          devices:
-            - driver: nvidia
-              count: all
-              capabilities: [gpu]
-    # Uncomment if no GPU:
-    # environment:
-    #   - OLLAMA_NUM_PARALLEL=2
+  # ── Local Ollama LLM Server (GPU ONLY — disabled by default) ─────────────
+  # Uncomment this entire block only on machines with a compatible NVIDIA or
+  # Apple Silicon GPU.  Then set LLM_BASE_URL=http://ollama:11434 in the
+  # gateway service below (or in your .env file).
+  #
+  # ollama:
+  #   image: ollama/ollama:latest
+  #   container_name: goclaw-ollama
+  #   restart: unless-stopped
+  #   ports:
+  #     - "11434:11434"
+  #   volumes:
+  #     - ollama-data:/root/.ollama
+  #   networks:
+  #     - goclaw-net
+  #   environment:
+  #     - OLLAMA_NUM_PARALLEL=2
+  #     - OLLAMA_MAX_LOADED_MODELS=2
+  #   # NVIDIA GPU support — uncomment if available:
+  #   # deploy:
+  #   #   resources:
+  #   #     reservations:
+  #   #       devices:
+  #   #         - driver: nvidia
+  #   #           count: all
+  #   #           capabilities: [gpu]

  # ── Go Gateway (Orchestrator + Tool Executor) ─────────────────────────────
  gateway:
@@ -80,21 +96,34 @@ services:
      - "18789:18789"
    environment:
      PORT: "18789"
-      OLLAMA_BASE_URL: "http://ollama:11434"
+      # ── LLM Provider ─────────────────────────────────────────────────────
+      # Cloud default (Ollama Cloud, OpenAI-compatible):
+      LLM_BASE_URL: "${LLM_BASE_URL:-https://ollama.com/v1}"
+      LLM_API_KEY:  "${LLM_API_KEY:-${OLLAMA_API_KEY:-}}"
+      # Legacy alias (still supported):
+      OLLAMA_API_KEY: "${OLLAMA_API_KEY:-${LLM_API_KEY:-}}"
+      # ── To use local Ollama on GPU node, set in .env: ─────────────────────
+      # LLM_BASE_URL=http://ollama:11434   (if ollama service above is enabled)
+      # LLM_BASE_URL=http://<gpu-host-ip>:11434  (external GPU machine)
+      # ─────────────────────────────────────────────────────────────────────
+      DEFAULT_MODEL: "${DEFAULT_MODEL:-qwen2.5:7b}"
      DATABASE_URL: "${MYSQL_USER:-goclaw}:${MYSQL_PASSWORD:-goClawPass123}@tcp(db:3306)/${MYSQL_DATABASE:-goclaw}?parseTime=true"
      PROJECT_ROOT: "/app"
-      REQUEST_TIMEOUT_SECS: "120"
+      GATEWAY_REQUEST_TIMEOUT_SECS: "120"
+      GATEWAY_MAX_TOOL_ITERATIONS: "10"
      LOG_LEVEL: "info"
    depends_on:
      db:
        condition: service_healthy
-      ollama:
-        condition: service_started
+      # ollama:              # Uncomment if using local Ollama service above
+      #   condition: service_started
    networks:
      - goclaw-net
    volumes:
-      # Mount project root for file tools (read-only in prod, rw in dev)
+      # Mount project root for file tools (read-only)
      - ..:/app:ro
+      # Mount Docker socket for docker_exec tool
+      - /var/run/docker.sock:/var/run/docker.sock
    healthcheck:
      test: ["CMD", "wget", "-qO-", "http://localhost:18789/health"]
      interval: 15s
--- a/docker/docker-stack.yml
+++ b/docker/docker-stack.yml
@@ -1,7 +1,23 @@
 ##############################################################################
 # GoClaw Control Center — Docker Stack (Docker Swarm Production)
 #
+# LLM Provider:
+#   By default the gateway uses Ollama Cloud (https://ollama.com/v1).
+#   Set LLM_BASE_URL and LLM_API_KEY via Docker secrets or environment.
+#
+#   To use a local Ollama instance on a GPU-equipped Swarm node:
+#     1. Uncomment the "ollama" service below.
+#     2. Add the label  gpu=true  to the GPU node:
+#          docker node update --label-add gpu=true <node-id>
+#     3. Change LLM_BASE_URL in the gateway service to: http://ollama:11434
+#
 # Deploy:
+#   # Create required secrets first:
+#   echo "rootpass" | docker secret create mysql-root-password -
+#   echo "pass"     | docker secret create mysql-password -
+#   echo "jwtsecret"| docker secret create jwt-secret -
+#   echo "ollamakey"| docker secret create llm-api-key -
+#
 #   docker stack deploy -c docker/docker-stack.yml goclaw
 #
 # Remove:
@@ -19,14 +35,10 @@ networks:
    attachable: true

 volumes:
-  ollama-data:
-    driver: local
  mysql-data:
    driver: local
-
-configs:
-  gateway-env:
-    external: true
+  # ollama-data:       # Uncomment when using local Ollama service below
+  #   driver: local

 secrets:
  mysql-root-password:
@@ -35,6 +47,8 @@ secrets:
    external: true
  jwt-secret:
    external: true
+  llm-api-key:
+    external: true

 services:

@@ -73,45 +87,59 @@ services:
      retries: 5
      start_period: 30s

-  # ── Ollama LLM Server ─────────────────────────────────────────────────────
-  ollama:
-    image: ollama/ollama:latest
-    volumes:
-      - ollama-data:/root/.ollama
-    networks:
-      - goclaw-net
-    deploy:
-      replicas: 1
-      placement:
-        constraints:
-          # Pin to GPU node if available
-          - node.labels.gpu == true
-      restart_policy:
-        condition: on-failure
-        delay: 15s
-      resources:
-        limits:
-          memory: 16G
-        reservations:
-          memory: 4G
-    # GPU support via nvidia-container-runtime
-    # Uncomment on GPU-enabled nodes:
-    # runtime: nvidia
-    # environment:
-    #   - NVIDIA_VISIBLE_DEVICES=all
+  # ── Local Ollama LLM Server (GPU ONLY — disabled by default) ─────────────
+  # Uncomment this entire block only on Swarm nodes with a compatible GPU.
+  # After uncommenting, also change gateway LLM_BASE_URL to http://ollama:11434
+  # and remove the llm-api-key secret from gateway (not needed for local Ollama).
+  #
+  # ollama:
+  #   image: ollama/ollama:latest
+  #   volumes:
+  #     - ollama-data:/root/.ollama
+  #   networks:
+  #     - goclaw-net
+  #   deploy:
+  #     replicas: 1
+  #     placement:
+  #       constraints:
+  #         # Pin to GPU-labelled node:  docker node update --label-add gpu=true <id>
+  #         - node.labels.gpu == true
+  #     restart_policy:
+  #       condition: on-failure
+  #       delay: 15s
+  #     resources:
+  #       limits:
+  #         memory: 16G
+  #       reservations:
+  #         memory: 4G
+  #   # NVIDIA GPU support — uncomment on GPU-enabled nodes:
+  #   # runtime: nvidia
+  #   # environment:
+  #   #   - NVIDIA_VISIBLE_DEVICES=all

  # ── Go Gateway (Orchestrator + Tool Executor) ─────────────────────────────
  gateway:
    image: git.softuniq.eu/uniqai/goclaw/gateway:latest
    environment:
      PORT: "18789"
-      OLLAMA_BASE_URL: "http://ollama:11434"
+      # ── LLM Provider ─────────────────────────────────────────────────────
+      # Default: Ollama Cloud (requires llm-api-key secret below)
+      LLM_BASE_URL: "${LLM_BASE_URL:-https://ollama.com/v1}"
+      DEFAULT_MODEL: "${DEFAULT_MODEL:-qwen2.5:7b}"
+      # LLM_API_KEY is injected via /run/secrets/llm-api-key (see below)
+      # ── To switch to local GPU Ollama, set: ──────────────────────────────
+      # LLM_BASE_URL: "http://ollama:11434"
+      # (and uncomment the ollama service above)
+      # ─────────────────────────────────────────────────────────────────────
      DATABASE_URL: "goclaw:{{MYSQL_PASSWORD}}@tcp(db:3306)/goclaw?parseTime=true"
      PROJECT_ROOT: "/app"
-      REQUEST_TIMEOUT_SECS: "120"
+      GATEWAY_REQUEST_TIMEOUT_SECS: "120"
+      GATEWAY_MAX_TOOL_ITERATIONS: "10"
      LOG_LEVEL: "info"
    secrets:
      - mysql-password
+      - source: llm-api-key
+        target: /run/secrets/llm-api-key
    networks:
      - goclaw-net
    ports:
@@ -119,6 +147,9 @@ services:
        published: 18789
        protocol: tcp
        mode: ingress
+    volumes:
+      # Docker socket for docker_exec tool
+      - /var/run/docker.sock:/var/run/docker.sock
    deploy:
      replicas: 2
      update_config:
--- a/gateway/config/config.go
+++ b/gateway/config/config.go
@@ -4,15 +4,35 @@ import (
 	"log"
 	"os"
 	"strconv"
+	"strings"

 	"github.com/joho/godotenv"
 )

+// Config holds all runtime configuration for the GoClaw Gateway.
+//
+// LLM Provider selection (via LLM_BASE_URL env):
+//
+//	Cloud (default):  https://ollama.com/v1          — Ollama Cloud API (requires OLLAMA_API_KEY)
+//	OpenAI-compat:    https://api.openai.com/v1       — OpenAI or any OpenAI-compatible endpoint
+//	Local GPU node:   http://localhost:11434/v1        — Local Ollama (only on machines with GPU)
+//
+// To use a local Ollama instance on a GPU-equipped agent node, set:
+//
+//	LLM_BASE_URL=http://<gpu-node-ip>:11434/v1
+//	LLM_API_KEY=                                      # (empty for local Ollama)
 type Config struct {
 	// Server
 	Port string

-	// Ollama / LLM
+	// LLM / Ollama Cloud
+	// LLM_BASE_URL — OpenAI-compatible base URL (without trailing slash, /v1 suffix included).
+	// Default: https://ollama.com/v1 (Ollama Cloud).
+	// For local GPU agents set LLM_BASE_URL=http://<host>:11434/v1
+	LLMBaseURL string
+	LLMAPIKey  string
+
+	// Deprecated aliases kept for backward-compat (mapped to LLMBaseURL / LLMAPIKey)
 	OllamaBaseURL string
 	OllamaAPIKey  string

@@ -29,17 +49,34 @@ type Config struct {
 }

 func Load() *Config {
-	// Try to load .env from parent directory (project root)
+	// Try to load .env from parent directory (project root) or current dir
 	_ = godotenv.Load("../.env")
 	_ = godotenv.Load(".env")

 	maxIter, _ := strconv.Atoi(getEnv("GATEWAY_MAX_TOOL_ITERATIONS", "10"))
 	timeout, _ := strconv.Atoi(getEnv("GATEWAY_REQUEST_TIMEOUT_SECS", "120"))

+	// Resolve LLM base URL — priority: LLM_BASE_URL > OLLAMA_BASE_URL > default cloud
+	rawLLMURL := getEnvFirst(
+		"LLM_BASE_URL",      // preferred new name
+		"OLLAMA_BASE_URL",   // legacy alias
+	)
+	if rawLLMURL == "" {
+		// Default: Ollama Cloud (OpenAI-compatible, requires OLLAMA_API_KEY)
+		rawLLMURL = "https://ollama.com/v1"
+	}
+	// Normalise: strip trailing slash, ensure /v1 suffix for bare Ollama hosts
+	llmBaseURL := normaliseLLMURL(rawLLMURL)
+
+	// Resolve API key — priority: LLM_API_KEY > OLLAMA_API_KEY
+	llmAPIKey := getEnvFirst("LLM_API_KEY", "OLLAMA_API_KEY")
+
 	cfg := &Config{
 		Port:               getEnv("GATEWAY_PORT", "18789"),
-		OllamaBaseURL:      getEnv("OLLAMA_BASE_URL", "https://ollama.com/v1"),
-		OllamaAPIKey:       getEnv("OLLAMA_API_KEY", ""),
+		LLMBaseURL:         llmBaseURL,
+		LLMAPIKey:          llmAPIKey,
+		OllamaBaseURL:      llmBaseURL, // backward-compat alias
+		OllamaAPIKey:       llmAPIKey,  // backward-compat alias
 		DatabaseURL:        getEnv("DATABASE_URL", ""),
 		ProjectRoot:        getEnv("PROJECT_ROOT", "/home/ubuntu/goclaw-control-center"),
 		DefaultModel:       getEnv("DEFAULT_MODEL", "qwen2.5:7b"),
@@ -47,16 +84,43 @@ func Load() *Config {
 		RequestTimeoutSecs: timeout,
 	}

-	if cfg.OllamaAPIKey == "" {
-		log.Println("[Config] WARNING: OLLAMA_API_KEY is not set")
+	if cfg.LLMAPIKey == "" {
+		log.Println("[Config] WARNING: LLM_API_KEY / OLLAMA_API_KEY is not set — cloud API calls will fail")
 	}
 	if cfg.DatabaseURL == "" {
 		log.Println("[Config] WARNING: DATABASE_URL is not set — agent config will use defaults")
 	}

+	log.Printf("[Config] LLM endpoint: %s", cfg.LLMBaseURL)
+
 	return cfg
 }

+// normaliseLLMURL ensures the URL has a /v1 suffix for bare Ollama hosts.
+// Examples:
+//
+//	"http://localhost:11434"   → "http://localhost:11434/v1"
+//	"http://localhost:11434/"  → "http://localhost:11434/v1"
+//	"https://ollama.com/v1"   → "https://ollama.com/v1"   (unchanged)
+//	"https://api.openai.com/v1" → unchanged
+func normaliseLLMURL(raw string) string {
+	u := strings.TrimRight(raw, "/")
+	if !strings.HasSuffix(u, "/v1") {
+		u += "/v1"
+	}
+	return u
+}
+
+// getEnvFirst returns the value of the first non-empty env variable from the list.
+func getEnvFirst(keys ...string) string {
+	for _, k := range keys {
+		if v := os.Getenv(k); v != "" {
+			return v
+		}
+	}
+	return ""
+}
+
 func getEnv(key, fallback string) string {
 	if v := os.Getenv(key); v != "" {
 		return v
--- a/todo.md
+++ b/todo.md
@@ -143,3 +143,11 @@
 - [ ] Write Go unit tests (gateway/internal/tools/executor_test.go)
 - [ ] Write Go integration test for orchestrator chat loop
 - [ ] Push to Gitea (NW)
+
+## Phase 10: LLM Provider Configuration
+- [x] config.go: default LLM_BASE_URL = https://ollama.com/v1 (Ollama Cloud)
+- [x] config.go: support LLM_BASE_URL + LLM_API_KEY env vars (legacy OLLAMA_* aliases kept)
+- [x] config.go: normaliseLLMURL() — auto-append /v1 for bare Ollama hosts
+- [x] docker-compose.yml: ollama service commented out (GPU only), LLM_BASE_URL/LLM_API_KEY added
+- [x] docker-stack.yml: ollama service commented out (GPU only), llm-api-key secret added
+- [x] docker/.env.example: 4 LLM provider options documented (Ollama Cloud, OpenAI, Groq, Local GPU)