AI MCP Gateway

# AI MCP Gateway - Environment Variables # Copy this file to .env.docker and customize with your values # ============================================ # LLM Provider API Keys # ============================================ # REQUIRED: At least one provider API key # You have OpenRouter - this enables L1 layer OPENROUTER_API_KEY=sk-or-openrouter-key-here # Optional: Add these to enable more layers # L1, L2: Uncomment to use OpenAI directly (or use via OpenRouter) # OPENAI_API_KEY=sk-your-openai-key-here # L2, L3: Uncomment to use Anthropic Claude models # ANTHROPIC_API_KEY=sk-ant-your-anthropic-key-here # NOTE: With only OpenRouter key, you can still use all models via OpenRouter proxy # OpenRouter supports: OpenAI models, Anthropic models, Google models, etc. # Just use the OpenRouter model names in your requests # ============================================ # OpenRouter Configuration # ============================================ # Free models for L0 layer (fallback) OPENROUTER_FALLBACK_MODELS=meta-llama/llama-3.3-70b-instruct:free,x-ai/grok-4.1-fast:free # Use OpenRouter to proxy OpenAI/Claude models (if you don't have direct API keys) OPENROUTER_REPLACE_OPENAI=openai/gpt-4o-mini OPENROUTER_REPLACE_CLAUDE=anthropic/claude-3.5-sonnet # Layer override - customize models per layer using OpenRouter # Format: LAYER_L0_MODELS, LAYER_L1_MODELS, etc. LAYER_L0_MODELS=meta-llama/llama-3.3-70b-instruct:free,x-ai/grok-4.1-fast:free LAYER_L1_MODELS=google/gemini-flash-1.5,openai/gpt-4o-mini LAYER_L2_MODELS=anthropic/claude-3-haiku,openai/gpt-4o LAYER_L3_MODELS=anthropic/claude-3.5-sonnet,openai/o1-preview # ============================================ # Task-Specific Model Configuration # ============================================ # Models for different tasks (comma-separated) # Chat - General conversation CHAT_MODELS=meta-llama/llama-3.3-70b-instruct:free,google/gemini-flash-1.5 # Code - Code generation and analysis (prefer models with 'code' or 'coder' in name) CODE_MODELS=qwen/qwen-2.5-coder-32b-instruct:free,deepseek/deepseek-coder-33b-instruct:free # Analyze - Code review and analysis ANALYZE_MODELS=x-ai/grok-4.1-fast:free,anthropic/claude-3-haiku # Create Project - Full project scaffolding CREATE_PROJECT_MODELS=qwen/qwen-2.5-coder-32b-instruct:free,openai/gpt-4o-mini # ============================================ # OSS/Local Model (Ollama) # ============================================ # Set to true if using Ollama (requires: docker-compose --profile with-ollama up) OSS_MODEL_ENABLED=false OSS_MODEL_ENDPOINT=http://ollama:11434 OSS_MODEL_NAME=llama3:8b # ============================================ # Database (PostgreSQL) # ============================================ POSTGRES_DB=ai_mcp_gateway POSTGRES_USER=postgres POSTGRES_PASSWORD=your_secure_postgres_password_here # ============================================ # Redis Cache # ============================================ # Leave empty for no password (development only) REDIS_PASSWORD= # ============================================ # Application Configuration # ============================================ LOG_LEVEL=info DEFAULT_LAYER=L0 ENABLE_CROSS_CHECK=true ENABLE_AUTO_ESCALATE=false MAX_ESCALATION_LAYER=L0 ENABLE_COST_TRACKING=true COST_ALERT_THRESHOLD=1.00 # ============================================ # Task-Specific Model Selection # ============================================ # Code generation tasks - prefer models with "code" or "coder" in name CODE_MODELS=qwen/qwen-2.5-coder-32b-instruct:free,deepseek/deepseek-coder-33b-instruct:free # General chat/conversational tasks CHAT_MODELS=meta-llama/llama-3.3-70b-instruct:free,x-ai/grok-4.1-fast:free # Code analysis tasks - prefer models with good reasoning ANALYZE_MODELS=qwen/qwq-32b-preview:free,x-ai/grok-4.1-fast:free # Project creation tasks - prefer models with both code and reasoning capabilities CREATE_PROJECT_MODELS=qwen/qwen-2.5-coder-32b-instruct:free,deepseek/deepseek-coder-33b-instruct:free