Architecture Pattern MCP

Host a copy with our keys and logs.

• GENERATOR_API_KEY — API key for the generator LLM (LiteLLM provider).
• GENERATOR_PROVIDER — LiteLLM provider prefix, e.g. openai, anthropic, openrouter, minimax.
• GENERATOR_MODEL — Model name, combined as <provider>/<model> via LiteLLM.
• GENERATOR_BASE_URL — OpenAI-compatible base URL (LiteLLM api_base).
• PATTERN_DIRECTORY — Directory with *-pattern.json files. The image bakes the 40-pattern catalogue at /app/pattern.
• EMBEDDER_PROVIDER — Embedding backend for pattern retrieval: tei, openai, ollama, vllm or hosted_vllm.
• EMBEDDER_BASE_URL — Required: the dense retrieval leg embeds the 40-pattern catalogue during startup warmup and the server refuses to start when the endpoint is unreachable. OpenAI-compatible embeddings endpoint ending in /v1: on the Docker host use http://host.docker.internal:8080/v1 (add --add-host=host.docker.internal:host-gateway on Linux); on a shared Docker network use http://pattern-tei-embed:8080/v1.
• EMBEDDER_API_KEY — API key for the embedder. Keyless TEI accepts any non-empty value; use 'not-needed' there.
• RERANKER_BASE_URL — Required: pattern retrieval runs a mandatory TEI cross-encoder rerank. On the Docker host use http://host.docker.internal:8080; on a shared Docker network use http://pattern-tei-rerank:8080.