hanyeol/model-compose

pip install model-compose
uv pip install model-compose

Usage

• •Create model-compose.yml defining controller, workflow and component blocks
• •export OPENAI_API_KEY=your-key && model-compose up
• •Switch an adapter from http-server to mcp-server to expose a workflow as an MCP server
• •Add a runtime block or Redis queue to deploy as a container or scale horizontally

Key features

• ✓One YAML file, any model (HuggingFace, vLLM, llama.cpp, OpenAI, Anthropic, Google, xAI)
• ✓Agents in YAML: ReAct loops, tool use, multi-step reasoning with human-in-the-loop approval
• ✓RAG pipelines with native drivers for Chroma, Milvus, Qdrant, FAISS, Neo4j, ArangoDB and Redis
• ✓MCP server generation with a one-line change; HTTP/WebSocket/MCP protocol adapters
• ✓Stream-native workflows, Gradio web UI, Docker runtime and Redis queue-based horizontal scaling

Best for

Deploying production-ready AI services, including MCP servers, without writing application code.

Caveats

• ⚠Requires model provider credentials (e.g. OPENAI_API_KEY) for hosted models
• ⚠Requires Python 3.10+