Skip to content
catalystcommunityPublic

About

Open Source Orchestration/OS for agents and AI, not just LLMs either.

Resources

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Magi

Magi is a composable automation and multi-agent system for personal, household, and business work. It is licensed under the Apache License, Version 2.0.

Magi is a set of separate primitives with their own APIs. The magi/ composition layer uses them to give people one conversation for questions, knowledge, and work. Clients (command line, web, Android) use the API.

  • Setup: set up a new installation, step by step.
  • Plan: goal, primitives, and slices.
  • Conventions: layout, rules, transport, storage, tests.
  • Conversations: the conversation graph.
  • Prompt injection: research and defense plan.
  • Speech interaction: research and plan.
  • Egress proxy: design of the outbound proxy with WASM plugins.
  • Host setup: requirements for the tool that makes model and speech hosts.
  • Each primitive has a README in primitives/<name>/.

Health endpoints

Every service answers GET /healthz with ok and no token. A service that depends on storage or engines answers 503 until that dependency answers, so the endpoint doubles as the Kubernetes readiness probe. The paths below are relative to the service's listen address (MAGI_LISTEN, default :8080).

Service Endpoint Checks
magi GET /healthz The API and web server process.
artifacts GET /healthz The process and the PostgreSQL database.
conversation GET /healthz The process and the PostgreSQL database.
knowledge GET /healthz The process and the PostgreSQL database.
workflow GET /healthz The process and the PostgreSQL database.
execution GET /healthz The process and the PostgreSQL database.
identity GET /healthz The process and the PostgreSQL database.
source GET /healthz The process and the repository directory.
speech GET /healthz The process and both speech engines.
release GET /healthz The controller process.
inference GET /healthz The host process.
envd (per environment) GET /healthz The environment daemon. No token.
Speech engine (STT/TTS) GET /health The engine process, device, and load.
Inference engine GET /health The model engine behind the host.

The apprentice worker claims workflow tasks and serves no HTTP endpoint.

Try it

To install Magi, read docs/setup.md. To use the command-line client, read magi/README.md.

About

Open Source Orchestration/OS for agents and AI, not just LLMs either.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages