Back to directory

Jazz

An agent harness that runs one user-defined agent, with its models, persona, tools and permissions set in a single JSON file, across many surfaces: a terminal REPL, a one-shot script command, a scheduled job, or a Telegram or Discord bot the user hosts themselves.

Jazz

Jazz separates the definition of an agent, its primary and companion models, persona, tools and permissions, from where it runs: the same configuration runs as an interactive terminal session, a one-shot command inside a script, a scheduled workflow, a GitHub Action that reviews pull requests, or a bot on a Telegram or Discord server the user controls. In a chat surface it asks for permission inline when a job needs it; from any other surface, it parks the job and sends a notification the user can approve from their phone.

It supports 23 model providers, including OpenAI, Anthropic, Google, Mistral and Groq with an API key or a ChatGPT subscription, plus self-hosted backends such as Ollama, llama.cpp, vLLM and SGLang with no API key required unless the server itself demands one; anything else connects through MCP. A single agent can bind different companion models to different media, keeping one model for reasoning and tools while routing image, audio or video understanding to models suited to each.

It installs as a single self-contained binary with no Node or npm dependency, or through npm, and is written in TypeScript with Effect.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.