Agent platform · CLI, Web & API
Muaz
Run agents anywhere: terminal, browser, API.
Muaz is an agent platform in a single binary. Chat with AI agents in your terminal, compose them into multi-agent pipelines, and plug in any model, whether that's OpenAI-compatible, Gemini, Ollama or Bedrock. Mix frontier, heavy and local models per step to tune cost and performance. Need a web UI or an HTTP API? Run muaz gateway and everything the CLI does is available in the browser and over the wire.
muaz — zsh
# chat with an agent in your terminal
$muaz chat
▸ default agent ready · type to start
# or launch the web UI + API
$muaz gateway
▸ gateway live · UI + API ready
// features
What you get
Any model, one platform
OpenAI-compatible, Gemini, Ollama and Bedrock all work the same way. Switch providers with a line of config.
Multi-agent pipelines
Chain agents in sequence, fan them out in parallel, or route work to the right one automatically.
Terminal or browser
Live in the REPL, or run muaz gateway to get the same agents in a web UI and over an HTTP API.
Custom agents & skills
Define your own agents, reusable skills and slash commands, tailored to how your team actually works.
Tools & MCP
Built-in tools plus any MCP server, so your agents can read, write and reach out to the systems you use.
Safe by default
Per-agent tool-approval policies, saved sessions and incognito mode keep you in control of every action.
// how it works
Config, not code
Define agents, skills and multi-agent pipelines as plain YAML, then run them in the terminal or call them over the gateway API. The docs cover every option.
research-and-write.yaml
name: research-and-write
pattern: sequential
steps:
- agent: researcher # gather the facts
- agent: writer # turn them into a brief
pass: full_history
# run it from the REPL: /run research-and-writeBuilding with Muaz?
We can help you design, ship and scale it. Tell us what you're working on.