Overview

muaz is a local, web-first LLM agent platform in a single binary. Bare muaz opens a browser UI where you chat with agents, run multi-agent pipelines, and install plugins; muaz chat is the terminal REPL for when you prefer the command line. Connect any model — OpenAI-compatible APIs, Google Gemini, local Ollama, or AWS Bedrock — and everything runs on your own machine.

Beta

muaz is in beta. Everything documented here works and is tested, but it is pre-1.0: configuration formats, CLI flags and APIs can still change between releases, and a release may drop a deprecated shape rather than carry it forward. Read the release notes before upgrading, and pin a version if you are building something on top of it.

What that does not mean: your data is local by default and stays where it is — see Security.

What you can do

• Chat with agents — in the browser or in an interactive REPL session. Switch models or agents mid-session without losing context.
• Stay model-agnostic — agents bind to a model through named provider profiles, so you can point everything at a different model in one place.
• Install plugins — add bundles of agents, skills, and pipelines from a .zip or the registry, and manage them with muaz plugins.
• Run pipelines — chain agents sequentially, fan out work in parallel, or route tasks to the right agent automatically.
• Extend with skills — give agents reusable instruction sets (skills) that they load on demand, keeping prompts focused and composable.
• Control tool access — decide exactly which built-in tools and MCP servers each agent can call, with per-pattern approval policies.
• Use the browser UI — bare muaz (or muaz gateway for a headless box) serves a web interface for sessions, agents, providers, and plugins.

Where to go next

• Installation — install the pre-built binary on macOS, Linux, or Windows.
• Quick start — your first chat session in under a minute.
• Concepts — agents, providers, plugins, tools, skills, and pipelines explained.
• CLI reference — every command and flag.