Quick start
-
Set a provider API key.
Terminal window export OPENAI_API_KEY=sk-...Or skip the export and store the key with muaz — run
muaz auth(or/authinside a chat session), or use the browser UI’s Providers page. It’s saved to your OS keychain and used automatically. See API keys & authentication.muaz supports OpenAI-compatible endpoints, Google Gemini, AWS Bedrock, and local Ollama. Agents are model-agnostic — define a provider profile once and set it as the default to use it everywhere. See Providers & models.
-
Open the browser UI.
Terminal window muaz # bare command → starts the gateway + opens the UIBare
muazis web-first: it starts a local gateway on127.0.0.1:7878and opens your browser (over SSH / no tty it prints the URL instead). A first-run wizard walks you through picking a provider, pasting a key, and dropping into your first chat. From there you get agents, pipelines, connectors, skills, and settings — all in one place. See Browser UI. -
Prefer the terminal? Use the REPL.
Terminal window muaz chat # chat with the default agent in the terminalmuaz chat --agent coding # open the built-in coding agentmuaz chat --incognito # don't save the session to diskmuaz chat --session <id> # resume a previous session by IDThe REPL is the power-user surface. Type
/helpfor every slash command:Command Description /helpShow all commands /agentsList available agents /switch <name>Load a different agent /toolsList tools and their approval policy /sessionsBrowse and resume recent sessions /usageToken and latency stats for this session /exitQuit muaz
Run a one-off turn (scripting)
Section titled “Run a one-off turn (scripting)”muaz run sends a single turn to an agent and prints the result — no REPL,
no interactive session. It reads from stdin when piped, making it easy to
wire into shell scripts or editor workflows.
# Ask a question directlymuaz run "what is the difference between a mutex and a semaphore?"
# Pipe context in alongside the promptgit diff | muaz run "review this change and flag any bugs"
# Summarise a filemuaz run "summarise this" < README.md
# Get structured JSON output (includes usage stats and tool calls)muaz run "list the files here" --output jsonServer mode
Section titled “Server mode”muaz gateway runs the same web server without auto-opening a browser — for
headless boxes, or a small trusted team sharing one instance (see the
Docker team-server image).
muaz gateway # server mode: prints the URL, no browser openmuaz gateway --readonly # auditors/demos: browse, but no state changesSee Browser UI and Security.
What next?
Section titled “What next?”- Guides — four end-to-end walkthroughs of the flagship workflows (coding agent, pipelines, your-data/RAG, connector automation).
- Connectors — add GitHub, filesystem, fetch, memory, and Linear in a couple of clicks.
Where to go next
Section titled “Where to go next”- Agents — configure prompts, tools, and models.
- Providers & models — connect OpenAI, Gemini, Ollama, or Bedrock.
- Plugins — install bundles of agents, skills, and pipelines.
- Pipelines — chain agents into multi-step workflows.
- Command reference — every command and flag.