Getting started
Quickstart
brigade on its own is shorthand for brigade tui — the chat TUI, which auto-starts a gateway if one is not already running. Here is everything you need to be productive on day one.
Set up and go#
Three commands take you from nothing to chatting with your crew:
# 1. Set up — storage, a provider + API key (or local Ollama), a model, web search$ brigade onboard # 2. Start the always-on gateway — your crew, channels, and cron jobs live here$ brigade gateway run # 3. Chat with your crew$ brigade tuibrigade tui (or just brigade) auto-starts the gateway if it is not already running, so for a quick local session you can skip step 2. To keep the gateway alive across reboots, run brigade gateway install — see The gateway.
Other entry points#
$ brigade agent -m "summarize ~/today.md" # one-shot turn, no TUI$ brigade connect # attach a client to a running gateway$ brigade status # config + sessions + gateway snapshot$ brigade doctor # health-check your installRun brigade --help or brigade <command> --help for full flags. The full list is on the CLI reference page.
Running from a source checkout?
npm run onboard, npm run tui, and npm run gateway (plus npm run connect, npm run status, npm run doctor). See Installation.In-chat commands#
When you are in the chat TUI (brigade or brigade connect), these slash commands work inline:
Slash commands are kept deliberately small — they are the operator-level actions the model cannot do for itself (it cannot switch its own model or reset its own session), so they fire locally and deterministically without burning a model call.
| Command | What it does |
|---|---|
/model <provider/modelId> | Switch the model for the next turn, e.g. /model anthropic/claude-opus-4-7 |
/model | Print the active provider/model for this session |
/thinking <off|low|medium|high> | Set the reasoning effort for the next turn |
/reset | Forget the running session and start fresh on the next turn |
/help | List the available commands |
An unknown /command is passed through to the model unchanged, so it never errors on a command you have defined elsewhere.
Switching models keeps your context
/model carries the full transcript onto the new model and re-anchors your thinking level to what it supports. See Models & web search for how cross-model continuity (Carrow) works.From a channel#
When you are talking to your crew over a channel (WhatsApp or Telegram), a separate, channel-only command set lets you steer which agent answers:
| Command | What it does |
|---|---|
/agent <id> | Switch which agent this chat is routed to |
/agents | List the agents available on this channel |
/whoami | Show the agent currently answering this chat |
/org | Show the org chart (supports sub-forms like departments / explain) |
A first real task#
Try delegating. Ask your main agent to do something multi-step and it can spawn sub-agents, schedule a follow-up with cron, or remember a fact with memory — all from inside the conversation. From there, add a second agent (Agents & isolation) or connect a channel so you can reach the crew from WhatsApp.