Quickstart
This walks you from a fresh install to a running agent. It assumes you've already installed the app and have at least one agent CLI (such as Claude Code) on your machine.
Start a new chat
Click New chat in the sidebar. The composer opens with your default agent and model selected, the ones you chose during setup.
Pick an agent and model (optional)
Use the Agent & model control to switch which agent runs and which model it uses. Selecting a model under Intelligence pins it; clicking the checked agent under Model clears the preference so the agent falls back to its own default.
Attach a project (optional)
Choose a folder to work in. With no project attached, the agent runs in a neutral scratch workspace. Nothing touches your repos until you attach one.
Send your first task
Type what you want done and send. Agent Relay launches the agent as a managed run and streams every turn into the chat as it happens.
Watching a run
Each chat shows the agent's turns live (messages, tool calls, and file changes) as the run progresses. You stay in control:
- Command approval decides how much the agent may do on its own. Set it to Ask (confirm each command), Approve (batch confirm), or Full auto in Settings → Launch defaults or per chat.
- On rate limit, the app can automatically switch to another agent to keep the conversation going, or stop and wait, your choice.
Slash commands
Type / in a chat to bring up commands:
| Command | What it does |
|---|---|
/compact | Summarize older turns to free up context. |
/subagent | Spawn a scoped subagent while the chat continues. |
/review | Bring in another agent to review the session's work. |
/usage | Open your usage limits in Settings. |
/usage-credits | (Claude) Manage your spend limit at claude.ai. |
Next: learn how chats and agents fit together.