Chat is the primary way you work with Crew. Type a message, and the agent responds with streaming text, live tool calls, and results — all in real time. Attach files, dictate prompts, interact with widgets, or let it run tools autonomously. The experience works the same whether you're in the desktop app, web dashboard, Slack, or the CLI.
@filename), or a voice memoEach chat tab is its own independent session. Sessions share your long-term memory (preferences, lessons, project context) but have their own conversation history and tool-approval state.
| Action | How |
|---|---|
| Attach a file | Type @ to open a fuzzy file picker scoped to your project |
| Send an image | Drag an image or paste from clipboard — the agent sees it via vision |
| Dictate a message | Press the mic button for streaming speech-to-text |
| Hear the reply | Enable TTS (speaker icon) for local Piper text-to-speech |
| Sharpen a prompt | Cmd+Shift+Enter (or the ✨ button) rewrites it before sending |
| Auto-approve tools | Toggle Autopilot per session — deny rules still apply |
| Go incognito | Session menu → nothing persists to long-term memory |
| Switch models | Header dropdown to change the model for this session |
| Set reasoning effort | Per-message low/medium/high for supported models |
Alongside the conversation, a side panel hosts context for the work at hand. It docks to the right or the bottom (toggle from the panel header) — docking to the bottom suits a tall or narrow monitor.
Opening a Jira link shows the issue's details right in the side panel instead of sending you to the browser. GitHub, GitLab, and Jira issue, PR, and MR links also render inline in chat as chips — an icon plus owner/repo#N.
The transcript renders code changes and files where you are already looking, so you don't switch to another tool to read them.
A launcher opens by keystroke and names each row's kind, so you can jump to a session, file, or action without a full-corpus scan on every keypress. It ships on by default.
Chat