Agents board

When you would want it

One assistant doing one thing at a time keeps you waiting. Several agents can fix a bug, write tests and polish a page at once while you keep chatting, and each box shows at a glance which one is busy, which one needs you and which one is done - no scrolling back through a transcript to find out.

Where to find it

• All Features → Agents board
• Open Agents
• New task

What you get out of it

• Several agents work at once while you keep chatting - a bug fix, tests and a UI polish no longer queue behind each other.
• Each agent works in its own copy of the project, so a half-finished edit never lands in your files.
• One glance says what is working, what needs you and what is ready - no scrolling back through transcripts.
• Cost shows on every box when the price is known, and a cap stops an agent before it spends more.
• A read-only Reviewer really cannot edit: tool access is enforced when the agent runs, not just written in its instructions.
• Proof, not promises: you see which checks actually ran and passed before you review a line, and risky changes are flagged or held.
• Agents that loop get stopped instead of burning your key, and parallel agents warn you before they step on each other.

How to use it, step by step

1. Choose Agents in the Vibe sidebar (in the IDE window, Agents in the More menu or Open Agents in the command palette).
2. Your agents starts with five specialists that use your current model. Edit one to pin a specific connection and model instead, or change its instructions, appearance and tool access.
3. Press New task. Describe what you want - for example: Add a dark mode toggle to the settings page.
4. Pick an agent, or keep Default agent to use your connected model with every tool. Optionally set a cost cap in dollars, then press Enter.
5. Watch its box under Working: the mascot thinks while it works, the status line says what it is doing right now, and the time, files, tokens and cost count up. Start more tasks straight away - they run side by side, and extra ones wait as Queued.
6. When it finishes: with Apply automatically on, the work lands in your project and the box offers Undo and Continue. With Review every change, the box moves to Ready - press Review to see the diff, then Apply, Apply & continue, or Discard.
7. Anything under Needs you is waiting on you: Resolve a clash with your own edits, Run again after an error or Open in chat to talk it through, Run with a higher cap when a cost cap stopped it, or Resume after Agentfly restarted.
8. Read the proof of work on each finished box: a tick or cross for every test, type check or linter the agent actually ran, and warnings for a possible secret, sign-in or payment code, removed tests, new dependencies or a very large change. With Apply automatically on, a failed check or a possible secret holds the run under Ready for you instead of applying it.
9. If an agent keeps trying the same failing thing, Agentfly stops it; press Try another way to run it again with a nudge to change approach. If two agents edit the same file, both boxes warn you before their work collides.
10. Select any box to see what the agent reported, the files it changed, its proof of work and every step it took.
11. Under Your agents, Run hands a saved agent a task, Edit changes its look, instructions, model and tools, and Select for a Team splits one task across two to four agents.

Why this beats the usual way

Chat-only assistants do one thing at a time and bury progress in a transcript. Cloud agent services run in parallel but bill by the minute and hide what the agent is doing until it opens a pull request. Agentfly runs parallel agents on your own machine and key, shows each one's live status and cost as a box, and brings its work back through the same review and Undo as the chat.

Find it in Universal Search

Agents board · The workbench · agents · agent board · parallel agents · background agents · multiple agents · run in background · kanban · boxes · tasks · ai tasks · queued · needs you · ready to review · apply and continue · cost cap · starter agents · reviewer · tester · bug fixer · like cursor agents · like devin