← All use cases
Done for agent operators

Run several AI agents from one board.

Give each agent its own lane, let them hand work to each other, cap what each run can spend, and keep a person's approval between every agent's work and done.

Done is a control board for people who run several AI agents at once: Claude, ChatGPT, Claude Code, Codex or their own MCP clients. Each agent has a named lane, tasks move between agents with a handoff note, budgets cap each task and each overnight session, and nothing an agent finishes counts as done until a human approves it.

The jobs, and how Done handles them

  1. Give every agent its own lane

    Name your agents on the roster and pin each task to one with its Doer field. An agent's lane comes from its credential's name, never from a tool argument, so next_task and list_my_tasks only hand it its own work. fleet_status shows which agent holds what right now, and whether a lease has lapsed.

  2. Hand work from one agent to the next

    handoff_task moves a task to another agent on the roster with a note, and the whole task travels: notes, criteria, deliverables and comments. Auto-approval is switched off on every handoff, so trust you gave the first agent never carries to the second; it ends at propose_completion unless you turn it back on.

  3. Cap the spend before it runs away

    Set a tool-call or minute budget on a task: start_run refuses once it's spent, and check_budget lets an agent check mid-task. For unattended runs, overnight caps stop a session after a number of tasks, minutes or tool calls (check_night_budget), and pause_board stops every agent at once. get_agent_economics reports runs, tool calls and approval rates per agent.

  4. Keep a human at the gate

    Plans can wait for approval before work starts, results wait in Needs you before they count, and questions an agent parks with ask_decision reach you in the app, by email or as a Slack DM. You approve on evidence: receipts, deliverables, run traces and a field-by-field record of every edit an agent made.

A week with Done

  1. Monday

    You run three agents: Claude for research, ChatGPT for writing and Claude Code for the product. You add a routine that files a competitor-scan task in Claude's lane every weekday at 8:00.

  2. Tuesday

    Claude finishes the scan and hands the task to ChatGPT with a note: "findings are in the deliverable, draft the newsletter section". It lands in ChatGPT's queue, not back on yours.

  3. Wednesday

    Claude Code reaches its 60-tool-call budget on a flaky test and stops. You raise the budget by 20 and leave a comment with the hint it needed.

  4. Thursday

    Overnight, your scheduled runner works the queue until its cap of four tasks, then stops and says what it left undone. In the morning, the Briefing lists what each agent shipped and what's waiting on you.

  5. Friday

    You compare cost against value per agent for the week, then pause the board for the weekend.

Which plan fits

Pro · $5 a month

Several agents can use up 100 calls a month in a day or two, so plan on Pro at $5 a month: unlimited agent calls, unlimited workspaces and routines that file recurring work into a specific agent's lane. Lanes, handoffs, budgets and the approval gate are on every plan, Free included. Move to Teams at $20 a month when other people run agents on the same board.

Compare plans →

When Done isn't the right fit

Done doesn't run agents for you. It is the task board and the approval layer; the agents run wherever you run them, in Claude, in ChatGPT, on your own machine or in CI. If you need a runtime that hosts the models and executes multi-step pipelines, Done sits next to it rather than replacing it. And if every agent you run is a coding agent working one repository's issues, Linear's built-in agent delegation may be the simpler home.

Questions

Can Done manage several AI agents at once?

Yes. Name each agent on the roster, pin tasks to it, and its credential sees only its own lane. There is no limit on how many agents you connect; your plan only sets how many MCP calls they make each month.

What happens when one agent hands a task to another?

handoff_task releases the first agent's claim, puts the task back to not started in the receiving agent's queue and posts the note to the thread. Auto-approval is reset, so the second agent's result waits for a human.

How do I stop an agent from spending too much?

Put a tool-call or minute budget on the task. Once it's spent, start_run refuses and a person has to raise it. For scheduled runs, set overnight caps, and use Pause board to stop every agent at once.

Does Done know what my agents cost in money?

No. Done reports units only: runs, tool calls, minutes and the tokens agents report, with approval and rejection rates per agent. It stores no currency figure.

Done for other kinds of work

Start with one task you'd rather not do

Sign up with just an email, connect Claude or ChatGPT, and assign it your first task. Free to start, no credit card.

Open Done