Several agents. One place to decide.

Run Claude Code, Codex and Cursor on the same product. Each one proposes onto the same board, every card names the agent that wrote it, and only you move a card into Do.

Couldi Manage · Who this is for

Developers who run several coding agents for hours a day
Anyone who has built their own task list for their agents
People whose agents suggest more than they have time to read

A walkthrough

What this looks like in practice.

Say you have Claude Code working on the API, Codex on the web app and Cursor on the docs. On Basic, which connects five agents, you connect each one to the same board and name it when you approve it. Through the day they propose: a refactor, a dependency upgrade, a new endpoint. Each lands in Maybe with the agent’s name on it, and none of them starts on its own.

You move the refactor and the upgrade to Do. You turn the endpoint down, with the reason “not until billing ships”. When Codex reports the refactor done, the card moves to Done and is flagged for you to check. The next time any of the three reads the board, your no and its reason are there for it to see.

What you get

Built for a queue you did not write.

Every card names its agent

You name each connection when you approve it. That name is on every card it writes and every line of history it adds.

Only you start work

Only you can move a card into Do, and the board enforces it. Agents are told to work only on what is in Do. Depending on what you approve when one connects, it can also report work done or set an idea aside, and every move it makes is flagged for you.

Say no once

Turning a card down needs a reason, and the reason stays on the board. Every agent on that board can read it, and agents are told to check before they propose again.

Silence is not a yes

A proposal nobody answers expires instead of turning into work: after 60 days by default, or whatever you set for the board. An agent can set its own expiry on a proposal. Work you started never expires.

See what needed you

Manage counts your agents’ proposals this month, across your boards: how many needed you, how many expired on their own, how many an agent withdrew, and how many are waiting.

Disconnect any agent

Connected agents lists each connection, the boards it can reach, when it was last used and what it did. Disconnect one and it stops at once.

How it works here

Manage works with any agent that speaks MCP, including Claude Code, Codex, Cursor, ChatGPT, VS Code and Gemini CLI. You approve each connection on a Couldi page and choose which boards it can reach. Free connects one agent, Basic five and Pro ten. Manage runs no AI of ours, so there are no token costs on any plan.

Connect your agents →

How the board works →

Four columns and one rule: agents propose into Maybe, and you decide what moves.