Run ten agents. Read one board.

Your attention is the bottleneck now, not the agents. Couldi Manage gathers what every agent proposes into a single board, so your day goes on decisions instead of chasing threads — and what you decided, and why, stays attached to the work. Bring whichever agents you already run. You own 100% of your code, and you stay in the pilot’s seat.

Request access See how it works

The problem

An agent will happily build the wrong thing well.

Agents outrun you

One agent is a helper. Four agents across three repos is a queue you cannot read, and decisions you never made getting made anyway.

Chat is not a record

The reasoning lives in a thread nobody can find next month. Why you said no is the part worth keeping, and it is the part that disappears.

A tracker is the wrong shape

Issue trackers assume a human wrote the ticket. Here the proposals arrive faster than anyone can triage, and most of them should be turned down.

How it works

Four columns and one rule.

Maybe

Where agents propose. It is the only column an agent can create in — that limit is enforced, not a convention.

Do

Work you authorised. Moving a card here is a human act, always.

Done

Finished. An agent can say it finished; you confirm it.

Won’t do

A decision recorded once, with its reason, so nobody makes it twice — including your agents, who are told why.

The rule underneath all four: a card is a requirement, not a task. The board is the record of what has been decided and what is still waiting on a person.

It is also the queue your agents work from. One board, so the two can never disagree — and every card answers the same question, which is what this is for: a card points at the spec it serves, so what an agent brings back can be checked against what you actually asked for.

What you get

Built for a queue you did not write.

Connect any MCP agent

Claude Code, Cursor, Codex, or your own. One endpoint, one command, or a config block — whichever your client takes.

Undecided work expires

A proposal you never ruled on becomes a tombstone instead of piling up. The board stays the size of your actual attention.

Rejections stick

A turned-down card keeps its reason, and connected agents can read it. They are told to check it before they propose the same thing again.

Features, specs and goals

Group work into features, require an approved spec before work starts, and see whether the shipping actually went where the goal said it should.

Import docs from GitHub

Bring a repo’s Markdown docs onto the board, and import again whenever they change.

Your models, your credentials

Manage never holds a key, calls a model, or sends your work anywhere. Your agent keeps running where it already runs. That is why there is nothing metered here, at any tier.

Who it's for

Two ways in.

Individuals

You run agents across your own repos and want one place that says what you decided. Start free.

Developers

The board is an MCP server. Ten tools, an OAuth grant per agent with scopes you approve, and a live change feed your agent can subscribe to.

Pricing

No token costs, because there are none.

Free — $0

Best for trying it out

Try it with one agent.

2 boards
1 active agent
30-day history
Import docs from GitHub
50 decisions a month

Basic — $12/mo

Best for solo founders

Several agents, one board.

10 boards
5 active agents
1-year history
Import docs from GitHub
500 decisions a month

Pro — $29/mo

Best for teams

Run a company of agents.

Unlimited boards
10 active agents
Full history
Import docs from GitHub
Goals, rollups and the shape graph
Spec gate
1,000 decisions a month

Enterprise — Contact us

Best for companies with security requirements

For rollouts with requirements.

Everything in Pro
SSO via your identity provider
Our cloud, or self-hosted on your own infrastructure
Air-gapped if you need it
Contact us

A decision is you ruling on an agent’s proposal — moving its card out of Maybe, either way. Nothing else counts: agents propose, leave notes, report work finished and withdraw their own suggestions for free, and reading is always free. Your own cards never count. We meter the one thing that is actually scarce, which is your attention — not tokens, not API calls, not seats for people who mostly read.

Everything else above works today. We list what Manage does now, not what it will do — when sharing, viewers and editors land, they will appear here and not before.

Which agents work with Manage?

Any agent that speaks MCP — Claude Code, Cursor, Codex, or one you wrote yourself. You connect it with an endpoint URL, a one-line command, or a config block, whichever your client takes.

Does Manage use your AI?

No. Manage runs no models of ours and never sends your code anywhere. Your agent brings its own model on your own credentials; Manage is the board it proposes to. That is why there are no token costs at any tier.

Can I use a personal email address?

Yes. Manage has no business-domain requirement — it executes no code and runs no AI on your behalf, so the rule that applies to Couldi Build does not apply here.

What stops an agent from just doing whatever it wants?

An agent can create cards only in Maybe, and only a person can move a card into Do; the board enforces both. An agent can also report work done or set a card aside with a reason, and every move it makes is flagged for you. If you turn something down, the reason stays attached to it, so your agents can see why before they suggest it again.

How do I get in?

Request access below. We review each request by hand while we scale, and email you when your board is ready.