Why Squirlnote: a board your agents can actually use

We kept running into the same problem: a long conversation with an AI coding agent across a dozen projects, and no shared, scannable record of what was pending, what was half-done, and what was actually finished. Notion is cloud-locked and awkward to script against. A markdown file in the repo drifts the moment two agents touch it at once. Project-management tools assume a human is the only thing moving cards.

So we built Squirlnote around a different assumption: an AI agent is a normal client of the board, not an afterthought bolted on with a webhook.

What that means in practice

Squirlnote is a Kanban board and a nested-notes tree, and it ships a built-in MCP server alongside the web app. The same task an agent creates with create_task shows up as a card you can drag, comment on, and label in the browser — and the reverse holds too. There's no sync job, no export step, no "markdown mirror" to keep matching by hand.

One rule we didn't compromise on

An agent can move a task forward — Pending, On-Going, For Review — entirely on its own. It cannot mark a task Done. That's a human call, always. We'd rather have a board that's slightly behind on bookkeeping than one where "done" quietly stops meaning "a person checked it."

Where it's headed

The next slice is public docs and a changelog (this page is part of that), so the project is legible to people outside the one account that's been dogfooding it. After that: billing, so teams beyond our own can actually run on it.

If you're managing more than one project with an AI agent doing real work inside it, this is the tool we wished existed before we built it.