Keelnote — powered by the open-source think-and-ship engine.
A roadmap that builds itself as your agents work.
Your team ships with AI agents. Keelnote is the shared, living record of why things were built, what was actually done, who asked for it, and what's next — written as the work happens, kept for the people accountable for it.
The problem
AI agents do more of your work every quarter. The context they create evaporates.
“Why is it built this way?”
The decision was made in a chat session that no longer exists. So your team re-litigates it — or repeats the mistake.
“Where are we, really?”
Status is hearsay collected in meetings, stale the moment it's typed. “Done” is a feeling, not a fact.
“Who asked for this?”
Customer and stakeholder requests scatter across inboxes and DMs — disconnected from the work they caused.
What it is
The system of record for AI-assisted work.
Git versions the code. Keelnote versions the intent behind it — the reasoning, the plan, the execution, and the stakeholder asks — as one connected, queryable history. It doesn't run your agents, and it isn't another tracker to fill in: the record is written as the work happens, and your people curate what it means.
Why
Every decision, the thinking behind it, and how confident it was — recorded as the work happens.
What
Tasks, actions, and quality gates. “Done” means a recorded check passed — not a ticked box.
What's next
A living plan that updates as work completes — and remembers why it changed.
What's asked
Every request from customers and colleagues, captured and tracked to a resolution.
All four stay linked, so the history reads as one story — a full audit trail from what the agent thought to what shipped.
Why teams adopt it
Built for people who are never in the same room.
Institutional memory, by default
Decisions keep their reasoning attached. The history lives in your git repo in an open format — reviewable in pull requests, readable in ten years, yours either way.
Cut the coordination tax
People and agents work the same project in parallel without overwriting each other. Status meetings become optional: the record is already current.
Everyone has a door
Engineers and agents work from the editor; leadership watches a live workspace; stakeholders file requests by plain email or a share link — no account, no training.
Requests don't interrupt — evidence does
Every ask is captured, researched by agents in the background, and surfaced only when it's worth a human's attention. An agent can't raise a guess.
AI proposes, your people decide
Nothing enters the roadmap without human approval. Agents write the record; humans curate it — enforced by the system, not a policy doc.
Dashboards that refuse to lie
Completion renders only from recorded quality gates. Missing data says “no data” — never a fabricated number.
Pricing
Tiers that match how teams actually grow.
Team
Per-seat billing that follows your roster.
- Per-seat licensing — no wall
- Unlimited retention
Enterprise
Self-host, SSO, zero-data-retention.
- Everything in Team
- Self-host or dedicated
- SSO + audit
Current prices are shown at checkout, in the app. Anonymized telemetry is opt-in on every tier — it is a posture, not a plan feature.
Open source
The engine is open. Your records are yours.
Keelnote runs on think-and-ship — an open-source (MIT) MCP server, published on npm and crates.io. It works with MCP-capable coding agents like Claude Code, Cursor, and Windsurf. Your reasoning and execution history can live in your own git repository, in a format compatible with the open Agent Trace standard — readable with or without us.
Privacy
Your content stays yours.
Opt-in, every tier
Anonymized telemetry is off by default on every tier. Nothing is collected or sent until you explicitly turn it on — and you can turn it off any time.
Never sent
The disclosure is explicit: prose, code, titles, descriptions, file paths, emails, ids, and secrets are never sent. Only structural shape — counts, statuses, topology.
Safe in aggregate
Cohort comparisons only unlock above an anonymity threshold of k = 5 contributors. Below it, nothing is readable — not even the contributor count. Self-hosted deployments emit nothing at all.