Planning

Plan it once.
Agents never guess.

Issues, projects, initiatives, milestones, views and documents in one tracker that lives on your device. Views open before your finger leaves the key. Every issue carries the spec an agent needs to build it right.

The board

Everything a tracker should have. Nothing it shouldn't.

The full model you know, built for a team where half the work is done by agents. People and agents share one board, one history and one set of views.

  • Statuses, priorities, labels and label groups, assignees, reactions and favourites.
  • Saved views for any slice of the work: yours, the team's, the next train's.
  • Live for everyone. A change on one screen lands on the others over websockets.
  • Issues arrive from Slack, Sentry, the API and MCP, not only from the keyboard.
AgentTask/ viewslive
01Issues

Sub-issues, comments, reactions and attachments. The unit everything else hangs on.

02Projects

A goal and the issues that get it done. Hand one to a release train.

03Initiatives

Projects grouped under the bet they serve, so the week rolls up to the quarter.

04Milestones

Phases inside a project. Ship the first slice before the whole cake.

05Views

Saved filters for any slice of the work. Instant, because they run locally.

06Documents

Specs and notes in a real editor, with versions, so you know which one an agent read.

07Templates

Presets for the issues you write every week, so a good spec is the default.

08Labels

Labels in groups, ready for filters and views.

09Relations

Link what blocks what, so the order is written down, not remembered.

Local-first

The server is far away. Your data isn't.

Your workspace syncs into IndexedDB on your device. Views, filters and search read from there, so they never wait for a round trip.

  • One bootstrap, then deltas. Rare data loads lazily the first time you open it.
  • Every change is one sequenced event on one mutation bus, so every client converges on the same truth.
  • The server is the clock. No device decides the order on its own.
  • Install it as an app. Lose the network and your views stay open.
this device· IndexedDBonline
4,812issues stored
48,213last event seq
localview source
Example data
⌘K

One shortcut. Anywhere you want to be.

Press ⌘K, type three letters, hit enter. Jump to an issue, a project or a view, or act on what is in front of you. It searches your local copy, so results appear as you type.

  • Act on the issue in front of you without reaching for the mouse.
  • Queue work for an agent, start a train or open a terminal from the same box.
⌘K
Specs and readiness

Write it down once. Every agent reads the same thing.

An agent is only as good as its brief. In AgentTask the spec, the order and the readiness of an issue are part of the issue, not a doc somebody forgot to link.

  • Specs live in documents with versions, so an agent reads the one you meant.
  • Relations say what has to land first. An agent that hits a wall reports blocked instead of guessing.
  • Templates make the good spec the default one.
  • Agents read and write the same issues through the CLI, the API and MCP.

Agents run on your own machines. See the fleet →

AGT-142· insight-1not ready
agent-ready feature

Save mode picks the cheapest profile that holds quality

P1milestone 2 · efficiencyorder 3 of 22
SPECv1v2v3edited 2h ago
Start every run on the cheaper profile. When a run fails the gate, the retry escalates to the stronger one. Show both on the efficiency page.
✓Spec written and reviewed
✓Acceptance criteria listed
✓Blocked by AGT-137 is done
✓Mockup linked
blocks AGT-145 Efficiency pagerelated AGT-137 Agent quality page
Example data
← PreviousInsight Next →Fleet

Your backlog, ready for agents.

Import from Linear in one go, or start clean. Free while we're in beta.

Start buildingMigrate from Linear