Your AI agents.
On track.

Run Claude Code, Codex and Cursor from one place. See who is doing what, approve what matters, and know the moment one drifts.

See how it works

Free and open source. Your agents keep running on your own computer.

Good evening

2 agents are working. 1 needs your OK.

CTO asks Reviewer to check the timer testsWaiting for your OK

CTOFixing the lobby timer · Opus 5.5Working

ReviewerChecking the match tests · GPT-6 AstraWorking

DesignerUpdated the pricing page · Sonnet 5.5Done

On track today94%46 of 49 checked steps matched what you asked for.

Example data

Works with the agents you already use

Claude CodeCodexCursorAny MCP agentProduction traces

A team you can actually see.

Every agent gets a name, a job and a record of what it has done.

Know who did what.

Every reply carries the agent's name and the model behind it.

CTO · monkey-post-3b Opus 5.5
The timer now follows the server's clock, so a slow phone can't drift.

Give each agent a job.

Set a title and a description. It keeps them across every chat.

CTOReviewerDesignerQA

Claude can ask Codex.

One agent asks another for help. Nothing starts until you read the request and say yes.

CTO asks Reviewer to check the timer tests.Approve

Stopped before it happens.

Secrets and new dependencies are blocked first, and the agent is told why.

Blocked: writing to .env
Allow once, or keep it blocked.

One number you can trust.

On track is what share of your agents' checked actions matched what you asked for. Counted, never guessed.

94%

Example figure. By project, by agent and by check, and with nothing checked, there is no number.

Leave your desk. Not your work.

Send a message from your phone and it continues on your computer, with the same history. Your computer checks every message before it runs.

Measured, not promised.

Every claim was tested first, and each write-up names its limits.

11/11times a blocked agent corrected itself when told why 14 msthe cost of every check, before every action 60/60held-out drifts caught by the judge 0 lineschanged in the core engine to add production

Read what it cannot do yet

Up and running with one paste.

Copy the prompt, paste it into your agent, and it installs Trackline for itself.

Full install guide

Open source under Apache-2.0

Your AI agents.On track.