Reads your code
On your own repository, through GitHub. Tracks reads what the product actually does, not what a ticket says it does.
Product
Tracks finds what to build next, puts it in front of you with the evidence, builds what you approve, and measures whether it worked. Four steps, one page.
01 · Find
Tracks reads your code and your usage data side by side and works out what to build next: the step where people leave, the error that reaches the most users, the screen nobody opens.
An example from Leafbox, a sample product
Reads
Finds
19 of 99 orders were charged less than their shipping cost.
Proposes
On your own repository, through GitHub. Tracks reads what the product actually does, not what a ticket says it does.
It reads your data and cannot change it.
Every finding comes with the number that shows why it matters, in plain words.
Tracks reads where your code, your usage data and your errors already live. It reads your data and cannot change it.
02 · Decide
Each idea lands on a board as a card: what is wrong, what to change, and the evidence behind it. In plain words, so the person deciding never has to read code. Open a proposed card to see one.
Leafbox is a sample product. One card crosses its board by itself.
What is wrong, what to change, and the evidence behind it. No code to read.
Approve it, ask a question, or reject it and say why. The team remembers the reason.
The strip at the top of the board lists what is waiting on a person. Everything else keeps moving.
03 · Build
Tracks's own teams run on Claude: they read your product, propose cards, challenge each other's proposals and build what you approve. Or add Tracks to the assistant you already use.
From approval to pull request
By Tracks's teams, or by your own assistant with its own tools.
Tracks checks every change before it reaches review. When a check fails, the builder reads why and fixes it.
A pull request on your own repository, explained for people who do not read code.
A coupon can make the plan free, never the total negative
How to try it
Tracks's own teams run on Claude, and they challenge each other's proposals before one reaches you.
MCP is the open standard for connecting assistants to tools. Your assistant reads the board, proposes cards, attaches designs and builds approved work.
Whoever does the work, Tracks keeps the record and checks every change before it reaches review.
04 · Measure
Afterwards Tracks reads the number again. Improved, unchanged or worse: the card says which. Every decision, by a person or by an agent, leaves a record you can go back to.
Orders with a wrong total
19 → 0
The card says which: ImprovedUnchangedWorse
05 · Control
Tracks finds the work and writes the card. Nothing is built by Tracks.
You approve a card. Tracks builds it, checks it and delivers the pull request.
Once you turn it on, small safe changes go out on their own. Everything else still waits for you.
At every level
| Step | Who does it |
|---|---|
| Finds the work | Tracks |
| Writes the card | Tracks |
| Decides | You |
| Builds and checks | You |
| Delivers | You |
| Measures the result | Tracks |
| Step | Who does it |
|---|---|
| Finds the work | Tracks |
| Writes the card | Tracks |
| Decides | You |
| Builds and checks | Tracks |
| Merges the pull request | You |
| Measures the result | Tracks |
| Step | Who does it |
|---|---|
| Finds the work | Tracks |
| Writes the card | Tracks |
| Decides on small safe changes | Tracks |
| Decides on everything else | You |
| Builds, checks and delivers | Tracks |
| Measures the result | Tracks |
every row is written to the record
Connect a repository and Tracks puts its first cards on your board. Start with propose only: nothing is built until you say so.
Get started