The whole release cycle, in one place

Track: your team's spine.

People and AI agents plan, build and prove work together. Tests that replay and pass sit at the centre: work is done when evidence says so, and a release ships when its proof is in.

Get started Sign in

Plan in goals→agents work under rules→Track proves it

Works with Claude Code · Playwright · Selenium · Cypress

Track (testing itself)
Roadmap · milestone goals

Trust and identity12 / 15

Track runs the team: people and agents26 / 39

Tests people can trust14 / 41

Studio people enjoy1 / 16

NowScore every test A to FClaude Codereplayed · passed

NowReadiness on each milestonea personneeds proof

NextApprove before an agent deletesAI agentwaits for approval

Illustration from Track's own board (goals and item counts as of 15 Oct 2026).

How it works

The whole release cycle is one loop.

Plan, build, prove, release, learn, and round again. Every step is connected to the others, so nothing gets lost between a goal and the release that delivers it.

1Plan

Goals first, then the work, then who does it.

A goal is an outcome you want. Work sits under it as Now, Next or Later, and every item has an owner: a person or an AI agent.

Releases → goals → items, with the Inbox for what Track found, requests and ideas anyone shared. "Plan it" puts an item under a goal.
A live roadmap: milestones with due dates, days left, a Gantt and a burn-up, and a readiness for each (on track, at risk, late).
A capability map of what people can do in your app, its risks, data and roles, with evidence linked to tests.

2Build

People and AI agents build together, under your rules.

Agents such as Claude Code connect to Track, read the guidelines, claim work, report progress with evidence and ask a person when unsure.

Guidelines: versioned rules, a definition of done, what never to do and what to ask a person first.
Approvals: deleting, changing people or secrets, changing a release or settings, or spending over the limit waits for a person. An approval works once, for exactly that action.
Identity: every agent has its own key, an owner and never more rights than that owner. Everything it does is attributed to it.
Messages and pause: agents and people talk in threads on the work; one press pauses an agent at once.

3Prove

Track explores your app and writes tests you can trust.

Track learns what your web app does, writes journeys, and a test only counts when it replays and passes.

Replays on Playwright, Selenium or Cypress, with the same verdict on all three.
Heals a test when the page changed, and replays every proven test on each check.
Grades every test A to F on proof, flakiness, selectors, secrets, length, waits and duplicates. Weak tests become work.
Failures become work: defects, quality and readiness findings land on the board and close themselves when a later check proves them fixed.

4Ship

Releases show progress next to the verdict.

Each milestone shows items done, by evidence and by people, beside the ship verdict and production readiness.

Evidence-closed work: an item is done with a commit, a test that replays and passes, or a passed check.
Production readiness: requests per page load (too many, duplicates, one per item, heavy, slow, refused) and several people at once.
Serious findings block the release: a test that proves nothing, or shows a secret, is not evidence a release can ship on.

5Learn

Everything is connected, and what you learn stays.

Open any work item, test or release and unfold Connected: what it is tied to, the numbers that roll up and what happened around it.

Connections across work, tests, results, runs, capabilities, releases, knowledge, people, agents, sessions, messages, approvals, commits and GitHub issues and pull requests.
Knowledge: live sections from real data plus what people write, with a page you can publish.
Measures that roll up along the links, and the activity around each thing.

Reliable tests at the centre

A passing test is not automatically a good test.

It may check nothing, check something that was true anyway, or carry a password in plain text. Track scores every test so people can see which ones to trust and what to change.

Every test gets a grade

Every test starts at 100 and loses points per finding. A test with no real proof, or a literal secret, is always an F, and serious findings block every release.

ABCDF

Proof35

Flakiness20

Selectors15

Secrets10

Length10

Waits5

Redundancy5

Tests people made are never changed by Track: their findings are suggestions, assigned to the test's author.

Runs where you already test

One proven test replays on three engines behind one interface. What an engine cannot run is said before a browser opens.

PlaywrightSelenium WebDriverCypress

Run checks on Track, or on your own machines and CI with runners. The engine, AI keys and secrets stay in Track; a runner only runs what it is given.

Economics: measured, not claimed

We measured Track's decision engine, Jev, against LLMs on 3 practice sites (69 goals, 36 runs, 2026-09-30), every result checked against human-written ground truth and replayed 3 more times.

Engine	Good tests	Per proven test
Jev (Track's default)	71%	$0.002
Claude Haiku 4.5	74%	$0.083
Claude Sonnet 5	72%	$0.119

LLM quality for about 1/40 of the cost. You can still choose any model for decisions, per organisation or per project.

Everything in it

All of it, in one product.

Planning, the people and agents doing the work, the proof and the release, without stitching tools together.

Roadmap and milestones

Computed live from the board: due dates, % done, a Gantt, a burn-up and a readiness for each milestone.

Capability map

What people can do in your app, its risks, data and roles, with evidence links to tests.

Assistant (Coming soon)

An optional add-on that will answer from your projects and act with your permissions.

Ideas from anyone

Anyone, even a viewer, can share an idea. It lands in the project's Inbox for the team to plan.

Claude Code plugin and MCP

A thin plugin and an MCP server with 83 tools. Sign in with OAuth, nothing to copy, or use a key.

Runners

Run checks on your own machines or in CI, such as GitHub Actions, with jobs and leases.

Three test engines

Playwright, Selenium WebDriver and Cypress, across Chromium, Firefox, WebKit, Chrome, Edge and Safari.

Roles and permissions

Fine-grained permissions, built-in and custom roles, project grants, teams and an audit log.

Single sign-on and two-step sign-in

Google and Microsoft sign-in, two-factor with an authenticator app and recovery codes, required org-wide if you like.

Billing and budgets

Plans with Stripe, a usage ledger attributed to org, team, person, project and run, and budgets that stop spending.

Your choice of AI model

The model for each job, picked per organisation from the live OpenRouter list, with live prices. Jev is the default for decisions.

Docs inside Studio

Every command with a real example, how each part works, and a short tour on your own data.

With Claude Code

Five minutes to a governed agent.

Connected to Track, Claude Code sees your project's plan, picks up work, records what it did and keeps its work under a goal you set. Each session starts with your open work and its guidelines.

/track:goalcreates a goal, or finds it, and makes it this session's goal

/track:statusthe plan: what is happening now, your goal's progress, what waits on you

/track:workyour work, and help claiming the next item

/track:ideashares an idea with the team; it lands in the project's Inbox

terminal

# 1. install the plugin (once per computer)
$ claude plugin marketplace add LETMETESTIT2/track
$ claude plugin install track@track

# 2. in Claude Code: sign in with Track
/mcp   → plugin:track:track → Authenticate

In Studio, Settings › Connect Claude Code shows the same steps with copy buttons and tells you when Claude Code has connected. Read the guide.

Pricing

Start free. Grow when your team does.

The organisation is the customer. Plans are about features and limits; payment runs through Stripe.

Free

For trying Track on a side project

Projects3
Members3
Runners—
Checks a day20
AI spend included a month$1

Get started

Team

For a team testing the apps it ships

$49 / month

or $490 a year

Projects20
Members15
Runners3
Checks a day500
AI spend included a month$25
Runners on your machines or CI
Several environments per project

Get started

Business

For organizations with many projects and stricter access

$199 / month

or $1,990 a year

Projectsno limit
Members100
Runners20
Checks a day5,000
AI spend included a month$150
Everything in Team
Single sign-on
Custom roles and per-project permissions
Live support sessions with the Track team

Get started

Every organisation starts on Free and upgrades in Settings › Billing, where Stripe handles cards, invoices and cancelling. Design partners and internal teams can be given a plan by the Track team.

Plan it. Build it. Prove it.

Give your people and your AI agents one spine to work along, with proof at every step.

Get started Sign in