Catch broken product analytics before and after it ships
Three tools for product and engineering teams on Mixpanel, PostHog or Amplitude. The Tracking Agent checks tracking on every pull request, Pravix alerts you when live data breaks, and the Datalyze MCP answers questions about it the Datalyze way.
- Tracking AgentPull request #4822 tracking changes proposed. The merge waits for a person to approve.$250 / month per workspace
- PravixDaily check, 05:00
checkout_startedis down 38% on last week.$49 / month - Datalyze MCPAsked in ClaudeWhy did week-4 retention drop? Start with activation by signup week.$49 / month
Teams Datalyze has worked with
A PR check for your analytics tracking
The Datalyze Tracking Agent checks every pull request for what it does to your analytics tracking, and holds the merge until a person approves. It runs on your own GitHub runner, on Claude with your own Anthropic key.
Datalyze Tracking Agent
2 tracking changes proposed
checkout_slot_selected: new step, no event yetorder_placed: now fires before payment. Breaking
Your reviewer
Approved in Datalyze
Datalyze Tracking Agent
Implemented and verified
Plan and code updated in your own tracking style. Every check passed.
Your reviewer
Approved the pull request
Nothing changes without a person's approval
Propose
The agent lists each tracking change the pull request needs, and flags anything it breaks.
Approve
Your reviewer edits or rejects items, then approves in Datalyze or with an
@datalyze applycomment.Implement and verify
The agent updates the tracking plan and the code in your own tracking style, then runs the plan's rules and an independent verifier.
Merge
The verified change merges through your team's normal review. There's no second Datalyze sign-off.
The tracking problems we find in every audit
- New behaviour, no eventA checkout step or feature that ships without tracking.
- Events that silently changeAn event renamed in place, or a call that now fires at a different moment.
- Events nothing firesCalls the change removed, and user properties nothing sets any more.
- Small slips with big costsPersonal data in a property, or
setwhere a first-time value needsset_once.
Your code never reaches Datalyze
| What | Where it goes | Does Datalyze see it? |
|---|---|---|
| Your code | Your GitHub runner, or a laptop | No |
| Claude's work | Anthropic, with your own API key | No |
| Analytics data | Your analytics tool, with your own access | No |
| Licence check | Datalyze | Only that your key was used |
Know the moment your data breaks
Pravix runs nine checks on the events in your Mixpanel or PostHog project, daily, weekly or on your own schedule, and tells you what broke and where. It keeps none of your data: each run's data is deleted when the run finishes.
- Missing events, properties that stop being set, and sudden shifts in volume or values
- Connects with a Mixpanel service account or a PostHog personal API key
- Leave out the events you don't care about, at any time
- Your first run is free, with no card
Pravix has its own sign-up on joinpravix.com for now. It's moving onto your Datalyze account.
- Missing
checkout_startedfell 38% against the last 7 days - Property
planis missing on 12% ofsignup_completed
Do analytics the Datalyze way in your AI tools
Connect the Datalyze MCP to Claude, ChatGPT, or Gemini. When you ask about your data, your AI follows Datalyze methods step by step.
Week-4 retention dropped after we shipped the new onboarding. Where do I look first?
Start with the step that changed, not the cohort curve:
- Compare activation by signup week, so you can tell a funnel change from a mix shift.
- Check that the new onboarding still fires the same events your retention metric depends on.
- Split week-4 retention by acquisition channel before you blame the flow.
Analytics questions, answered like a senior analyst would
Will this event schema hold up?
Paste your plan and get naming, property and identity problems flagged before they reach production.
Why don't my channel numbers add up?
Trace where double counting, missing UTMs and identity gaps are bending your attribution.
Which cohort is actually churning?
Get the cuts worth running first, and how to read the result without fooling yourself.
How should we define activation?
Pick a metric that predicts retention, and see how teams at your stage usually define it.
Each product is its own plan
Pay for the ones you use. Monthly, no seats, cancel any time.
Tracking Agent
- Every pull request checked
- Unlimited reviewers
- Runs on your own runner and Anthropic key
A workspace is your team and the repos you link to it, and everyone in it is covered. Claude's usage is billed by Anthropic to your own key.
Get the Tracking AgentPravix
- Unlimited runs and all nine checks
- 100k rows a month included, then $0.001 a row
- First run free, no card
Signs up on joinpravix.com for now, with its own account.
Try Pravix freeDatalyze MCP
- Every Datalyze workflow and the full knowledge base
- Works in Claude, ChatGPT, Gemini and coding agents
- One plan, no tiers
For one person. Each teammate subscribes from their own account.
Get the MCPQuestions about the Datalyze products
Do I need all three?
No. Each product works on its own and is its own plan. Start with the one closest to your problem: tracking that breaks as the code changes (the Tracking Agent), bad data already in your analytics tool (Pravix), or analysis that's slow or inconsistent (the MCP).
Is it one account?
The Tracking Agent and the Datalyze MCP share one Datalyze account, with one sign-in and one Billing page. Pravix has its own sign-up on joinpravix.com for now, and is moving onto the same account.
Which analytics tools do they work with?
The Tracking Agent works with any tool, because it writes new calls through your own tracking functions; it reads live data from Mixpanel, PostHog and Amplitude. Pravix checks Mixpanel and PostHog. The MCP works with whatever data connectors your AI app has.
Does Datalyze see our code or our data?
No. The Tracking Agent runs on your own GitHub runner with your own Anthropic key, so your code never reaches Datalyze. Pravix deletes each run's data when the run finishes. The MCP sends your AI the method; your AI queries your data through its own connectors.
What does Pravix check for?
Pravix runs nine detectors covering missing events, properties that stop being set, shifts in distributions and other anomalies, and reports each with where the fix is needed.
What is an MCP?
The Model Context Protocol is an open standard that lets an AI app such as Claude, ChatGPT or Gemini use outside tools. The Datalyze MCP gives your AI Datalyze's analytics method and knowledge base.
Do reviewers need a Datalyze account?
No. Only the person who pays needs one. Reviewers approve in Datalyze with their GitHub sign-in, or with an @datalyze apply comment on the pull request. Anyone with write access to the repo can approve, unless your config names a reviewers list.
What happens to our pull requests if the Tracking Agent subscription lapses?
They pass. The check reports that the Tracking Agent is inactive and the pull request wasn't checked, so a missed payment never blocks a release.