Docs
Data ownership and privacy
Your events live in a bucket in your own cloud account, in open formats. This page says what you own, what Torque Lift can reach and what it can't, and how to leave.
What you own
The events, the tables, the growth models, the tracking code and the flags are yours, and you can open any of them at any time. Apache Iceberg tables in your bucket are the system of record. The Torque Lift engine reads from them and is never the only copy.
Handing the stack to your own data team is a permissions change, not a migration. The columns are in Event tables.
What Torque Lift can reach
Your code, through a GitHub App
The App works on the repositories you select, and nothing else. It can read your repo's details, checks and commit statuses. It can write code to its own new branches and open pull requests. It has no access to workflows, secrets, administration settings, Actions or members, and it sends no webhooks.
Every change is a pull request. A person on your team reviews and merges each one. Torque Lift can't merge or approve a pull request, or push to your default branch.
Your bucket, through a few narrow roles
The setup stack creates one S3 bucket in your account, and a few roles Torque Lift uses to reach only that bucket. Each role trusts one Torque Lift identity, needs its own secret ID, and lasts at most 1 hour per session. Your CloudTrail logs each use. You can read the whole template on the AWS page before you click Create stack.
What the code agent can't reach
Your data. The agent that edits your code runs in a sandbox with no data credentials and no GitHub token. Your own typecheck, tests and build run in that sandbox before the pull request opens.
Your events go through an independent pipeline instead. It lands them in your bucket, builds your growth tables and measures each experiment.
Personal data
The rule is default deny. A property is stored only if your tracking plan declares it, with an allowed type, and none of the types holds free text. A blocklist of personal fields is enforced twice: in the tracking Torque Lift writes, and again as events land. It covers emails, names, phone numbers, addresses, locations, secrets and card numbers.
- A field Torque Lift doesn't recognize is dropped or set aside, never passed through.
- A value shaped like an email, a phone number or a card number is removed, even from an allowed property.
- Each row records how many values were removed, and the names of undeclared properties. It never records the values.
- A call that can't be kept goes to a turned-away table with a reason, and never the call itself.
The full blocklist, with the property types, is in Tracking.
Consent
Torque Lift's tracking sends nothing from a visitor's browser until your app records that they've agreed to analytics.
AI models
To draft a change, Torque Lift sends the code it needs to one AI model, through its own gateway. The provider's commercial terms apply: no training on that data by default, and limited retention. We don't promise the provider keeps nothing. Your pilot terms name the repositories this covers.
Anything else Torque Lift sends to a model is reduced first to counts, names and rates. Never user IDs, rows, emails or free text.
Pause, leave, or have data erased
- Stop Torque Lift at once. Suspend or uninstall the GitHub App: in GitHub, under Settings, then Applications.
- End its access to your bucket. Delete the stack in CloudFormation. That removes every role. The bucket and your data stay with you.
- Clean up on your side. Delete the setup token on the machine that ran the command,
~/.config/torque-lift/setup.json, and any service account you made in your analytics tool for the import. - Have data erased. For one person's data, or all of Torque Lift's copies of yours, write to us. We delete within 30 days and confirm in writing. Your bucket stays untouched.