How we handle your data

What is recorded

Your source code is not stored. For the few commits Procket cannot place from recorded events, it reads the diff from GitHub to compute a fingerprint, then discards it. Diffs never reach our database or logs.

Only repositories your workspace connects are recorded. The client does not capture anything in other repositories on the same machine.

Where it is stored

• On Amazon Web Services in Ireland (eu-west-1). Customer data stays in the EU.
• Encrypted in transit with TLS 1.2 or later, and at rest.
• Prompt text is additionally encrypted with keys specific to each workspace, person and month. Paying workspaces get a dedicated key in AWS Key Management Service.
• Backups are encrypted and kept for at least 7 days.

Who can read it

• Reports follow GitHub permissions: you see prompts for repositories you can read on GitHub, checked each time you sign in.
• Engineers see their own prompts in the team report. There are no leaderboards or per-person rankings.
• The credential on each laptop can record events but cannot read anything. A stolen laptop exposes nobody's prompts.
• Every read of prompt data is logged, and workspace admins can see the log.
• Our staff access production data only to investigate an incident or a support request you make. That access is recorded in an audit log.

Comments on pull requests are visible to anyone who can read the repository. On public repositories, Procket posts only a summary by default, never prompt text.

How long it is kept

• Prompt text is erased automatically after 12 months by default. Admins can choose from 3 months to 10 years.
• An admin can erase everything recorded for one person at any time, for example when they leave.
• When a workspace closes, its data is exported on request and deleted 30 days later.

Tamper evidence, and its limits

Every event is hash-chained to the one before it, in a database where the application cannot update or delete rows, so an edited record shows. Each pull request comment also publishes the latest hash of every chain it covers, on GitHub, in your own repository, where we cannot change it. If records shown there were later removed or rewritten, the chain no longer contains that hash, so anyone holding the comment can see it. Records written after the last comment have only the chain's own protection.

Two things this cannot show. A prompt sent while the hooks were turned off was never recorded; the commits it produced appear as “not in the ledger”, which is itself visible on the pull request. And linking a prompt to a commit by the files it changed shows where code came from, not who wrote every line.

Redaction

Private keys, cloud and API keys, tokens for GitHub, GitLab, Slack, Stripe and npm, JWTs, passwords in URLs and assignments, and high-entropy strings are removed before a prompt is saved or sent, and again on the server. Pattern matching cannot recognise every secret, so treat any credential pasted into an AI tool as exposed and rotate it.

Your responsibilities

Recording employees' prompts is workplace monitoring under UK and EU data protection law. Before rolling Procket out, you will usually need a data protection impact assessment, and you must tell your engineers what is recorded and why. On request, we provide a description of the processing to help with both.

Sub-processors

Amazon Web Services hosts the service. The full list, with what each one can access, is published with our data processing agreement.

Reporting a vulnerability

Email security@procket.dev. We will acknowledge your report, keep you updated, and not take action against good-faith research that avoids other customers' data.