Security and data handling

codqual reads your source code to review it. Here is exactly what it sends to the model, how long it keeps anything, and who can see the results.

Your code and the model

codqual reviews your code by sending it to LLM providers. Two things people ask first: whether we train on it, and what we send.

No — codqual has no training pipeline and never uses your code to train any model. To produce a review, codqual sends your code to LLM providers (Anthropic, and others via OpenRouter) solely to generate that review; how each provider handles data is governed by its own terms.

The diff of your pull request plus bounded excerpts of the surrounding code — a window of lines around each changed or cited location, not entire files or your whole repository. codqual sends only what a reviewer needs to judge the change in context.

How long we keep your code

codqual is built to hold as little of your code as possible.

codqual keeps no standing copy of your repository. The working copy used for a review runs in a temporary workspace that is deleted the moment the review finishes.

Who can see your code and findings

Access is scoped at every layer: which repositories codqual can read, who on your team sees the results, and how to cut off access entirely.

Only the ones you choose. codqual installs as a GitHub App with access scoped to the repositories you select during install — and you can change that selection any time from GitHub.

Owners and admins see every repo and finding in the org by default — a centralized view meant for governance. Regular members see findings only for repositories they've been explicitly granted access to, directly or through a team; without a grant, a member sees nothing for that repo. Access is closed by default, never shared automatically across the org.

Disconnecting a GitHub organization removes its repositories from codqual right away and stops all future reviews. codqual also asks GitHub to revoke its own access.