Ozone documentation
Ozone is an AI security engineer for your codebase. It reviews every pull request the moment it opens, digs through the changed code the way a human reviewer would — reading files, tracing call paths, running searches — and reports real, actionable findings back to the pull request and your dashboard.
What Ozone does#
- Reviews pull requests automatically. Connect a repository and every new PR gets a security review. The result lands as one comment on the PR, updated in place, ranked by severity, with triage actions only your team can use — and in your dashboard, and as a check you can require.
- Maintains what it finds. When your code or findings move, an agent rechecks open findings against current code, spots duplicates, and closes what it can prove, leaving the doubtful ones open.
- Analyses whole projects. Sweep every repository in a project on demand or on a schedule — nightly, weekly, whatever cadence suits you.
- Answers questions. A built-in assistant digs through your runs and findings, starts reviews, manages schedules, and passes harder code questions to the review agent.
- Learns your codebase. A knowledge library your team writes, and an agent keeps current as the code changes, gives later reviews your conventions and your past decisions.
How it works#
Each review runs in an isolated, disposable sandbox. Ozone clones the repository, and an agent examines the change with real tools — file reads, greps, git history — rather than skimming a diff in one pass. Everything the agent does is recorded in an activity feed you can inspect afterwards, finding by finding, tool call by tool call.
Where to start#
- Getting started — connect GitHub and get your first review in a few minutes.
- Projects & repositories — control what triggers reviews with per-repository rules.
- Reviews & runs — what happens during a review, and how to steer it from the PR.
- Findings — triage, activity, and the finding lifecycle.
Building an integration instead? See API & MCP and the API reference.