Owlet lens

AI code reviewer for pull requests. It posts findings as inline review threads, reads the developer's replies, and closes threads that are fixed or declined

See our solution in action.

The problem we set out to solve

Human reviews are slow and inconsistent, PRs wait for reviewers, repetitive issues consume time that could be spent on deeper technical feedback, and the same mistakes keep appearing. Generic AI review bots add comments but rarely manage the review process, leaving unresolved threads and extra work for developers.

From problem to possibility

Owlet Lens uses AI to automate pull request reviews, identifying code issues and managing review discussions. It analyses code changes against relevant files and team coding standards, then posts findings as inline comments. It also tracks developer replies and automatically updates thread statuses and a summary comment. Running directly in GitHub Actions, it supports multiple LLM providers without requiring a separate server or database.

Key Feature & Modules

Build around what matters

Inline AI review on every pull request

Inline AI review on every pull request

Each new or updated PR is reviewed automatically, and findings appear as inline threads on the exact lines concerned, plus one summary comment kept up to date. Developers get feedback in about a minute or two, before a human reviewer opens the PR.

It talks back and cleans up after itself

It talks back and cleans up after itself

The tool reads each developer reply and decides whether the issue is fixed, will not be fixed, or is still open. Fixed threads close, regressions reopen, and the thread list always matches the code instead of piling up as stale comments.

Learns your team's standards

Learns your team's standards

Findings that the team keeps declining are mined and turned into proposed coding-standards rules in a pull request. A person always approves the PR; nothing is merged automatically.

Runs where you are, with your choice of AI model

Runs where you are, with your choice of AI model

Works on GitHub, GitLab, Bitbucket Cloud and Azure DevOps, with several LLM providers, including local models, and nothing to host. Secret-shaped text is redacted before the diff is sent, and a dry-run mode previews results without posting.


See How Your Code Connects With Owlet Lens

Introduction

How It Performed

Processing time

Processing time

Findings typically appear within a minute or two of the PR being opened (README guidance). Observed CI review jobs on our repositories took from about 45 seconds to 2 minutes.

Capacity

Capacity

Runs per PR in CI, so capacity follows the CI runners. A typical PR takes about two model calls.

Pipeline Performance

Pipeline Performance

PR opened or pushed → fetch the diff and context → redact secrets → LLM review → reconcile with existing threads (post / reply / reopen / close) → update summary comment. Developer reply → LLM verdict (fixed / wontfix / active) → update thread.