Home / About

About J-Bot Review

J-Bot Review is an open-source (MIT) agentic pull-request reviewer built by PGUP AI. It runs as a single GitHub Action inside your own CI, reviews with a model you already pay for, and posts diff-anchored findings back on the pull request. There is no reviewer SaaS in the loop and no per-seat bill.

Last updated August 29, 2026

Who makes it

PGUP AI is a small independent software project based in Canada. It publishes its work under the pgup-ai organisation on GitHub. J-Bot Review is the project's main public release: the Action that users install lives at pgup-ai/jbot-review-action and is listed on the GitHub Actions Marketplace. The code is MIT licensed, so you can read it, fork it, and run it without asking anyone.

Why it exists

Most AI code reviewers are hosted services. You pay per seat, and your diff leaves your infrastructure for a vendor's. J-Bot Review takes the opposite position: the reviewer is a container action that GitHub runs on your runner, reading your checkout read-only. The only place your diff goes is the model provider whose key you configured, under your own account and its retention terms.

That has a practical consequence for cost. The action adds no charge of its own — you pay your model provider, or nothing at all if you use a provider's free model route or reuse a coding-agent subscription you already hold, plus your normal CI minutes.

How it works

• One workflow file, one secret. The standard path is a single YAML file in .github/workflows plus one repository secret.
• Bring your own model. More than 30 backends are supported: direct model APIs and gateways, coding-agent CLI subscriptions, or any OpenAI-compatible endpoint you run yourself. You switch with one repository variable.
• Findings are verified before they post. A second model re-checks every blocking finding; refuted findings are dropped and uncertain ones are demoted, so what lands on the pull request is signal rather than a wall of nits.
• It reads your house rules. The reviewer discovers AGENTS.md, REVIEW.md, .coderabbit.yaml, greptile.json, and Cursor rules, and reviews against them.
• Least-privilege by default. contents: read, with write scopes only for pull-request comments and reactions.

Track record

The maintainers run J-Bot Review on their own work. As of August 29, 2026, its engines had metered 6,350 review runs and 1,059,006,673 tokens since June 2026, covering 4.2M+ diff lines across runs. A narrower GitHub-derived snapshot (August 29, 2026) counted 3,103 successful pull-request-triggered reviews across 8,524 files and 2,258,452 diff lines, each pull request's diff counted once (798 unique pull requests), in the two main private production repositories — narrower because it counts only successful pull-request-triggered runs in those two repositories. All of it is first-party dogfooding, not a customer adoption figure, and it is stated that way deliberately.

What is released and what is not

How we write about the product

Provider support, model availability, free routes, and pricing move quickly, and this site tries not to outrun them. Claims are checked against the action's source before publishing, guides carry the date they were verified, and routes that were read from a provider's catalogue but not yet exercised end to end say so. Where a free tier depends on a provider's terms — including data-retention terms that may not suit private code — the guide says that too.

Read more