Automated AI pull request review for GitHub

Crongent reviews GitHub pull requests with an AI agent: your CI starts a run through the API, the agent reads the diff and the surrounding code with repo-scoped GitHub tools, and returns its findings by severity. The result goes to your webhook and the run page, and every tool call is logged.

The problem

Human review is the bottleneck on most teams, and generic AI review bots comment on style instead of risk. You want a reviewer you control: your prompt, your model, your repos, your rules.

How to set it up in Crongent

  1. Create a routine scoped to the repo, with your review instructions.
  2. Call the Crongent API from CI when a pull request opens or updates.
  3. The agent reads the diff and changed files with read-only GitHub tools.
  4. Findings land on the run page and go to your signed webhook; turn on writes to let it comment on the PR.

Example routine

Name
PR review
Trigger
API trigger on pull_request
Model
anthropic/claude-sonnet-5.5
Instructions
Review this pull request for correctness, security and data-loss risks. Ignore style. Cite file and line for every finding.

Who uses it

Frequently asked questions

Does the AI reviewer push code?
No. Crongent has no tool that pushes code. Routines are read-only by default; with writes turned on they can comment and edit issues, and each run gets a GitHub token limited to the routine's repos.
Which models can review pull requests?
Any model on the Vercel AI Gateway, including Claude, GPT, Gemini and open models, using your own key.

Related use cases