# Entelligence Code Review

> Entelligence checks every pull request against your codebase, team rules, and resolved incidents, and cites the precedent behind each finding.

Source: https://entelligence.ai/code-review

Entelligence Code Review checks every diff against your codebase, standards, and resolved incidents, citing exact precedent so the same class of bug never ships twice.

## Every review reads past the diff

Callers, dependencies, Guidelines, Learnings, linked repositories, and production incidents.

- **Only high-signal findings reach the PR.** Set the bar per repository: critical only, medium and above, or everything. Cap inline comments or post one summary instead.
- **Review against your Guidelines.** Rules that always apply. Write them, link docs you already keep, or generate them from review history. Assign per repository.
- **Follow dependencies across repos.** Finds every caller of a changed function, including callers in linked repositories the PR didn’t touch.
- **Every finding comes with evidence.** Each finding names the file, the line and the call sites it checked, so you can verify it before you act.
- **Enforce your merge policy.** Set the conditions a PR must meet before Entelligence approves it. It never merges on its own.
- **Learn from how your team reviews.** Generate Guidelines from the comments your engineers already leave. Findings they mark unhelpful become Learnings and come up less.

## One pull request, followed from push to merge

- **Push your PR.** Nothing is configured at this point and nobody has been assigned. The review is queued because the repository is connected.
- **It maps the change.** The review reads past the diff: every caller of the changed function, the rule that covers it, and the last incident it caused.
- **It runs the review.** The checks run in parallel. The rule and the incident point at the same line, so they rank as one critical finding, not two comments.
- **Findings land on the PR.** The comment sits on the line it is about, with a suggestion you can commit and a prompt you can hand to your coding agent.
- **Fix or merge.** Every push is re-reviewed. If you turn it on, a 5/5 review submits an approval through your Git provider. Required checks still decide the merge.

## Review wherever the code gets written.

Findings follow the code into your editor, your terminal and your coding agent, and every new commit gets a fresh review.

- **Hand a finding to your agent.** Each finding carries a ready prompt for Claude Code or Cursor.
- **See reviews in your editor.** Connect Cursor, Claude Code, Codex, or VS Code over MCP.
- **Get reviewed before you push.** Run ellie review on your local diff, in the terminal.
- **Each commit is re-reviewed.** The score updates with each fix. Approval at 5/5 is optional.

## Benchmark

Entelligence ranks #1 of 8 AI code reviewers on the [2026 AI Code Review Benchmark](https://entelligence.ai/code-review-benchmark-2026.md): F1 47.2%, precision 50.0%, recall 44.8%, 30/67 golden comments found.

## Frequently asked questions

- **What does it read besides the diff?** The code around the change: callers, dependencies and repository structure. It also reads the guidelines you assign, repository Markdown such as AGENTS.md, accepted Learnings, linked repositories and, where connected, your production incidents. That is how a finding can point at a call site in another repo, or at the incident the change would repeat.
- **Which Git providers and languages does it support?** GitHub, GitLab and Bitbucket, including supported self-hosted GitHub and GitLab. Language-aware checks cover Python, JavaScript and TypeScript (JSX and TSX too), Java, Go, C, C++, C#, Ruby, Rust and Dart, and cross-language checks apply beyond that list.
- **Will it flood our pull requests with comments?** Each repository sets its own bar: critical only, medium and above, or everything. You can cap the number of inline comments, switch to a single summary comment, and choose short or detailed explanations. Most teams start at medium and above and adjust after a week of real pull requests.
- **How do we teach it our standards?** Guidelines hold the rules that should always apply. Write them, point at the docs you already keep, or generate them from your review history, then assign them per repository. Learnings handle exceptions: when an engineer marks a finding unhelpful and says why, it becomes a Learning and the same false positive comes up less in later reviews.
- **Can it approve or merge a pull request on its own?** It never merges. If you turn on approval at 5/5, it submits an approval through your Git provider once the confidence score reaches 5 out of 5. Branch protection, required checks and required reviewers still decide the merge, and the setting is off until you turn it on for a repository.
- **Can I run a review before I open the pull request?** Yes. Run ellie review in your terminal to check local changes, or add --prompt-only to hand the findings to Claude Code, Cursor or Codex as the next task. It is the same review engine that comments on your pull requests. [Review local changes](https://docs.entelligence.ai/code-review/cli-review)
- **What happens to our code?** Your code is used only for your organization and never to train models. Secrets are encrypted at rest, and you can deploy in your own cloud with your own model provider keys. [How we handle your code](https://entelligence.ai/security)
- **How do we start, and what does it cost?** Connect one repository, open a pull request and comment @entelligence review. The first review arrives in minutes and there is no credit card to start. Plans are priced by team size and review volume. [See pricing](https://entelligence.ai/pricing)

Pricing: [Pricing](https://entelligence.ai/pricing.md). Docs: https://docs.entelligence.ai/code-review/overview