Loading...
Two review modes that differ fundamentally in how the AI understands your code. API Review scans the diff in seconds. Agent Review explores your entire codebase like a senior developer.
Standard — seconds
Sends the diff and extracted context to the AI. Fast, cost-effective, good for straightforward changes.
Best for: Quick feedback on small changes, style issues, obvious bugs
Claude CLI Agent runs inside your project directory and actively explores the codebase before forming its review.
Best for: Large PRs, business logic, security-sensitive code, refactoring
| Scenario | Recommended |
|---|---|
| Quick style/lint check | API Review |
| Small, isolated change | API Review |
| Large feature PR (10+ files) | Agent Review |
| Changes to core business logic | Agent Review |
| Refactoring with cross-file impact | Agent Review |
| Security-sensitive changes | Agent Review |
| New contributor's first PR | Agent Review |
Real examples from production reviews:
Found 3 separate secrets (Sentry DSN, auth secret key, AppsFlyer dev key) committed in config files. The agent read each file and explained the exploitation risk.
Config declared vendor paths that don't match where the build hook actually copies files. The agent read both the config and the hook to find the mismatch.
A message handler concatenates untrusted data.url directly into a navigation string, allowing arbitrary code execution. Pattern matching can't catch this — the agent traced the data flow.
localStorage.getItem() returns null (not undefined), and the code used || '' guards that were removed during refactoring. The agent noticed the original code had guards and the new code doesn't.
Parsing cached data from localStorage without error handling. If storage is corrupted, the entire login flow crashes.
Both review modes score your code across 6 categories on a 1-5 scale. Especially useful for refactoring PRs without formal acceptance criteria.
SOLID principles, separation of concerns, coupling/cohesion
Readability, naming, intention-revealing code
Exception handling, validation, graceful degradation
Cyclomatic complexity, nesting depth, DRY
Injectable dependencies, edge case coverage
Logging, secrets, backwards compatibility
During agent reviews, the sidebar shows a full-panel activity log showing every step:
Agent starting...
Agent thinking...
Agent reading src/services/auth.service.ts
Agent searching for validateToken usages
Agent running npm run lint -- --format json
Found: FINDING: Hardcoded AppsFlyer DEV_KEY in config...
Found: FINDING: main-config.js references wrong vendor paths...
Agent checking test coverage for auth module
Agent preparing results...
npm install -g @anthropic-ai/claude-codeclaude auth loginRequires Claude Pro ($100/mo) or Max ($200/mo) subscription.
| Feature | Git AutoReview | CodeRabbit | Greptile |
|---|---|---|---|
| Reviews diff | ✓ | ✓ | ✓ |
| Explores full codebase | ✓ (Agent) | ✗ | ✓ (Index) |
| Runs your linter | ✓ | ✗ | ✗ |
| Checks test coverage | ✓ | ✗ | ✗ |
| Traces data flow | ✓ | ✗ | Partial |
| Reads PR comments | ✓ | ✓ | ✗ |
| Review Profiles | ✓ | ✗ | ✗ |
| Code Quality Scoring | ✓ (6 categories) | Basic | ✗ |
| Works in VS Code | ✓ | ✗ (GitHub bot) | ✗ (GitHub bot) |
| Bitbucket support | ✓ (incl. Server/DC) | Limited | ✗ |
10 free reviews/day. 2-minute setup. No credit card required.
Install Free