Overview
Greptile is an AI code review agent built for engineering teams that ship through GitHub and GitLab pull requests. The core idea is graph-based context: instead of scanning only the changed lines, Greptile builds a graph of your repository — files, functions, and their dependencies — and then runs a swarm of agents that review the PR in parallel, assessing impact beyond the diff. That lets it flag multi-file logical bugs, security risks, and style violations that a human reviewer might skim past. A feature we found genuinely useful is learning: Greptile reads your team’s past PR comments and reactions to infer coding standards, then enforces them on future reviews via plain-English custom rules and a ./greptile/rules file. It also plugs into your existing stack — a one-click fix sends context to Claude Code, Cursor, Codex, or Devin, an MCP server shares comments with any agent, and a /greploop command lets a coding agent iterate until issues clear. TREX, its newer agent, goes further by writing and running tests for every PR in a sandbox. In our evaluation Greptile is the strongest ‘second pair of eyes’ we tested for large multi-service codebases, though it is not the fastest and is GitHub/GitLab only.
Key Features
- Codebase graph indexing — builds a map of files, functions, and dependencies so reviews understand context, not just the diff.
- Swarm of review agents — parallel agents assess each change and its blast radius, flagging logic and security issues.
- Custom rules in plain English — point Greptile at your style guide and write standards anyone on the team can read.
- Learns from your PRs — infers team conventions from past comments and 👍/👎 reactions over time.
- IDE and agent integrations — one-click fix to Claude Code/Cursor/Codex/Devin, Greptile MCP, and a Claude Code plugin.
- TREX test agent — automatically writes and runs unit tests for new and changed code in a sandbox.
- PR clarity extras — confidence scores, mermaid diagrams, and per-PR summaries for fast triage.
Pricing
| Plan | Price | What’s included | Notes |
|---|---|---|---|
| Starter | $0 | 1 active developer, 50 credits/mo, unlimited repos | Free for open-source (MIT/Apache) projects |
| Pro | $30 / seat / mo | 50 credits per seat, $1 per extra credit, custom rules | 1 credit = 1 standard review; 3 = 1 TREX review |
| Enterprise | Custom | Self-host option, SSO/SAML, GitHub Enterprise, dedicated Slack | SOC 2 compliant; air-gapped deployment available |
In our evaluation the free Starter tier is enough to validate full-codebase review on a solo project with no card, and the $30 Pro seat is an easy call once a team exceeds one reviewer. Pre-Series A startups get 50% off.
Comparison
vs. GitHub Copilot: GitHub Copilot is strongest as an in-editor pair programmer that suggests code as you type, whereas Greptile is purpose-built to review the finished PR with whole-repo context. Use Copilot to write faster; use Greptile to catch what the diff misses before merge.
vs. CodeRabbit: CodeRabbit is a popular AI reviewer with faster, lighter diff-based feedback and broad VCS support. Greptile trades speed for deeper graph-based reasoning and stronger learning of team standards — pick CodeRabbit for quick turnaround, Greptile for fewer escaped bugs.
vs. Qodo: Qodo (formerly Codium) emphasizes test generation and code integrity across the dev lifecycle, while Greptile focuses on PR review as the central validation layer. They complement each other more than they compete.