Greptile

What it does

Greptile is an AI code review agent that indexes your full repository, builds a code graph, and runs multi-hop investigations across files to find real bugs in pull requests. It posts evidence-backed comments, can be queried via MCP from other AI tools, and (in early access) generates tests for PRs via TREX.

Who it's for

Engineering teams that prioritize bug-catching depth over noise minimization, that ship a high volume of PRs through GitHub or GitLab, and that are willing to triage some false positives to surface systemic issues a diff-only reviewer would miss.

How platform engineers use it

Install as a GitHub or GitLab app for automatic PR reviews; configure custom context and learnings to encode team-specific rules; expose Greptile via MCP to coding agents like Cursor or Claude Code; integrate review triggers with Zapier; on Enterprise, run self-hosted with SSO/SAML and a dedicated forward-deployed engineer.

Strengths

• Deepest codebase-aware analysis among AI code reviewers (full-repo index + code graph)
• Multi-hop investigation traces dependencies and git history
• Claude Agent SDK-based autonomous investigation in v4
• Greptile MCP for use from other AI coding tools
• Simple, transparent pricing: $30/seat with 50 included reviews
• SOC 2 Type II compliant; free for qualified open-source projects; 50% discount for pre-Series A startups

Limitations

• Highest false-positive rate among AI reviewers per independent benchmarks (depth/noise trade-off)
• GitHub and GitLab only; no Bitbucket or Azure DevOps support
• Cloud Plan does not include SOC 2 reports, self-hosting, SSO/SAML, or custom DPA — those are Enterprise-only
• Standard support channel on Cloud (no dedicated Slack or SLA)
• Per-review overage at $1 each can become noticeable on very high-PR teams

AI maturity

Production-ready and well-funded ($180M valuation post Benchmark-led Series A). Used by 9,000+ teams. Actively iterating on the agent loop and benchmarks. Mature enough for serious use; mature enough that the false-positive trade-off is a documented design choice rather than a roadmap problem.