Greptile
Codebase-aware AI reviewer that catches cross-file bugs other bots miss.
About
Greptile indexes your entire codebase into a semantic graph, then sends a swarm of review agents at each pull request with full architectural context. Its TREX reviewer goes further and writes plus runs tests per PR in a sandbox, and the Learning feature adapts review style from your team's PR comments over time. On its own benchmark it reports an 82% bug catch rate, well ahead of diff-only reviewers, but the cost is noise: roughly 11 false positives per run versus 2 for CodeRabbit in one independent test. Pricing is $30 per seat per month with 50 credits included and $1 per credit after, so high-velocity teams need to model the overage.
What it can do
- Full codebase graph indexing for architectural context
- Multi-agent parallel PR reviews
- TREX reviewer that writes and runs tests per PR
- Learning feature that adapts to team feedback
- Custom review rules with directory scoping
- Confidence scores on comments
- MCP server for external context (Jira, Notion)
- API for triggering reviews programmatically
Setup
- Sign up at greptile.com and connect your GitHub or GitLab repositories.
- Configure review rules and effort tiers (Base 1 credit, Plus 3, Apex 10).
- Reviews run automatically on PR open, push, or rebase; drafts are skipped by default.
- Use @greptileai in a PR thread for on-demand reviews.
Prerequisites
A GitHub or GitLab account; a Greptile account with review credits.
Use cases
Works with
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Solid traction in the AI review space; Series A led by Benchmark in 2026 at a reported $180M valuation.; Actively developed: TREX test-writing reviewer, Learning feature, MCP server, redesigned app.; Closed-source commercial product; no public repository.; Honest caveats: independent testing showed ~11 false positives per run vs 2 for CodeRabbit; credit-based pricing needs cost modeling.. License: Proprietary SaaS; no public source license..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Solid traction in the AI review space; Series A led by Benchmark in 2026 at a reported $180M valuation.
- · Actively developed: TREX test-writing reviewer, Learning feature, MCP server, redesigned app.
- · Closed-source commercial product; no public repository.
- · Honest caveats: independent testing showed ~11 false positives per run vs 2 for CodeRabbit; credit-based pricing needs cost modeling.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.