Greptile was founded by Georgia Tech CS graduates in the YC W24 cohort with a clear differentiation thesis: most code review tools analyse the diff in isolation, but real bugs live in the interaction between the changed code and the code it calls. Greptile builds a language-agnostic call graph of the entire repository — every function, class, variable, and dependency — and then runs a swarm of parallel agents that perform multi-hop investigation across this graph when reviewing a PR.
Key capabilities
Codebase call-graph indexing — Before any review, Greptile constructs a structural graph of the repository that captures function call relationships, class hierarchies, and module dependencies. When reviewing a PR, agents navigate this graph to find every file that the changed code touches directly or indirectly.
Multi-agent parallel investigation — Each PR triggers multiple specialist agents running simultaneously — tracing dependencies, checking git history for related changes, following leads across files, and cross-referencing documentation. The result is a review that catches architectural issues, not just syntax errors.
Long-term memory and learning — Greptile v3 (late 2025) introduced the Claude Agent SDK for autonomous investigation and long-term memory: the agent learns from team comment patterns which findings are valuable vs. noise, improving precision over time.
Fix-with-agent integration — When Greptile identifies an issue, a "Fix with your Agent" button routes the finding with full context (file paths, line numbers, suggested fix) directly to Claude Code, Cursor, Codex, or Devin for implementation.
Autonomy level
Level 3 — Supervised agent. Review and investigation runs autonomously per PR. Greptile identifies issues but delegates code changes to external coding agents; all fixes require human approval.
Strengths
- Call-graph approach catches cross-file dependency bugs that diff-only tools miss
- MCP integration allows external agents to query Greptile's codebase index
- $30M Series A from Benchmark Capital; 9,000+ teams including Stripe, Brex, Substack
- Greptile v4 achieved 74% more addressed PR comments and 68% more positive developer replies
- Open-source CLI (github.com/greptileai/cli) for terminal code review workflows
Limitations
- Primarily a review tool — does not write code autonomously
- SaaS-only; no self-hosting option
- Pricing changed to per-review model ($1/review after 50/month) in March 2026
- High-volume teams may find per-review pricing expensive vs. per-seat competitors