Greptile
by Greptile (Tabnam Inc.)
AI code review agent that indexes the whole codebase to catch cross-file bugs before merge
Greptile is an AI code review agent that builds a dependency graph of an entire repository and then runs swarm agents over every pull request to flag logic bugs, security risks and standards violations that span multiple files. It is aimed at engineering teams whose review queue has become the merge bottleneck now that AI coding assistants generate far more pull requests than humans can read carefully.
Greptile is an AI code review agent from Y Combinator W24 company Tabnam Inc., founded in 2023 by Daksh Gupta, Soohoon Choi and Vaishant Kameswaran and based in San Francisco. Unlike diff-scoped reviewers, Greptile first constructs a graph index of the full repository, then dispatches parallel swarm agents against each pull request to evaluate both the change and its downstream impact across files the diff never touches. It posts findings as pull-request comments, roughly a third of them as one-click GitHub suggestion blocks, and can hand issues off to Claude Code, Cursor, Codex or Devin for the fix. Teams encode their own conventions as custom rules written in plain English, and the agent adapts over time by analysing which comments developers act on versus dismiss. A feature called TREX autonomously writes and runs tests in a sandbox so a suspected defect arrives with a reproduction rather than a static assertion. Greptile Agent v4 shipped 5 March 2026 with published A/B results: addressed comments per pull request rose 74% from 0.92 to 1.60 and the comment addressing rate rose 44% from 30% to 43%. The same announcement moved pricing to a base-plus-usage model, which provoked significant public criticism. Greptile raised a $25M Series A led by Benchmark in September 2025 after a $4M round in June 2024, states 22,000+ teams on the platform including Substack, Klaviyo, Retool, NVIDIA, Brex, Scale, PostHog, Zapier and WorkOS, and offers SOC 2, SSO, audit logs and self-hosted air-gapped deployment inside a customer's own AWS account for defense, healthcare and financial services buyers.
A VP of Engineering or platform lead at a 50-500 developer organisation where AI coding agents have multiplied pull-request volume and human review has become the merge bottleneck.
Cross-file bugs are caught on the pull request instead of in production, with review comments developers actually act on — Greptile's own A/B test put addressed comments at 1.60 per PR after v4, up from 0.92.
At a Glance
- Category
- Developer Tools
- Pricing
- Subscription, Usage-based, Freemium
- Target Market
- CTOs, VP Engineering, Enterprise Developers, Platform Engineering Teams
- Deployment
- Cloud-first, Self-hosted, API-based
- Founded
- 2023
- Headquarters
- San Francisco, United States
- Team Size
- 11-50
- Customers
- 22,000+ teams (vendor-stated); named customers include Substack, Klaviyo, Retool, NVIDIA, Brex, Scale, PostHog, Mintlify, Zapier and WorkOS
Key Features
- ✓Full-codebase graph index
Builds a dependency graph of the whole repository so reviews reason about code the diff never touches.
- ✓Swarm agent pull-request review
Runs multiple agents in parallel over each pull request to assess the change and its downstream impact.
- ✓Custom rules in plain English
Teams write engineering standards as natural-language rules that Greptile then enforces on every subsequent pull request.
- ✓TREX autonomous testing
Writes and executes tests inside a sandbox to confirm a suspected bug is real before commenting.
- ✓Learning from review history
Analyses which comments developers address or dismiss and adapts to that team's tolerance over time.
- ✓Coding-agent and IDE handoff
Routes findings directly into Claude Code, Cursor, Codex and Devin so remediation starts where the developer works.
- ✓Self-hosted air-gapped deployment
Enterprise customers run Greptile inside their own AWS account with SSO, audit logs and SOC 2 controls.
Capabilities
Use Cases
- •Gating AI agent output
An organisation running Cursor and Claude Code routes every agent-authored pull request through Greptile so a machine reviews machine-written code before a human signs off.
- •Cross-file regression catching
A payments team edits a shared helper and Greptile flags the unrelated call sites in other modules that the diff silently breaks.
- •Standards enforcement without a review rota
A platform team encodes logging, error-handling and migration conventions as plain-English rules rather than repeating the same review comments every sprint.
- •Regulated self-hosted review
A defense or healthcare contractor deploys Greptile into an air-gapped AWS environment so proprietary source code never crosses the network boundary.
- •Bug triage with a reproduction attached
TREX writes a failing test in a sandbox so the review comment arrives with proof rather than a speculative warning a developer must verify.
Ideal For
Best For
- ✓Reviewing AI-generated pull requests at volume, where human reviewers cannot keep pace with agent output
- ✓Catching multi-file and cross-module logic bugs that diff-only reviewers miss because they never see the rest of the codebase
- ✓Enforcing team-specific engineering conventions written in plain English as custom rules instead of repeating them in review comments
- ✓Regulated engineering organisations that need code review running self-hosted inside an air-gapped AWS environment
- ✓Handing verified findings straight to a coding agent such as Claude Code, Cursor, Codex or Devin to apply the fix
Not Ideal For
- ✗High-throughput teams on the Pro plan — the 50-review monthly allowance is consumed quickly and overage runs $1 per review, which is exactly what triggered the April 2026 public backlash
- ✗Teams that want one vendor for both authoring and reviewing code; Greptile deliberately does not generate features, arguing a reviewer that writes the code cannot be an independent check
- ✗Teams already sensitive to AI reviewer noise — independent tests disagree sharply on its false-positive rate, from zero across 120 findings in one three-week trial to 11 against a competitor's 2 in another
Integrations
Deployment
Market & Ratings
22,000+ teams (vendor-stated); named customers include Substack, Klaviyo, Retool, NVIDIA, Brex, Scale, PostHog, Mintlify, Zapier and WorkOS
Market Analysis
Pros
- ✓Whole-codebase indexing catches cross-file and architectural bugs that diff-scoped reviewers structurally cannot see
- ✓Posted the lowest finding density of four reviewers in an independent 146-PR, 3.5-week trial — 2.2 findings per pull request, roughly 92% of them substantive bugs rather than naming nits
- ✓About 32.5% of findings arrive as one-click GitHub suggestion blocks, so fixes apply without leaving the pull request
- ✓Self-hosted air-gapped deployment plus SOC 2, SSO and audit logs make it viable for defense, healthcare and financial services buyers
- ✓Publishes measured v4 improvements rather than asserting them: addressed comments per PR up 74%, comment addressing rate up 44%
Cons
- ✗The March 2026 shift to $1-per-review overage drew sustained criticism, including a Hacker News post titled 'Greptile's New Pricing Is Predatory' and a dedicated rebuttal site arguing the 50-review allowance sits below what a normal team already ships
- ✗Open-source maintainers reported being billed despite the advertised free-for-OSS policy, costing Greptile community goodwill in the same episode
- ✗False-positive rate is benchmark-dependent and genuinely disputed — one comparison recorded 11 false positives against CodeRabbit's 2, while a separate three-week trial recorded zero across 120 findings
- ✗The GitHub check-run integration intermittently failed to register during that independent trial, forcing a manual audit workaround
- ✗Auto-generated sequence diagrams were found inaccurate or unhelpful often enough that the same reviewers recommended disabling the feature
- ✗Does not generate code by design, so teams wanting a single vendor for authoring and review need a second tool
Pricing
Starter
$0
- ✓Unlimited repositories
- ✓50 credits per month
- ✓One active developer
- ✓1 credit = 1 standard review, 3 credits = 1 TREX review
Pro
From $30/seat/mo
- ✓50 credits per seat per month
- ✓$1 per additional credit
- ✓Unlimited repositories and users
- ✓Custom rules
- ✓Unlimited app integrations
- ✓14-day free trial
Enterprise
Contact for pricing
- ✓Self-hosting
- ✓SSO/SAML
- ✓GitHub Enterprise support
- ✓Dedicated Slack support
- ✓Custom DPA
Greptile moved to a base-plus-usage model on 5 March 2026 alongside Agent v4: $30 per developer per month includes 50 reviews, then $1 per additional review. A seat is any developer who received a review in that billing period, so cost tracks activity rather than headcount. Greptile said fewer than 10% of active users would exceed the allowance; critics on Hacker News showed that stacked-PR and agent-driven workflows blow through it in days, with one developer projecting $500+ from 571 pull requests in 30 days. Enterprise self-hosting is quote-only. Open-source projects under MIT, Apache or GPL licences get free reviews and pre-Series A startups under $2M revenue get 50% off, though both concessions are documented in footers rather than advertised.
Security & Compliance
Connect
Sources
This page was written from 8 sources, 5 on domains other than greptile.com.
- 1.greptile.com — greptile.comvendor
- 2.greptile.com — pricingvendor
- 3.greptile.com — greptile v4vendor
- 4.ycombinator.com — greptile
- 5.news.ycombinator.com — item
- 6.dev.to — best ai code reviewer in 2026 we ran 4 in parallel for 3 wee
- 7.thesaasnews.com — greptile raises 25 million series a
- 8.github.com — greptileai
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
GitLens
The Git workbench where human and AI-agent changes get reviewed, composed and merged in one view
OpenHands
Open-source, self-hosted control center for running coding agents on real engineering work
Confident AI
Hosted LLM evaluation, observability and red teaming built on the open-source DeepEval framework
Respan
Observability, evals and an LLM gateway for AI agents in one control plane