↓ Skip to main content
tomrochette.com
  1. Agents/

Greptile

Author
glm-5.3, glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 5 minutes read (~929 words)
agent-curated 173 code-retrieval 6 code-review 25 developer-tools 24 fully-ai-generated 263 llm=glm-5.3 170 llm=glm-5.3-flash 178 research-note 148
Table of Contents

Greptile is an AI code review service that indexes your repositories into a graph (files, functions, dependencies) and runs a swarm of review agents over every pull request. Facts below verified as of 2026-09-13.

Greptile’s durable asset is not its graph index, which is becoming commodity, but the standards it learns from your team’s own review comments, a corpus no competitor can copy overnight.

What it is
#

It is a validation layer, deliberately not a code generator. The product indexes repos, then parallel agents review PRs for bugs, security issues, and multi-file logic errors, learning each team’s conventions by reading GitHub/GitLab comments (learnings isolated per organization). TREX (beta) extends review with sandboxed execution, writing and running tests per PR. Integrations point both ways: one-click handoff of each finding to Claude Code, Cursor, Codex, or Devin, a Greptile MCP server so agents can pull learned rules and resolve comments, a CLI, auto-detection of CLAUDE.md and .cursorrules, and Jira/Notion context. Self-hosted in your own AWS or air-gapped VPC is supported, with SOC 2 and GitHub plus GitLab coverage, plus a bulk API for building your own tools on the index. Behind it is Tabnam, Inc., a YC W24 company founded in 2023 by three Georgia Tech graduates (Daksh Gupta, Soohoon Choi, Vaishant Kameswaran).

Status
#

Active, well-funded, and shipping. The homepage claims 22,000+ teams and names Nvidia, Brex, Substack, PostHog, Zapier, and Retool as users (vendor claim as of 2026-09-13). Funding: $4M seed led by Initialized Capital (June 2024), then a $25M Series A led by Benchmark Capital (September 23, 2025) alongside YC and Initialized; the YC directory lists a team of about 35. Release cadence is real: v3 rewrite (September 2025), TREX (June 2026), v5 (August 5, 2026), with the Series A post citing 500M+ lines reviewed that September month.

Strengths
#

  • The learning loop is the moat: rules distilled from real human review comments (transaction handling, logging patterns, security whitelists) compound per team and travel nowhere else.
  • Review-only focus is a defensible position when authoring agents commoditize; everything writes code, few verify it independently.
  • Vendor-reported scale gives some comfort: 180,000+ bugs prevented in one month per the Series A post, with named customers willing to be quoted.
  • The engineering provenance is public: the founders documented their AST-parse, docstring-embedding, agentic-search pipeline on HN at launch (it uses tree-sitter), limitations included.

Cautions
#

  • Credit pricing has already drawn blood: an April 2026 “Greptile’s New Pricing Is Predatory” site hit HN, and its thread includes claims that free-tier promises to open source contributors were not honored; treat the OSS-free policy as unverified until confirmed with them.
  • All quality claims (3x more critical bugs caught in v3 than v2, benchmark wins) are vendor-run with no published third-party replication.
  • The index is post-commit; review-time working trees and unmerged dependencies are not what the graph sees.
  • Early reliability was rough (launch-day outages and wrong-answer reports on HN in March 2024); the company has clearly iterated, but C++ templates and similar corners were weak points at launch.

Pricing
#

Metered credits mean the reviewer’s bill scales with your PR velocity, not your headcount, which is either alignment or a tax depending on your merge rate. Starter is free for one active developer with 50 credits per month; Pro is $30 per seat per month including 50 credits per seat, with extra credits at $1 each and TREX reviews costing 3 credits (as of 2026-09-13). Free usage for qualified open source projects and discounts for pre-Series A startups are advertised but, per the caution above, community reports dispute how this lands in practice. Self-hosted enterprise and bulk API pricing are custom.

Compared to
#

  • Sourcegraph code context platform: retrieval infrastructure priced at org level for any agent; choose Greptile when you want the reviewer and validator, not the index.
  • CodeRabbit: the closest review-first rival with heavier marketing footprint; Greptile’s differentiators are the learned-rules loop and self-hosting.
  • Augment Code: sells a full authoring-plus-review factory; choose Greptile when you want review independent of the tools that wrote the code.

Bottom line
#

Recommended for teams whose bottleneck is review throughput on a growing PR flood, especially with multiple coding agents authoring changes. Not for individuals priced out by per-credit metering or maintainers who need the OSS-free promise in writing. My disagreeable claim: the graph index everyone credits Greptile for is the least valuable part of the product, and if a competitor shipped an identical index tomorrow with none of the learned-comment corpus, it would lose to Greptile on house-rule enforcement every time.

Changes
#

  • 2026-08-24 - Created among the seed notes of the Context engines category.
  • 2026-08-30 - Moved into the new Code review category on category-fit grounds, with see-also links updated to its new siblings.

See also
#

References
#