We build machine learning models that find bugs in code. We can be used proactively as part of the development process by both humans and AI agents to eliminate problematic code before it's submitted for review. We can also do checks at time of code review through our GitHub agent that adds comments to PRs (Pull Requests - essentially just submissions of code), to ensure nothing slips through. We allow developers to enforce their own taste on the code that either the AI or their teams contribute to the codebase by providing Recurse Rules. These are written in markdown and are descriptions of bad patterns that you don't want present in your codebase (e.g. the concept of DRY - do not repeat yourself).
Recurse ML uses machine-learning models to automatically find bugs in code and comment on GitHub pull requests.
- for
- Developers and teams using AI-generated code who want automated bug detection.
- pricing
- freemium · free trial
works withClaude CodeCursorGitHub
Key features
5 features of Recurse ML
- Full codebase understanding — Analyzes interactions across the entire repository to spot upstream and downstream bugs.
- Expert library knowledge — Provides up-to-date documentation links for external and internal libraries.
- Custom Recurse Rules — Allows teams to define markdown rules describing patterns to reject.
- GitHub PR review bot — Posts automated comments on pull requests highlighting detected issues.
- CLI tool rml — Command-line interface for running analyses locally.
Use cases
- Automatically comment on GitHub PRs with bug findings before merge
- Enforce team-specific coding standards via custom markdown rules
- Detect breaking changes and suggest fixes during development
- Analyze SQL queries for hidden bugs and edge cases
- Provide real-time feedback to AI coding assistants like Claude Code
Recurse ML vs alternatives
 Recurse ML |  Claude Code |  OpenAI Codex |  Bugzy AI |  Devin | |
|---|---|---|---|---|---|
| Best for | AI-assisted code review with custom rules | Agentic coding in terminal/IDE | Code generation assistance | Release-time QA | AI-driven code creation |
| Pricing | Freemium | Subscription | Subscription | Subscription | Subscription |
| DevHunt upvotes | 0 | 0 | 0 | 19 | 0 |
| Launched | Jan 2026 | — | — | Mar 2026 | — |
- Recurse ML vs Claude Code: Claude Code is an agentic coding assistant, not a dedicated bug-finding review bot.
- Recurse ML vs OpenAI Codex: OpenAI Codex provides code generation, whereas Recurse focuses on bug detection and PR comments.
- Recurse ML vs Bugzy AI: Bugzy AI offers QA testing for releases, while Recurse targets early bug detection in PRs.
- Recurse ML vs Devin: Devin is an autonomous AI software engineer that writes code, not a review tool like Recurse.
Recurse ML FAQ
How does Recurse analyze my code?+
It runs machine-learning models in real time, then discards all code, keeping no retention.
Does Recurse store my code?+
No, it has a zero-retention policy and does not keep any submitted code.
What languages are supported?+
The site lists support for major frameworks like React, Vue, Angular and SQL, implying broad language coverage.
What are the usage limits?+
The pricing page mentions a free 14-day trial and a one-time $25 Pro plan; specific limits are not detailed.
Is Recurse SOC 2 compliant?+
Recurse is currently pursuing SOC 2 compliance.
Summarized by DevHunt from recurse.ml · Oct 6, 2026. Details may change; check the official site.
.jpg?width=64)
