Pre-merge verification.
Only comments when it has proof.

LogoMesh is a GitHub App that breaks your PRs before production does. It generates adversarial fuzz tests, runs them in an isolated Docker sandbox, and posts findings only when it can reproduce a failure.

Request Beta Access View on GitHub

🏆 1st Place — UC Berkeley RDI AgentBeats Competition
Software Testing Agent Track · Evaluated across 20 complex engineering challenges

How it works

1. PR opened

LogoMesh triggers on pull request events via GitHub webhook. No CI configuration required.

2. Adversarial fuzzing

A Red Agent hunts for vulnerabilities using Monte Carlo Tree Search, generating edge-case inputs targeting the changed code.

3. Sandbox execution

All tests run inside an ephemeral Docker container. Ground-truth pass/fail — no hallucination, no static guessing.

4. Evidence-only comments

LogoMesh posts inline comments only when it can reproduce a failure: input shown, expected output, actual output. No noise.

Why not just use CodeRabbit, Greptile, or Copilot?

Every existing AI code reviewer reads code. LogoMesh executes it. Static readers hallucinate vulnerabilities that don't exist and miss bugs that do. Alert fatigue kills adoption. LogoMesh only speaks when it has a reproducible failing test.

Tool	Execution	False Positive Risk	Workflow Lock-in
LogoMesh	✅ Docker sandbox	Extremely Low (deterministic)	None (any GitHub repo)
CodeRabbit	❌ Static only	Medium	None
Greptile	❌ Static only	Highest	None
Cursor Bugbot	✅ VM-based	Low-Medium	Cursor IDE required

Team

Joshua Hickson

Co-Founder / CEO

Platform engineering, developer experience, technical product management. Led LogoMesh from concept to 1st-place AgentBeats win.

Oleksandr Voievodin

Co-Founder / CTO

AI/ML infrastructure, PyTorch, LangChain, Docker orchestration, Prometheus. Builds the adversarial engine and sandbox environments.

Looking for the competition-era research? View the pre-competition research archive →

Pre-merge verification.Only comments when it has proof.