- L2: Embedding recall for stale APIs and deprecated patterns
- L3: LLM deep scan for context coherence
Key difference from traditional linters: it's built specifically for AI code failure modes — hallucinated packages, context window breaks, over-engineering patterns. ESLint/SonarQube don't even look for these.
100% local (Ollama), self-hostable, free forever. GitHub Actions + GitLab CI integration included.
Would love feedback on what other AI code patterns you've seen that we should detect!
Hi HN! I built this because I kept finding phantom npm packages in Copilot/Cursor output that passed ESLint just fine.
What it does:
- L1: AST-based structural analysis (hallucinated imports, logic gaps, security anti-patterns) — runs in <10s
- L2: Embedding recall for stale APIs and deprecated patterns
- L3: LLM deep scan for context coherence
Key difference from traditional linters: it's built specifically for AI code failure modes — hallucinated packages, context window breaks, over-engineering patterns. ESLint/SonarQube don't even look for these.
100% local (Ollama), self-hostable, free forever. GitHub Actions + GitLab CI integration included.
Would love feedback on what other AI code patterns you've seen that we should detect!