Tetrad

Brings three AI code validators (OpenAI Codex, Google Gemini CLI, and Alibaba Qwen) into Claude's workflow to enforce quadruple consensus before accepting any code changes. Written in Rust with parallel execution, it exposes tetrad_review_code and related tools that automatically score and approve or reject code snippets based on unanimous agreement across all four models (the three external evaluators plus Claude itself). The system maintains a ReasoningBank with SQLite persistence to learn patterns across sessions, caches results in an LRU store, and runs as a stdio JSON-RPC server. You'd reach for this when shipping code where a single AI's judgment isn't enough and you want multiple models catching edge cases, security issues, or logic errors before anything gets written to disk.