← All projects

Multi-vendor AI coding team

Live

Multi-model agent team across Claude, OpenAI Codex, Gemini, and Grok with role separation: Architect, Implementer, Reviewer, Gate, QA, Auditor. Claude Code (via Microsoft Foundry) as executor; the Gate role blocks merge until review passes. An adversarial testing agent on the Jev typed-decision API scores outputs with calibrated probabilities and gates releases on pass/fail thresholds.