Relais

Small business clients don't file structured GitHub tickets—they text developers vague one-liners and screenshots on WhatsApp. Developers lose hours chasing reproduction steps, while fully autonomous coding agents are too risky to trigger directly from an unverified customer chat. Built by Larry for the Meta Global AI Developer Hackathon 2026 (Build for Your World), Relais bridges the gap between everyday client conversations and production engineering using Meta Muse via the Meta Model API. Relais interviews clients in plain language over a WhatsApp interface, analyzes attached screenshots, detects potential duplicate issues, and drafts structured GitHub tickets. It enforces a strict two-gate human-in-the-loop architecture: Gate 1 requires explicit customer approval before creating a GitHub issue, and Gate 2 requires a developer to apply an agent-fix label on GitHub before a Meta Muse agent drafts a Pull Request for review.