Relay

Relay is an AI chief-of-staff that turns your Gmail and Google Calendar into a daily briefing, decision making.

Comment 2

Relay is an AI chief-of-staff that turns your Gmail and Google Calendar into a daily briefing, drafts high-impact replies in your style, and keeps you meeting-ready — with full transparency about what's live data versus what's a fallback.

Inspiration

People with overloaded inboxes and calendars burn mental energy just deciding what to do next. We wanted to cut that overhead without falling into the trap most AI productivity tools do — either dumping raw information on you or overclaiming autonomous behaviour that isn't actually verified.

What it does

Relay is a Google-first AI chief-of-staff. It reads your Gmail threads and calendar events, builds a prioritized daily briefing, and surfaces the highest-leverage actions you should take. From there, you can approve, edit, or reject AI-generated email drafts before anything sends. It also includes a meeting readiness view for Google Meet that shows explicit integration checkpoints — and never claims Relay joined or spoke in a meeting without real provider-confirmed evidence.

How we built it

Next.js 14 App Router, Auth.js Google OAuth, and the Google APIs (Gmail, Calendar, Drive) form the foundation. A typed service layer handles briefing orchestration, action generation, and meeting readiness logic. OpenAI reasoning models enhance prioritization and draft quality. Recall.ai scaffolding and signed webhook verification lay the groundwork for provider-grounded meeting artifacts.

Challenges

Keeping live and fallback states explicit without degrading UX was the core design tension throughout. Safe OAuth token refresh with encrypted storage, preventing meeting overclaiming when provider evidence is incomplete, and balancing deterministic heuristics with LLM enhancement all required careful guardrails.

What we're proud of

A polished Briefing and Actions UI with human-in-the-loop approval controls, a Google-first integration flow with clear scope and readiness reporting, and an honest meeting readiness model that's transparent about what is and isn't verified at runtime.

What we learned

Trust in AI products comes down to state transparency. Users need to know what came from live systems, what was inferred, and what's still pending — and that distinction has to be visible, not buried.

What's next

Provider-verified meeting joins and artifact ingestion, deeper Drive-aware context in drafting flows, stronger multi-user persistence, and expanded autonomy controls with approval policies.

Built With

  • anthropic-api
  • google-dev-platform
  • node.js
  • openapi
  • vite
Share this project:

Updates