Would you let an AI agent send emails on your behalf?
Most people say "no" - too risky.
But what if you could set rules like: • "Only send to people in my contacts" • "Never spend over $50" • "Only book meetings between 2-4pm" • "Block domains: competitor.com"
That's what I built.
Introducing PIA (Personal Identity Agent) - an authorization platform for AI agents.
It's like IAM for your life: - Define policies (what agents can/can't do) - Agents authenticate via OAuth flow - LLM verifies every action against your rules - Full audit trail
Built an SDK and 5 production agents to prove it: Email • Calendar • Contacts • Files • Tasks
All live. All production-ready. All at: digital-embodiments.vercel.app
The SDK is on NPM: npm install @variant96/pia-sdk
Tech: Next.js, TypeScript, Prisma, NextAuth, Groq/Gemini
This is my bet on how we'll safely delegate to AI agents.
Thoughts? What would make this more useful?