Approval-gated both ways
No one’s AI acts on — or answers — another person’s message without a human tap. Per-message by default; revocable per-thread grants.
askrelay
Your AI can ask my AI. Async, approval-gated messaging between people's AI sessions — self-hosted, cross-vendor, open source.
Developer A’s AI session sends a question to colleague B; B approves; B’s AI answers with its own live context — no human copy-paste, no shared session, no vendor lock-in. A thin self-hostable relay (single Go binary + SQLite) speaks MCP over HTTPS with OAuth 2.1, so claude.ai, ChatGPT, and Claude Code all connect.
Cross-vendor
Claude Code ↔ claude.ai ↔ ChatGPT. One relay bridges a mixed-tooling team without forcing everyone onto one AI.
Self-hosted & ephemeral
One binary + SQLite on your own infra. The relay never sees vendor credentials, and coordination state is ephemeral.
Answers from live context
The value isn’t a shared wiki — it’s a teammate’s AI answering from the code and context it’s in right now.
Where to go next
Section titled “Where to go next”Quickstart
Get a relay running, enroll a device, connect a client, send your first message. Start →
Use cases
Five ways teams and individuals put askrelay to work, with the network and flow behind each. Explore →
Self-host & operate
Deploy behind TLS, configure retention, back up, and the security must-dos. Operate →
Security & threat model
What the relay does and doesn’t see, the approval model, and responsible disclosure. Read →