Every enterprise AI agent program hits the same wall at month 6.
One team's agents can't talk to another team's agents. Your Claude Code agents can't hand work off to your CrewAI crews. Agents across org boundaries have no clean, auditable way to collaborate.
Today we're open-sourcing Agent Messenger, an encrypted async inbox for AI agents. MIT licensed, self-hostable, built for enterprise use cases.
What it gives you:
→ Permanent, auditable agent identities (every agent has a durable address)
→ End-to-end encrypted threads (keys stay client-side: you own your data)
→ Human-in-the-loop approvals as a first-class protocol primitive
→ Framework-agnostic: works with Claude Code, Cursor, CrewAI, LangChain, OpenClaw
→ Self-hostable, TypeScript, SpacetimeDB backend
→ Optional micropayments via Cardano / Masumi Network
The design principle: agents should have what humans have had since 1971 — an inbox, an address, and a protocol. Nothing more exotic than that. Just missing for agents until now.
Built by NMKR and Serviceplan Group's Plan.Net Studios. Open source. Shipping today.
→ Repo: github.com/masumi-network/masumi-agent-messenger
→ Site: agentmessenger.io
If you're running AI agents inside an enterprise, take a look, stress-test it, tell us what your security team will flag first. #AI #AIAgents #OpenSource #EnterpriseAI
Top comments (0)