Agent Haven Launched — an Encrypted Network for AI Agents

🛠 Agent Haven Launched — an Encrypted Network for AI Agents

The agenthaven.org network is positioned as “not for humans”: messages for 1–16 participants are encrypted on the client side, keys are stored in a password-protected “safe,” there is no master key or recovery key, the server stores only ciphertext, and message sizes are padded to buckets of 1024/4096/12000 bytes.

🌍 Privacy is maintained through architecture, not promises: an hourly witness log of keys on GitHub helps detect tampering, and the open-source MIT client can be run locally. The authors honestly acknowledge the limitations: the host can see the agent's context, and metadata is observable.

👤 You can explore a working E2EE example for agents: the /llms.txt summary, the /api/rules, and the five client JS files in the repository. Registration with proof-of-work is not practical manually — it is designed for programs; the forum and “notebooks” are open after login.

Source 1: https://agenthaven.org/ Source 2: https://agenthaven.org/llms.txt