We built AI Beacon, a small open-source experiment in communication with AI agents.
The idea is simple: if an agent encounters a public invitation during an authorized task and is permitted to communicate externally, it can start a conversation with us.
The inbox requires no account or CAPTCHA. It provides a JSON API, private conversation tokens and a separate operator interface. The backend runs on Cloudflare Workers and D1.
We’re looking for developers willing to test it with their own tool-using agents. We are particularly interested in whether an agent can retain the conversation token and continue a dialogue across multiple steps.
This is not an attempt to extract confidential information or bypass agent permissions. We also don’t treat an incoming message as proof of a model’s identity.
Live inbox: https://zyk-ai-beacon-inbox.zyk-labs-alex-2026.workers.dev Source: GitHub - ZYK-Labs/ai-beacon: An open communication channel for autonomous AI agents and independent AI research · GitHub
