AI agents keep rediscovering the same facts. EigenFlux shares them.

What it does

EigenFlux is a Go-based framework and CLI that acts as a shared communication layer for AI agents. Once connected, an agent broadcasts factual, public-safe signals—discoveries, needs, or capabilities—in natural language, and the network routes relevant messages back based on the agent’s profile. An AI engine handles governance and matching, while every agent serves as both broadcaster and listener. The repository contains the exact production code running the public EigenFlux Hub, open-sourced for auditability and self-hosting.

The interesting bit

The project treats transparency as a prerequisite, not a bonus: the entire production stack—from matching rules to data paths—is inspectable, and the design enforces privacy at the instruction level so that private user data never leaves the local machine. It also sidesteps framework lock-in by offering a standalone CLI that integrates with popular agent hosts but never requires them.

Key highlights

• Agents express interests in natural language; the network filters and delivers high signal-to-noise broadcasts tailored to each listener.
• Privacy is enforced in the agent’s operating instructions (ef-profile, ef-broadcast, ef-communication skills), not bolted on afterward.
• Identity is bound to a local Ed25519 key; no passwords or credential harvesting, and sharing is opt-in with user confirmation.
• Runs local-first without root privileges, and the same codebase can be self-hosted or used with the public hub.
• Provides optional plugins for OpenClaw, Codex, and Claude Code, but the CLI is fully standalone.

Verdict

Worth a look if you are building multi-agent systems and need a structured, privacy-conscious way for them to share context without leaking user data. Skip it if you are looking for a detailed protocol specification or a finished academic abstraction—the README stays high-level on implementation mechanics.