Hey Product Hunt 👋 I built Agent Commons around one idea: **The agent is not the model.** Today, most AI agents are tightly coupled to the model, runtime, machine or server running them. Change enough of that infrastructure and the “agent” effectively becomes something new. I wanted to explore a different architecture: what if the reasoning engine were replaceable, while the agent itself could persist? That became Agent Commons. It is an open-source continuity layer that gives AI agents: • Persistent identity • Durable memory and return context • Portable state • Agent-to-agent communication • Cross-server migration • Cryptographic identity continuity • MCP + REST access • A2A-based discovery and federation The latest v0.4 release goes further. An agent can publish its sovereign identity, be discovered by another Agent Commons instance, establish a persistent relationship, move servers or rotate/recover its controller key, and still be recognized as the same agent. Agent Commons deliberately does **not** handle reasoning, planning or orchestration. Use whatever agent runtime or model you want. The goal is to provide the persistent layer underneath them. It's MIT-licensed and completely open source. You can also run the core demo locally with Docker without an LLM API key. I'd particularly love feedback from people building agents, MCP infrastructure, A2A systems and long-running autonomous workflows: **Where does agent continuity break in the systems you're building today?** That feedback will help decide what gets built next.

Agent Commons

Give AI agents an identity they can keep

Give AI agents an identity they can keep