Agent2Memory (A2M) Protocol

[dibenedetto]

I whould like to take this repo as a good channel to show a problem I recently faced. When working on multiple agents that use different underlying framework, i noticed a lack in the communication chain that i think should be addressed.

LangChain, Agno, n8n, CrewAI, AutoGen — each ships its own memory model. Agents from different frameworks cannot share state, history, or knowledge, even when running inside the same workflow.

A2M is a thin, open protocol that lets any agent framework read and write to a shared memory store through a single REST interface — without modifying existing agents.

I think we should start a discussion on it :)

[aniruddhaadak80]

Shared memory across frameworks is a real gap, especially once teams stop living inside one orchestration stack. The tricky part is that interoperability fails less on the transport and more on the semantics: write scope, consistency rules, identity, retention, and conflict resolution all mean different things across frameworks.nnIf A2M moves forward, I think the strongest first step would be a very small common contract for append, read, provenance, and namespace isolation. If the initial surface stays thin, people can actually test cross-framework adoption before the protocol tries to standardize every memory pattern at once.

[Vect0rdecay]

@dibenedetto it looks like you've only completed 1 of 5 adapters. Do you have plans to finish the rest and if so, do you have a projected timeline?