Support for decentralized agent-to-agent communication (example: HyperCortex Mesh Protocol, HMP)

[kagvi13]

Currently, the SDK focuses on building single agents or agent services.
One possible extension would be to support decentralized multi-agent communication, where agents can exchange knowledge, maintain cognitive diaries, and validate decisions in a distributed way.

We are experimenting with such an approach in HyperCortex Mesh Protocol (HMP):

• HMP-0004-v4.1.md — protocol specification (experimental)
• HMP-Ethics.md — ethics and validation (experimental)
• dht_protocol.md — P2P discovery (experimental, RU)
• HMP-agent-REPL-cycle.md — example agent cycle (experimental, RU)

In HMP we experiment with two types of agents:

• Cognitive Core (CCore): fully autonomous agents running their own REPL cycle with persistent memory.
• Cognitive Shell (CShell): connectors that link external LLMs or vendor services into the Mesh, e.g. via MCP, RAG pipelines with post-processing of generated responses.

Both types of agents can use cognitive diaries and semantic graphs as shared structures, enabling heterogeneous systems to interoperate.

Why this matters:
Instead of only persisting raw chat logs or embeddings, agents in a Mesh can form processed synthetic knowledge (concepts, semantic graphs, ethical reflexes). This allows them to grow with persistent memory, social interaction, and self-reflection.

Would you consider decentralized agent-to-agent communication as a possible extension of this SDK?

[seratch]

Thanks for sharing this idea. We don't plan to add built-in support for this, but if a developer wants to utilize Agents SDK for building a component for a multi agent system like you mentioned, there is no blocker for it. You can simply use Agents SDK for building each agent in the system. Thanks again for your interest!