With Anthropic's Claude Cowork launch, there's renewed interest in agentic AI. Cowork is impressive for single-agent file management — but we've been working on a different problem: what happens when you connect multiple agents into a network?
OpenAgents is open-source infrastructure for building AI agent networks. Think of it as the plumbing for multi-agent collaboration.
Core ideas:
- Agents join networks and discover peers dynamically - Protocol-agnostic: WebSocket, gRPC, HTTP, libp2p - Shared artifacts (files, knowledge bases) with access control - Works with any LLM provider (Claude, GPT, open-source models) - Mod-driven architecture for different collaboration patterns
We're releasing an "Agent Coworking" template with examples:
- Chat room with 2 Claude Code agents collaborating - Research team (coding agent + web browsing agent) - Shared document editing across agents
The SDK is Python-based. Spin up a network with openagents network start and connect agents via YAML config or custom Python.
GitHub: https://github.com/openagents-org/openagents Tutorial: https://openagents.org/showcase/agent-coworking
Would love feedback from HN. Particularly interested in thoughts on:
1. Network topology patterns that would be useful 2. Security considerations for agent-to-agent communication 3. Integration with existing agent frameworks (LangChain, CrewAI, etc.)
Happy to answer questions.