Back to directory

OpenAgentIO

A lightweight runtime bus that bridges MCP servers, Matrix rooms and HTTP/SSE endpoints into one set of Invoke, StreamInvoke and Publish/Subscribe flows, giving heterogeneous agents cross-process, cross-language collaboration with session and trace propagation.

OpenAgentIO

OpenAgentIO is a runtime integration layer for distributed, multi-agent AI systems built from heterogeneous workers, tools and platforms. Rather than replacing an agent framework's prompting or orchestration logic, it normalizes how separate components talk to each other underneath: it translates external protocols into structured Invoke, StreamInvoke and Publish/Subscribe flows on one bus model, with session propagation and OpenTelemetry tracing context carried across every hop.

Bridges are provided for MCP servers, exposing their tools as native bus Invoke targets; for the Matrix chat protocol, turning room messages into publish/subscribe bus events and routing agent output back into the room; and for OpenAI-compatible HTTP/SSE streaming endpoints, wrapped into unified StreamInvoke lifecycles. The project positions itself explicitly as not another agent framework like LangGraph or CrewAI, but as the communication and observability substrate underneath such frameworks.

It is an early developer preview; the bridge layer and multi-language SDKs are still under active development.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.