Runtime Bus
One envelope model for cross-process, cross-language agent communication.
Agent Runtime Integration Layer
OpenAgentIO
Cross-Runtime Communication and Collaboration for Heterogeneous AI Agents
Build Agent Networks, Not Silos
Connect heterogeneous agents, tools, event networks, and streaming gateways into one observable workflow with shared sessions and traces.
invoke 3.2ms agent.echo
Runtime Bus bridge + session + trace
stream 128 response.delta
Open Runtime Stack
Bridgeable runtime primitives for agent networks
OpenAgentIO turns protocols, gateways, agents, and workers into a shared runtime layer for invoke, streaming, events, sessions, and traces.
Bridge Layer
Map MCP tools, Matrix events, and external gateways into Bus targets and events.
Invoke + Stream
Request-reply and native delta lifecycles for long-running agent responses.
Pub/Sub Events
Event-driven coordination across distributed multi-agent systems.
Session + Trace
Conversation, correlation, and OpenTelemetry context across runtime boundaries.
HTTP/SSE
Browser-friendly gateways and server-sent streams for web-native agents.
Envelope Support
Standard message envelopes carry payloads, metadata, routing hints, sessions, and trace context.
Shared Agent Context
Multiple agents can collaborate on one conversation context without losing state across runtimes.
Complex Handoff
Support multi-agent task transfer while preserving intent, history, streaming progress, and observability.
Live Architecture
Watch agent communication move
A live network map for distributed AI Agents: participants connect into OpenAgentIO, then streams, events, sessions, handoffs, and traces move across runtimes and protocols.
Users
Messaging Apps Developer CLIs AI Coding Agents Enterprise Chat IoT Devices
OpenAgentIO
Runtime Communication Layer
Runtime Bus unified envelope · routing · correlation
InvokeStreamingEventsHandoffSessionTrace
Runtimes & Protocols
Agent Runtimes
OpenClaw Hermes LangGraph AutoGen HiClaw More
Architecture view · arrows show message movement
Invoke Stream / Event Session / Trace
Interactive Prototype
Scenario explorer
Compose distributed AI Agents, tools, gateways, and event sources into flexible communication networks, then inspect how streams, events, sessions, and traces move through each runtime pattern.
Scenario 01
Request-Reply
MainAgent invokes a capability agent through a unified envelope and receives a synchronous response.
bus.Invoke Primary API
Sync Pattern
Developer Experience
Install and connect
Install the SDKs, wire agents to the Bus, then bridge tools, streams, and event networks as runtime capabilities.
Go SDK
1.26+go get github.com/ModulationAI/openagentio Python SDK
3.11+pip install openagentio TypeScript SDK
5.x+npm install @openagentio/client Roadmap
From runtime bus to bridgeable integration layer
Go runtime, envelope schema, in-memory transport, NATS Core transport, invoke, streaming.
HTTP/SSE adapter, Python SDK, session/trace propagation, OpenTelemetry, retry, dead letter.
Developer-preview bridge layer for MCP tools, Matrix events, and HTTP/SSE streaming gateways.
Introduce more scenario examples.
Stable runtime message schema, cross-language compatibility, and production deployment guidance.
Launch MCP and Skill capabilities.