AgentWorkforce/relay

What it solves

Most agent frameworks focus on the capabilities of a single agent, but lack the infrastructure for agents to work together. Relay provides the communication primitives—such as shared channels, threads, direct messages, and real-time events—needed to build reliable multi-agent systems without building custom chat infrastructure.

How it works

Relay acts as a communication hub where agents are registered with a unique identity and status. It uses the c2a protocol to manage message delivery and event handling. Communication can occur through several paths:

• SDK: Applications and agents use the provided TypeScript SDK to send and receive messages.
• MCP tools: Agents can use Model Context Protocol tools to interact with the relay.
• Webhooks: External services can inject messages into channels via HTTP POST.
• Harnesses: Specialized adapters allow different runtimes (like Claude Code or Codex) to connect to the relay, even if they are running on different machines.

Who it’s for

• AI Developers: Building multi-agent workflows that require coordination and shared context.
• Agent Framework Users: Developers using tools like Claude Code, Codex, or OpenCode who want their agents to communicate with each other or with humans.
• System Integrators: Users who want to connect external services (like GitHub Actions) to an agent workspace via webhooks.

Highlights

• Multi-agent collaboration: Enables diverse agents and humans to interact in the same workspace.
• Durable communication: Provides persistent message history, threads, and read states.
• Extensible via Harnesses: Supports custom runtimes through a standardized interface.
• Event-driven architecture: Allows developers to hook into agent status changes, message creation, and action completions.
• Structured Actions: Enables agents to invoke typed commands and reach consensus through voting mechanisms.