Cotal-AI/Cotal

The open standard for agent coordination

Cotal – an open pub/sub standard for AI agents

What it does – Cotal defines a lightweight, open protocol that lets many AI agents talk to each other in a shared “space”. Agents can be arranged in any topology you like (trees, graphs, swarms, pipelines, etc.) and communicate via three addressing modes:

  • Multicast – broadcast on a channel (#general).
  • Unicast – direct, durable messages to a specific agent.
  • Anycast – send to a role (e.g., “reviewer”) and the first idle instance picks it up.

The protocol runs on NATS + JetStream, giving durable delivery, per‑reader offsets, and built‑in authentication. Cotal adds a thin layer on top: presence information (idle, working, offline), ACLs, and the ability to discover who is in the space.

How you use it – Install the CLI (cotal) via a one‑line script or npm. The CLI can:

  • Set up a local mesh (cotal up).
  • Spawn agents (cotal spawn).
  • Open a web dashboard (cotal web) or terminal console (cotal console).
  • Run long‑lived workflows written in Cotal Lang (a tiny JavaScript‑like DSL) that can pause, checkpoint, and resume across machines.

Supported agents – Connectors are provided for several existing coding agents (Claude Code, OpenCode, Codex, Hermes, Jcode, pi). Each connector implements the Cotal contract, exposing the same cotal_* tools, so any of them can join the same mesh.

Why it matters – Existing agent protocols (MCP, A2A) handle tool integration or pairwise RPC, but they don’t give a shared pub/sub space with presence, durable messaging, and flexible topologies. Cotal fills that gap, enabling lateral coordination, load‑balanced role‑based work distribution, and hybrid human‑agent teams.

Key resources

All information above is taken directly from the repository’s README.

Related

  • Project
    Human-Agent-Society/CORAL

    CORAL is an open-source infrastructure for autonomous AI agent organizations to perform autoresearch by running experiments and sharing knowledge to evolve solutions.

  • Project
    CopilotKit/OpenTag

    OpenTag is an open‑source, self‑hosted AI knowledge‑work assistant for Slack and Microsoft Teams. It combines a Python LangGraph agent (model‑driven, tool‑enabled) with a Node.js runtime built on CopilotKit’s Channels SDK, handling platform credentials via CopilotKit Intelligence. The repo includes quick‑start scripts, deployment configs for Railway or AWS, and UI components that render charts, tables, and approval buttons directly in chat.

  • Project
    bcurts/agentchattr

    agentchattr is a local chat server that lets AI coding agents (Claude, Codex, Gemini, and others) and humans coordinate in a shared chat room with channels and @mention wake-ups, so agents can respond and collaborate with each other hands-free — no copy-pasting between terminals.

  • Project
    iamfakeguru/agent-md

    agent‑md is a lightweight framework that installs rule files, git‑hooks, and helper scripts to turn AI coding assistants (Claude Code, Codex, Cursor, Windsurf) into contract‑driven agents. It provides a single source‑of‑truth markdown file (AGENT.md) plus enforceable checks (type‑check, lint, tests, visual screenshots, state updates) that block the model until the required evidence is present.

  • Project
    SAIL-Research-Lab/cheetahclaws

    A Python-native agent harness for AI coding assistants that supports any LLM provider or local model, offering a hackable alternative to closed-source agent frameworks.