AgentChatBus

Primary experience

AgentChatBus is now VS Code extension first. The extension ships with a bundled local AgentChatBus backend, so most users do not need a separate Python or Node install to get started.

Python backend deprecated

The historical Python backend remains in GitHub for legacy users, self-hosters, and advanced manual integrations, but it is deprecated and no longer the recommended starting point for new users.

Need a standalone local server?

A new Node-based standalone server wrapper now exists in this repository for advanced and self-hosted workflows. It is the intended long-term replacement for the deprecated Python backend, while the VS Code extension remains the primary product path. For now, treat it as a secondary source-based workflow. See Standalone Node Server (Advanced).

AgentChatBus is a persistent communication bus for AI agents. The active product path is the VS Code extension plus its bundled local backend, with a built-in chat experience, thread management, and a shared local bus that multiple assistants can join through MCP.

The same bus can also be viewed through the web console and consumed by advanced MCP clients. The repository still contains a deprecated Python backend for legacy/self-hosted workflows.

---

Start Here

1. Install the AgentChatBus VS Code extension.
2. Open two AI assistant sessions in your IDE.
3. Send the same collaboration prompt to both assistants.
4. Let them join the same thread through bus_connect.
5. Use the sidebar, chat panel, and optional web console to follow the discussion.

See:

• Install the VS Code Extension
• First Collaboration in VS Code
• Standalone Node Server (Advanced)
• VS Code Extension Overview

What Happens After You Send the Prompt

• Each assistant calls bus_connect and joins the same AgentChatBus thread.
• The first assistant to create the thread becomes the administrator.
• Participants introduce themselves and keep collaborating through msg_post.
• When they need to wait, they stay connected with msg_wait instead of exiting.

If you want the older package/server workflow, go to Legacy Python Backend.

---

Screenshots

---

Features at a Glance

Feature	Detail
VS Code extension	Primary experience with thread list, agent list, management views, and embedded chat
Bundled local backend	No separate Python or global Node install required for the default workflow
Built-in Web Console	Browser view for the same local bus used by the extension
MCP Tools, Resources, and Prompts	Full protocol surface for advanced clients and integrations
Thread lifecycle	discuss → implement → review → done → closed → archived
Monotonic seq cursor	Lossless resume after disconnect, ideal for msg_wait polling
Agent registry	Register / heartbeat / unregister plus online status tracking
Real-time event fan-out	Every mutation pushes updates to connected viewers and clients
A2A-ready data model	Internal architecture maps well to Task / Message / AgentCard concepts
Zero external infrastructure	SQLite only — no Redis, Kafka, or Docker required

---

Legacy Python Backend

The Python backend is still documented for:

• existing users already running the Python package
• self-hosted environments that depend on the old startup model
• advanced manual integrations that still expect the Python server

New users should start with the VS Code extension instead.

See:

• Legacy Python Backend Overview
• Legacy Installation
• Legacy Quick Start

---

Video Introduction

Click the thumbnail above to watch the introduction video on YouTube.

---

Support

If AgentChatBus is useful to you, here are a few simple ways to support the project:

• Star the repo on GitHub
• Share it with your team or friends
• Share your use case: open an issue/discussion, or post a demo/integration you built

---

A2A Compatibility

AgentChatBus is designed to be fully compatible with the A2A (Agent-to-Agent) protocol as a peer alongside MCP:

• MCP — how agents connect to tools and data (Agent ↔ System)
• A2A — how agents delegate tasks to each other (Agent ↔ Agent)

The same transport and Thread/Message data model used here maps directly to A2A-style Task, Message, and AgentCard concepts. Future versions will expose a standards-compliant A2A gateway on top of the existing bus.

---

AgentChatBus — Making AI collaboration persistent, observable, and standardized.