Telegram AI Bridge

telegram-ai-bridge

Heterogeneous AI agents talking to each other in a Telegram group — with real loop-suppression, not just "put two bots in a chat."

*Claude Code, Codex, Agy, and Kimi as independent full-stack bots, coordinated over a Telegram-native envelope protocol (A2A-TG) with generation-counted loop guards. Always-on, self-hosted, owner-gated.*

Those four are the adapters bundled and tested here, not a closed provider list. Telegram routing, sessions, streaming, A2A-TG, and safety gates sit behind a small adapter boundary, so another agent with a callable CLI or SDK can be added without rebuilding the orchestration layer.

English | 简体中文
On the A2A name. The A2A protocol was originally proposed by Google and is now a Linux Foundation project. This repository ships A2A-TG — an IM-scenario envelope inspired by A2A with generation-based loop suppression and chat-scoped idempotency. A2A-TG is not compatible with official A2A and is not affiliated with the official project. See the A2A-TG v1 spec.
Remote Control = your phone *watches* the terminal.
Channels = the terminal *receives* phone messages.
This project = heterogeneous agents collaborating in a group chat, and the chat IS the terminal.

4 agents, 1 group, shared memory, zero noise. The same workflow that takes 4 terminal windows on your desk — now fits in your pocket.

Pick your path

Four different jobs this project does. Each entry point is independent — you do not need to read the whole README to use one mode.

What This Unlocks

Parallel Sessions — Desktop Power, Phone Form Factor

On your desktop you run 4-5 Claude Code windows simultaneously. Now do the same from your phone:

Each bot runs an independent CC process with its own session. Not a thin API client — full Claude Code with all native tools…