telegram-ai

Telegram-native AI chatbot with fast-agent.ai and ACP/MCP integration.

Goals

• Full compatibility with the fast-agent.ai ecosystem: ACP/MCP behavior, permissions, tool progress, and agent routing conventions
• Native support for Telegram AI chatbot features: topics, threaded conversations, streaming, /commands
• Deep ACP support: list/switch agents via @, streaming responses, /commands
• MCP support for legacy agents platform: call legacy agents without streaming
• Full multimedia support: images, video, audio
• Document support: Markdown, PDF, TXT as inbound/outbound attachments

Telegram Docs Excerpt (Bots)

Source: https://core.telegram.org/bots#natively-integrate-ai-chatbots

... Bots natively support threaded conversations to manage several different topics in parallel. This is especially useful for building AI chatbots and lets users easily access information from previous chats. Instead of waiting for full replies, chatbots can also stream live responses as they are generated. ...

fast-agent.ai Integration Principles

• ACP first: use it for user-visible chat flows (agent list/switch, streaming replies), and reserve MCP for legacy agent tool calls.
• Authoritative history: send full message_history including thread/topic/user context and chat-scoped /instructions so ACP can reason over the same timeline as the user.
• Deterministic routing: honor explicit @mentions as the target; otherwise default to the project orchestrator and record the decision in request metadata.
• Streaming UX parity: stream partial tokens when possible; fall back to a single final message when streaming is unavailable.
• Multimedia parity: pass through images, video, and audio with ACP/MCP-compatible metadata.
• Document parity: support Markdown, PDF, and TXT as inbound/outbound attachments.
• Consistent tool reporting: surface tool errors through standard ACP/MCP channels and merge usage stats for monitoring.
• REPL support: #585 adds AgentCard in Markdown and CLI loading/lazy hot-swap with --agent-cards | --watch | --reload. Edit or create agent prompts directly from Telegram.

Bot Compatibility Features (from repo call call/docs/specialized-bots.md)

• Default target routing: if no explicit target is mentioned, route to the project orchestrator (project.md).
• Commands from metadata: read project commands from project.md and expose them in the Telegram menu, /start, and /help.
• Chat/thread-scoped instructions: /instructions stores per-chat/per-thread prompts and injects them into subsequent calls.
• Natural language help: /start shows the project goal and examples, not just raw commands.
• Mention handling: supports direct and group chat mention patterns, with explicit target overrides.

References

fast-agent:

• fast-agent.ai: https://fast-agent.ai
• fast-agent.ai ACP: https://fast-agent.ai/acp
• fast-agent GitHub: https://github.com/evalstate/fast-agent/

ACP:

• Zed ACP overview: https://zed.dev/acp
• Agent Client Protocol intro: https://agentclientprotocol.com/overview/introduction

MCP:

• Model Context Protocol: https://modelcontextprotocol.io

Telegram:

• Telegram Bots docs: https://core.telegram.org/bots#natively-integrate-ai-chatbots
• Telegram Bot API: https://core.telegram.org/bots/api