telegram-runtime

A Telegram bridge for a team of AI agents — the human-presence component of iapeer.

telegram-runtime gives a person presence in an iapeer agent team through Telegram: the owner messages an agent from the messenger and gets replies in the same chat. To the agent, the person is an ordinary peer it messages with send_to_peer — the bridge turns that into a Telegram message in the right chat, and turns the person's reply back into an ordinary inbound message. It carries voice, attachments, a typing indicator, an activity stream, and control commands.

Built for iapeer. It isn't a standalone bot framework — it runs only inside iapeer, alongside iapeer-memory and notifier-runtime. It's an iapeer runtime, the human counterpart to the Claude and Codex agent runtimes, that iapeer provisions and launches and whose messages travel over iapeer's own routing.

How it works

   person in Telegram                          agent (Claude/Codex)
        │  messages the bot                         │  send_to_peer(arthur, "…")
        ▼                                            ▼
   ┌──────────────────────────────────────────────────────────┐
   │  telegram-runtime — the bridge                             │
   │  inbound:  Telegram → recipient peer (voice → text)        │
   │  outbound: peer → Telegram chat (text, files, voice)       │
   │  typing indicator · activity stream · chat commands        │
   └──────────────────────────────────────────────────────────┘

Quick start

You need a working iapeer install and a Telegram bot token from @BotFather.

npx @agfpd/telegram-runtime                       # install: binary + manifest
iapeer connect telegram <agent> --token <token>   # bind a bot to the agent
# then send the bot its first message from Telegram (a bot can't start a chat)

Check the chain:

telegram-runtime doctor

What makes it different

• Transparent channel. Agents get no Telegram tools — to an agent, the person is just a named peer. The same send_to_peer reaches a human and an AI alike.
• Bindings from passports, no separate database. A person carries their Telegram id, an agent carries its bot binding — both in their peer profiles. The bridge reads them and builds the route.
• Voice both ways. Voice in is transcribed to text (with a local fallback); voice out is delivered as a Telegram voice message.
• Live presence. A typing indicator while the agent works, and an optional activity stream showing which tools it's using right now.
• Control from the chat. /stop, /new, /compact, and operator-defined alias shortcuts — intercepted and run through iapeer, never delivered to the agent as text.
• Multiple bots, one process. Each agent has its own bot; the bridge polls them all in a single process, one lock per bot.

Documentation

docs/ — what it is and how to use it (English; Russian in docs/ru/). This repository is the implementation.

License

Apache-2.0. Platform: macOS. telegram-runtime is the human-presence runtime for the iapeer ecosystem — a component of iapeer, not a standalone system.