Overview
The librarian can be reached from a chat app while an autorag tui session runs on your machine. This page compares the three supported bridges and points at each page's setup.
Messenger connections
| Messenger | Status | Start here |
|---|---|---|
| Telegram | Supported | autorag tui --telegram |
| Supported (unofficial client) | autorag tui --whatsapp | |
| Discord | Supported (no steering yet) | autorag tui --discord |
| Slack, others | Not supported | — |
Each bridge is an extension of the running TUI process: a question you send the bot runs through the same Jev routing and two-phase search as the same text typed in the terminal, and the answer comes back in the chat. If the TUI exits, the bridge stops answering. A chat message sent while a search is running steers that search where the bridge supports it (Telegram, WhatsApp); Discord does not support steering yet.
A bridged session has the same shell and file tools as the terminal, so treat every chat that can drive it as a credential for a shell on your machine, and allow only chats you control. Chat content is untrusted input to the model. A bridged tui defaults to full permission because there is no terminal to answer an approval dialog; see Permission modes.
How the bridges are loaded
| Bridge | Extension | Pinned version |
|---|---|---|
| Telegram | @llblab/pi-telegram | 0.58.3 |
whatsapp-pi | 1.0.70 | |
| Discord | pi-discord-remote | 0.2.4 |
The Telegram bridge loads on demand from the installed package. The WhatsApp extension is installed into Pi's agent directory by autorag tui --whatsapp on first run, then loaded. The Discord extension ships with AutoRAG and is hosted in-process; it creates one text channel per session and the fast answer and the verified final answer are posted as they land.
