AnythingLLM 1.12.0: Telegram channel setup

Timothy Carambat introduces Telegram Channels in AnythingLLM 1.12.0 for Desktop and self-hosted installations. The walkthrough shows how to access a local LLM from a phone while the model runs on a computer at home. Telegram is the only supported channel in this release; Discord and Slack are requests for future support.

Setup starts in Settings under Channels. Carambat creates a Telegram bot through BotFather, then pastes its token into AnythingLLM. He recommends a complex bot username to reduce discoverability. A user's first message triggers a pairing code that needs approval in AnythingLLM before that user can chat with the bot. The desktop computer must remain on to respond.

The demonstration covers streamed replies and slash commands for changing models, workspaces and conversation threads. Carambat sends a photo to a vision model, then asks it to research AnythingLLM and create a PDF. He approves the tool call through Telegram and receives the file on his phone. He says existing agent skills, custom skills, agent flows and MCPs remain available through the channel.

The closing example shows bidirectional chat synchronization, including the photo and generated PDF. Carambat also claims end-to-end privacy, but the tutorial does not establish the security properties of the complete Telegram connection.