Happen to be a regular user of telegram and part of some large academic group chats with a lot of discussion, links, and attachments thrown around. Many of them get overwhelming with the amount of raw content and messages there are every day, and keeping track and up to date is a daunting task.
I was inspired by Karpathy's AI post about building a second brain/knowledge base from a raw repo of unrefined data, and I thought I'd take a crack at it with the most dense data source I had that I actually cared about, and that was Telegram chats/channels.
TeleBrain is a desktop app that connects to your Telegram account, imports a chat's history into a local SQLite DB, and uses an LLM to produce two things:
1) a wiki — each recurring topic becomes an article-style page, and every claim carries an inline citation back to the exact source message
2) Ask — natural-language Q&A across the whole history, with clickable citations that jump to the original message in Telegram
Data stays private. Synthesis does send messages to an LLM. Gemini in the trial, and if you buy the license, you can use openrouter to send it wherever you want.
You can see a very quick demo here:
https://x.com/brothaAkhee/status/2077429337528074348
Would love to get any feedback or suggestions!