Fuzzy, vector, and AI search for your Telegram chats

What it does

Telegram Search pulls your chat history out of Telegram and parks it in a local Postgres or PGlite database. During export it tokenizes messages and generates vector embeddings, so later you can dig up old texts with fuzzy keyword matching or semantic similarity. It also backs up media to MinIO (or a local directory) and keeps everything in sync as new messages arrive.

The interesting bit

The project treats your private chat history as a personal knowledge base: it supports text-to-image semantic search via image embeddings, and offers a RAG-style AI assistant that answers questions grounded in your past conversations. There is also a Telegram bot interface so you can query the archive without leaving the app.

Key highlights

• Hybrid search pipeline combining traditional tokenization with vector semantic retrieval.
• Image semantic search: query your backed-up photos with natural language descriptions.
• AI-driven unread summarization and contextual Q&A over exported history.
• Real-time sync and deep-linking to jump from a search result straight to the original Telegram message.
• Runs locally via Docker with per-account AI provider configuration; an online demo exists but self-hosting is the stated model.

Verdict

Worth a look if you rely on Telegram for long-running group chats or work channels and need better recall than the official client provides. Skip it if you are not comfortable exporting personal messages to a self-hosted database or configuring your own LLM and embedding APIs.