groupultra/telegram-search

What it solves

Telegram's native search can be limited, particularly for Chinese language users or those struggling to locate specific messages within massive chat histories. This project provides a way to back up personal Telegram messages and search them using advanced techniques like vector search and AI-driven summarization to find information more accurately and quickly.

How it works

The system exports and backs up Telegram chat histories into databases (PGlite or PostgreSQL). It processes these messages by applying tokenization and creating vector embeddings. This allows the tool to perform both keyword-based fuzzy matching and semantic vector searches. It also supports image embeddings for text-to-image search and integrates with LLMs to provide RAG-based (Retrieval-Augmented Generation) Q&A and intelligent summaries of unread messages.

Who it’s for

Users who want a powerful, self-hosted archive of their Telegram conversations with advanced search capabilities, as well as those who want to use AI to summarize their chat history or ask questions based on their own message data.

Highlights

• Hybrid Search: Combines intelligent tokenization, fuzzy matching, and vector semantic search for high precision.
• AI-Powered Features: Includes RAG-based intelligent Q&A and automatic summaries for unread messages.
• Multimodal Search: Supports image embeddings to enable searching for images using text.
• Flexible Deployment: Offers a TypeScript CLI for local-first management, Docker Compose for full-stack deployment, and a browser-based mode.
• Automated Backup: Automatically backs up media resources to MinIO object storage and synchronizes new conversations in real-time.