asktg

Local desktop search for your Telegram chats. Index selected conversations, search with full-text and semantic retrieval through a built-in GUI, and optionally expose results to LLM agents via an MCP server.

Built with Go, Svelte, and Wails. Windows-first.

Features

• Full-text search — SQLite FTS5 with BM25 ranking and advanced query syntax (AND, OR, NOT, phrase, prefix)
• Hybrid search — combines FTS with HNSW vector search using reciprocal rank fusion (RRF); powered by any OpenAI-compatible embeddings API
• Per-chat policies — fine-grained control over which chats to index, history depth, URL/PDF extraction, and embeddings
• URL & PDF indexing — extracts and indexes text from links and PDF attachments found in messages
• Realtime sync — periodic background sync plus instant updates for new, edited, and deleted messages
• MCP server — read-only HTTP server on localhost with tools for search_messages, get_message, list_chats, and index_status; connect it to Claude or any MCP-compatible LLM agent
• Windows integration — system tray icon, close-to-tray, autostart, single-instance restore
• Backup & restore — full data backup/restore and purge from the UI
• Security — secrets protected with Windows DPAPI; URL fetch guard against SSRF; MCP origin validation (localhost only)

Prerequisites

• Windows 10/11
• Telegram API credentials (API ID and API Hash)
• (Optional) An OpenAI-compatible embeddings API key for semantic search

Installation

Download the latest version from the Releases page. asktg.exe - standalone portable binary — no installation required, just run it

Alternatively, build from source (see Development).

Getting Started

1. Launch asktg.exe. The onboarding wizard will guide you through:

   • Entering your Telegram API ID and API Hash
   • Logging in with your phone number and verification code (2FA supported)
   • Discovering and enabling at least one chat for indexing

2. Search — once indexing completes, use the Search tab to query your messages. Toggle between simple and advanced (FTS5 syntax) modes, filter by chat or date range.

3. Configure chats — in the Chats tab, set per-chat policies:

   • History — off / lazy (new messages only) / full (backfill entire history)
   • URLs — off / lazy / full — controls URL text extraction
   • Embeddings — enable semantic search per chat

4. Enable semantic search (optional) — go to Settings, enter your embeddings API endpoint, model, and API key. Enable embeddings on desired chats.

5. Connect an LLM agent (optional) — enable the MCP server in Settings, copy the endpoint URL, and add it to your agent's MCP config. The server exposes read-only search tools on 127.0.0.1.

Data Directory

Default: %LOCALAPPDATA%\asktg

You can pick a custom data directory during onboarding or later in Settings. Override for the current process:

$env:ASKTG_DATA_DIR = "D:\my-asktg-data"

Contents:

• app.db — SQLite database (messages, chat policies, task queue)
• vectors.graph — HNSW vector index
• telegram/session.json — Telegram session state

Development

Requires Go 1.24+, Node.js 18+, and the Wails CLI.

# Run in dev mode (hot-reload UI)
wails dev

# Run tests
go test ./...

# Production build
wails build

Output binary: build/bin/asktg.exe

Telegram API Credentials for Development

Create tg_secrets.local.json in the project root (gitignored):

{
  "id": "YOUR_APP_ID",
  "hash": "YOUR_APP_HASH"
}

Then generate the encrypted secrets file:

go generate ./...

License

MIT