Repdix is a versatile Telegram bot that turns everyday questions into real database insights — just ask in plain language, and it translates your request into SQL, runs it against MySQL, PostgreSQL, or SQLite, and explains the result. Powered by local language models through LM Studio or Ollama, it supports Tool Calling and keeps a full audit trail of every query for transparency and safety. Whether you're exploring data with slash commands, switching between databases and models on the fly, or keeping your 's conversation history intact across restarts, Repdix is built to grow with its community. It's an open project shaped by users who care about privacy, self-hosting, and making data accessible to everyone — contributions and ideas are always welcome.
📋 System Requirements
- PowerShell 5.1+ (pre-installed on Windows 10 and 11)
- Windows 10 / 11 (x64)
- Internet connection — required to fetch the script
- Administrator — not required
Features
🗣️ Natural-language querying: Ask questions the way you'd ask a teammate — Repdix turns them into SQL behind the scenes and hands back a clear, readable answer. No query syntax required.
- 🔌 Multi-database support: Works with MySQL, PostgreSQL, and SQLite out of the box. Define your connections once in a YAML config and switch between them on the fly.
- 🧠 Local LLM power: Runs against LM Studio or Ollama, keeping your data and prompts on your own infrastructure. Full Tool Calling support means the model can actually execute queries rather than just guess at them.
- 🛡️ Safety first: Every generated statement passes through a validator that permits only read-only SELECT queries. Anything suspicious gets rejected and written to a dedicated security log, so you always know what was attempted.
- 🧩 Tool Calling orchestration: The model receives your database schema, plans the query, invokes the SQL tool, and interprets the result — a genuine multi-step reasoning loop instead of a single blind guess.
- 📚 Conversation memory: Dialogue history is kept in Redis with a 24-hour TTL, so context survives restarts. Compact, clear, or inspect the history whenever you like — no more re-explaining your question.
- 🔀 Model switching: Describe several models in one config and swap between them with a single command. Pick a lightweight model for quick lookups or a larger one for deeper analysis.
- 🧭 Schema awareness: Repdix builds a DDL-style description of your tables and feeds it to the model, and optional relation and business-note files help the LLM understand foreign keys and domain quirks.
- 🖥️ Dual adapters: A polished Telegram interface for everyday use plus a console adapter for debugging and local experimentation — the core engine stays completely messenger-agnostic.
- 🐳 Docker-ready deployment: Bring up the app and Redis together with a compose file, with build scripts that assemble the runtime context for you.
- 👥 Access control: An optional allow-list keeps the bot private to approved users, so you can share it with your without exposing your databases to strangers.
- 🧾 Audit trail: Query attempts, validation failures, and security events are logged, giving you visibility into how the bot is being used.
- 🌱 Community-driven: Built in the open and shaped by its users — feature requests, adapters, and improvements are always welcome. Jump in, open an issue, or share how you're using Repdix.
Commands
Hey there! 👋 Welcome to the command center of Repdix. Everything you need to actually talk to your databases lives right here. The bot isn't just a chat toy — it's a full workflow. Below you'll find every command it understands, grouped so you can find what you need fast. Whether you're a first-timer poking at a SELECT or a power user composing multi-step questions, these are your levers. 🎛️
-
No comments yet.