DEV Community

Cover image for Username Studio — find short Telegram usernames with a local LLM
sattop 02
sattop 02

Posted on

Username Studio — find short Telegram usernames with a local LLM

If you've ever tried to find a short, clean Telegram username —
you know how painful it is. Everything good is taken.
You end up typing random combinations for 20 minutes.

I built a local tool to fix that.

What is Username Studio?

Username Studio generates, scores, and checks short Telegram
usernames (5–6 chars) using a local LLM. Everything runs on
your machine — no cloud, no subscriptions.

How it works

  1. Generate — LM Studio (local LLM) creates username candidates in three styles: brandable, russian transliteration, multilingual
  2. Score — each username gets rated on readability, brandability, meaning, and rarity
  3. Filter — only valid 5–6 char lowercase latin usernames pass
  4. Check — optionally verify availability via Telegram API (Telethon), with dry-run mode for safety
  5. Track — everything saved to local SQLite with full status history

Stack

  • Python 3.10+
  • LM Studio — local OpenAI-compatible LLM API
  • Telethon — Telegram MTProto client
  • SQLite — local storage
  • Flask — local web dashboard

Web dashboard

The main interface is a local browser dashboard at
http://127.0.0.1:8080. You can browse batches, see scores,
pick candidates and run Telegram checks — all without touching
the terminal.

CLI mode is also available for terminal lovers.

Safety first

Telegram actions are intentionally explicit:

  • --no-telegram flag disables all Telegram connections
  • dry-run mode previews actions without real requests
  • channel creation requires a typed CHECK confirmation
  • FloodWait handling respects Telegram rate limits

Quick start (Windows)

git clone https://github.com/sattop/username-studio
cd username-studio
.\START.bat

START.bat handles everything: creates venv, installs deps,
copies .env.example.env, launches the app.

No LM Studio? No problem — the project has fallback generation
and scoring built in.

What's next

  • Linux / macOS support
  • More generation styles
  • Configurable username length filter
  • Export to CSV

Links

GitHub: https://github.com/sattop/username-studio

Would love feedback — especially if you've built something
similar or have ideas for scoring improvements.

Top comments (0)