(FIRST install zip then extract zip to start or and clone the repo I peronally would pick to clone the repo then start on this)
Remade for infinite persistent chats with SQLite DB (chats.db), 1024 token context window, /projects shows ALL files, free better AI via HF API.
- Persistent Chats: History saved forever across runs. Multiple sessions.
- Context Window: Exactly 1024 tokens max (truncated smartly).
- ALL PROJECTS ON SCREEN:
/projectslists my_ai_trainer/ + Desktop files. - Free Better AI:
python chat_ai.py --api(DialoGPT-large online, no key). - Commands:
/list,/projects,/new,/switch,/delete,/clear,/reset-rate(web rate limit reset).
Default when you click run_ai.bat? FREE ONLINE AI (like ChatGPT, connects to web brain, auto-fixes "too many asks" with waits/resets).
3 Brains Explained Like Candy:
- Web (--api/default) 🍭 Free online super-smart (DialoGPT-large), but sometimes "wait 4 quota". Auto-retries + /reset-rate = new asks!
- Simple (--simple) 🧁 Instant local small brain (DialoGPT-medium), no internet/train.
- Trained (--trained) 🧠 YOUR brain!
run_ai.bat trainfirst (teach on convos, ~5min GPU), then super personal.
Steps:
- Click
run_ai.bat→ Web chat! Say hi. /new MyChatnew talk,quit.- Want yours?
run_ai.bat train→python chat_ai.py --trained.
Saved forever in chats.db. /projects = see files!
First run: Enter session name e.g. "MyChat". Chats saved!
You: /projects
AI: Lists all your files/projects!
You: Hello
AI: Hi! (context-aware)
quit
Next run: History loads automatically!
run_ai.bat train # Merged full train (~5min GPU, hh-rlhf 5k samples, LoRA r=32)
python chat_ai.py --trained
chats.db: sessions + messages (role, content, tokens).- Infinite history, auto-truncate to 1024 tokens.
- Models: HF DialoGPT-large (--api), local DistilGPT2/LoRA.
- DB: SQLite full CRUD.
- Token truncate: Real tokenizer len.
Your mini-GPT with persistence + file browser! 🎉