🔍
Codebase-aware
Indexes your .gd files with ChromaDB. Reads the files it actually needs before writing.
🔒
100% local
Your code never leaves your LAN. Ollama runs on your own hardware.
🛡️
Git safety net
Every patch is committed to a session branch. Multi-level /undo always available.
⚡
Truncation protection
Three layers prevent the AI from writing corrupt partial files.
Install
git clone https://github.com/ja33man/godot-ai-dev.git cd godot-ai-dev pip install -r requirements.txt
Run
# Windows python godot_ai.py C:\Users\you\MyGame --host http://192.168.2.49:11434 # Mac / Linux python3 godot_ai.py ~/MyGame --host http://192.168.2.49:11434 # Ollama on same machine python3 godot_ai.py ~/MyGame --host http://localhost:11434
Models (pull on your Ollama server)
ollama pull nomic-embed-text # ~274 MB — indexes your codebase ollama pull qwen2.5-coder:14b # ~9 GB — writes and fixes GDScript ollama pull qwen3.5:2b # ~2.7 GB — fast chat answers
Commands
| Command | What it does |
|---|---|
| /fix | Paste a Godot error → diagnose → patch files |
| /feature | Describe a feature → generate GDScript → write files |
| /chat | Ask anything about Godot or your project |
| /run | Launch Godot here — errors trigger auto-fix offer |
| /watch | Watch Godot's log file for errors in real time |
| /retry | Re-run last command with a stricter prompt |
| /undo | Revert last AI patch via git reset |
| /models | List Ollama models and switch code/chat model |
| /help | Full command reference |