A Google Stitch-like tool that generates complete HTML/CSS/JS UIs from natural language prompts, powered by your local Ollama models.
- Describe any UI in plain English → get live-preview HTML instantly
- Streaming token output from Ollama
- Edit mode — iterate on existing generated UIs
- History — revisit any previously generated UI in the session
- Copy or download the generated HTML
- Auto-detects available Ollama models
- Works with any Ollama model (llama3, mistral, codestral, qwen2.5-coder, etc.)
1. Start Ollama (must be running locally):
ollama serve
ollama pull llama3 # or any model you prefer2. Run STick:
bash run.sh| Env var | Default | Description |
|---|---|---|
OLLAMA_BASE_URL |
http://localhost:11434 |
Ollama API base URL |
OLLAMA_MODEL |
llama3 |
Default model to use |
OLLAMA_MODEL=qwen2.5-coder bash run.shFor best UI generation results:
qwen2.5-coder— excellent at HTML/CSS/JScodestral— strong coding modelllama3/llama3.1— good general purposemistral— fast and capable
| Endpoint | Method | Description |
|---|---|---|
/ |
GET | Web UI |
/api/models |
GET | List available Ollama models |
/api/generate |
POST | Generate UI (SSE streaming) |
{
"prompt": "A dark dashboard with revenue charts",
"model": "llama3",
"context": "<optional: existing HTML to iterate on>"
}