# Ollama Hello World
Local LLM inference, tools, vision, and embeddings with Genkit — all running
privately on your machine via [Ollama](https://ollama.com/).
## Prerequisites
- **Ollama** installed and running locally.
## Quick Start
The `run.sh` script handles everything — it installs dependencies, pulls
required models, and starts the Dev UI:
```bash
./run.sh
```
## Manual Setup
1. **Install Ollama**: Download from [ollama.com](https://ollama.com/download).
2. **Start the server**:
```bash
ollama serve
```
3. **Pull models**:
```bash
ollama pull gemma3:latest # General generation
ollama pull mistral-nemo:latest # Tool calling
ollama pull llava:latest # Vision / image description
ollama pull nomic-embed-text # Embeddings for RAG
```
4. **Run the sample**:
```bash
genkit start -- uv run src/main.py
```
## Testing This Demo
1. **Open DevUI** at http://localhost:4000
2. **Test basic flows** (uses gemma3):
- [ ] `say_hi` — Simple generation
- [ ] `say_hi_stream` — Streaming response
- [ ] `say_hi_constrained` — Structured output (HelloSchema)
- [ ] `generate_character` — Structured output (RpgCharacter)
- [ ] `generate_code` — Code generation
3. **Test tools** (uses mistral-nemo):
- [ ] `calculate_gablorken` — Tool calling demo
- [ ] `currency_exchange` — Currency conversion tool
- [ ] `weather_flow` — Weather tool
4. **Test vision** (uses llava):
- [ ] `describe_image` — Image description from URL
5. **Test embeddings & RAG** (uses nomic-embed-text + gemma3):
- [ ] `Pokedex` — Ask questions about Pokemon using local RAG
6. **Example queries for Pokedex**:
- "Tell me about fire-type Pokemon"
- "Which Pokemon can fly?"
- "What's the strongest water Pokemon?"
7. **Notes**:
- gemma2:latest does NOT support tool calling; use mistral-nemo
- Vision requires `ollama pull llava` before use
- First run may be slow (model loading into memory)
- All processing happens locally — no API calls
8. **Expected behavior**:
- Responses generated locally (no external requests)
- Streaming shows incremental output
- Tools work with compatible models only
- Vision describes images accurately
- Embeddings computed locally, similarity search finds relevant Pokemon
- RAG combines retrieval with generation
