Ollama

Ollama runs open-source AI models locally on your machine. Free, private, no API key needed.

Setup

Install Ollama: ollama.com/download
Pull a model:
ollama pull llama3.1
Pawz auto-detects Ollama at localhost:11434

That’s it — no API key or configuration needed.

Configuration

Setting	Default	Notes
Base URL	`http://localhost:11434/v1`	Change if Ollama runs on another machine
API key	(none)	Not needed for local Ollama
Default model	—	Set to your preferred model

Recommended models

Model	Size	Best for
`llama3.1:8b`	4.7 GB	General use, fast
`llama3.1:70b`	40 GB	High quality, needs GPU
`gemma2:9b`	5.4 GB	Good balance
`mistral:7b`	4.1 GB	Fast, European
`codellama:13b`	7.4 GB	Code generation
`deepseek-coder-v2`	8.9 GB	Code generation

Embedding model

Pawz uses nomic-embed-text for the memory system:

ollama pull nomic-embed-text

Pawz auto-pulls this model if it’s missing.

Remote Ollama

To use Ollama on another machine:

On the remote machine, set OLLAMA_HOST=0.0.0.0:11434
In Pawz, change the base URL to http://<remote-ip>:11434/v1

Tips

Ollama uses your GPU automatically if available
Models are downloaded to ~/.ollama/models/
Run ollama list to see installed models
Run ollama rm <model> to delete a model
Ollama models use the : syntax (e.g., llama3.1:8b) — Pawz auto-routes these

OpenAIUse GPT-4.1, o3, o4-mini, and other OpenAI models.