Ollama local LLM

When the in-browser object detector is unsure, the API server can ask a local vision LLM via Ollama. Nothing leaves your LAN. You provide the hardware.

Install Ollama

macOS, Linux, or Windows — installers at ollama.com/download.

# macOS / Linux
brew install ollama       # or use the installer
ollama serve              # starts on http://localhost:11434

Pull a vision model

Model	Size	Latency	Best for
moondream	~1.7 GB	~500 ms	Fast triage
llava:7b	~4 GB	2–5 s	Detailed analysis
llama3.2-vision:11b	~7 GB	5–10 s	Complex reasoning

ollama pull moondream
ollama pull llava:7b

Wire to SafeOS

Add the Ollama host URL to your API server's .env:

OLLAMA_HOST=http://localhost:11434

If Ollama runs on a different machine on your LAN, use that machine's IP. Open port 11434 in its firewall.

Verify it works

From the package root, run the bundled health check:

pnpm ollama:check    # GETs http://localhost:11434/api/tags

Then pick a model under Settings → AI Models. If it's detected, the dropdown will list the model name.