LOCAL AI SETUP

Use Ollama with Olungu

Ollama lets Olungu classify pages with a model running on your own computer. The main setup detail is allowing the Olungu browser extension to call your local Ollama server.

Endpoint

In Olungu, enter http://localhost:11434. Leave out /api/generate.

Origin access

Ollama must allow chrome-extension://*, otherwise Chrome extension requests return HTTP 403.

Model quality

Small models are fast but can misclassify pages. Use a 7B or 8B model when accuracy matters.

Recommended setup

These steps are for the Ollama macOS app. They allow browser extensions to call Ollama, then restart the app so the setting is applied.

launchctl setenv OLLAMA_ORIGINS "chrome-extension://*,http://localhost:*,http://127.0.0.1:*"
pkill -9 Ollama
pkill -9 ollama
open -a Ollama

If you run Ollama from the terminal instead of the macOS app, start it with:

OLLAMA_ORIGINS="chrome-extension://*,http://localhost:*,http://127.0.0.1:*" ollama serve

Configure Olungu

1. Open Dashboard. Go to AI configuration.
2. Select Local Model / Ollama. Use the Ollama local endpoint provider.
3. Set base URL. Use http://localhost:11434.
4. Set model. Start with llama3.2:3b, then move to a stronger model if classifications are too loose.

Verify locally

A successful test returns HTTP 200 and an Access-Control-Allow-Origin header. HTTP 403 means the origin setting is still not applied.

ollama pull llama3.2:3b
ollama ps
curl -i http://localhost:11434/api/generate \
  -H 'Origin: chrome-extension://test' \
  -H 'Content-Type: application/json' \
  -d '{"model":"llama3.2:3b","prompt":"Return only {\"ok\":true}","stream":false,"format":"json"}'

Model choices

llama3.2:3b

Fastest starting point

Good for testing that Olungu can reach Ollama. Accuracy can be weaker on ambiguous pages.

llama3.1:8b

Better balance

A stronger local option when the machine has enough memory and GPU headroom.

qwen2.5:7b

Good reasoning for classification

Often better than tiny models for focus relevance checks, but slower than 3B models.

Troubleshooting

Olungu shows HTTP 403

Ollama is running, but it rejected the browser extension origin. Set OLLAMA_ORIGINS with chrome-extension://* and fully restart Ollama.

Olungu shows HTTP 404

Use the base URL in Olungu: http://localhost:11434. Do not enter /api/generate in the dashboard field.

Dashboard says available, but checks are poor

The connection is working, but the model may be too small. Try an 7B or 8B model and keep temperature low.

curl works without Origin but fails with Origin

That is a CORS/origin issue, not a model issue. Re-apply OLLAMA_ORIGINS and restart the Ollama app or service.

What to expect

Local AI keeps page classification on your computer.
Speed and accuracy depend on your hardware and selected model.
Managed AI is usually more consistent; Ollama is best when local privacy is the priority.

Download Ollama Ollama origin docs Contact support