First run
What happens the first time you open Lumen.
On first launch, Lumen downloads the local language model, mlx-community/Llama-3.2-1B-Instruct-4bit (roughly 700 MB), from Hugging Face and caches it on-device. This is the only time Lumen talks to Hugging Face and is the only component of the app that ever leaves your network boundary.
The model fetch is a one-time transaction. Once cached, it is reused forever and no further requests hit Hugging Face.
What to expect
| Phase | What you see | How long |
|---|---|---|
| Cold start | Lumen icon, empty browser | Instant |
| Model fetch | Knowledge panel marked Preparing | 1 to 3 minutes on Wi-Fi |
| Ready | Knowledge panel shows the sparkle | Stays ready after that |
Optional: pre-warm your library
If you want the Ask panel to answer useful questions from day one, browse five to ten articles you care about before opening it. Lumen classifies, embeds, and summarizes each one in the background. The retrieval step then has something to grip on.
Pages you read once and never return to also get stored. The library trims stale content automatically after 90 days of non-engagement.