Does it really work with no internet?
Yes, for any model you've already downloaded. Ollama, LM Studio, and llama.cpp all run entirely offline once a model is on disk: no telemetry, no phone-home, no API calls. The only steps that ever touch the internet are installing the software and downloading a model. After that, turning off Wi-Fi or enabling airplane mode changes nothing about how the model runs.
What you actually need before going offline
- Install Ollama, LM Studio, or another local runner while connected.
- Download the model(s) you want while still connected; this is the only large data transfer involved.
- Disconnect. Test a prompt before you actually need it offline, not after, so you're not troubleshooting mid-flight.
See the beginner guide for the full install-and-download walkthrough if you haven't done this yet.
Where this actually matters
- Flights and dead zones: a downloaded model keeps working with zero connectivity, unlike any cloud AI service.
- Privacy-sensitive work: nothing you type leaves your machine, which matters for some regulated or confidential work regardless of connectivity.
- Air-gapped environments: some workplaces don't allow internet-connected machines at all; a local model can run entirely within that constraint once transferred over.
- Avoiding ongoing cost: no per-token API bill, since there's no API involved at all.
What still needs a connection
A few things stop working offline, worth knowing before you rely on this: downloading a new or different model, any tool that does live web search from within a chat, and anything explicitly built to call a cloud API (some apps offer a mix of local and cloud models; only the local ones keep working offline). Everything else, the actual conversation with the model, runs the same with or without a connection.