Brave Leo and Ollama: set up GPT-OSS 20B locally

Brave's tutorial connects a local model to Leo through Bring Your Own Model. It uses Ollama in the desktop browser, with GPT-OSS 20B as the example configuration in the description.

The speaker lists Brave version 1.69 or higher, at least 8 GB of RAM and storage requirements that depend on the model. The walkthrough starts with installing Ollama, checking its version in a terminal and downloading a model. These are the video's stated prerequisites; the selected model determines the actual hardware needed.

In Brave Settings, open Leo and scroll to Bring Your Own Model. Automatic sync detects local Ollama models. Manual setup takes a label, an exact model request name, a server endpoint and a context size. The supplied example uses gpt-oss:20b, http://localhost:11434/v1/chat/completions and a context size of 4000.

Select the model in Leo's sidebar, then turn off Wi-Fi and ask a question to check offline operation. Local inference and remote providers have different data handling: connecting a remote model sends requests to that provider. The offline test checks that the downloaded model can respond without an internet connection.