- The ones the user sees (like a sidepanel). These often use LLM API's like OpenAI.
- The browser API ones. These are indeed local, but are often very limited smaller models (for Chrome this is Gemini Nano). Results from these would be lower quality, and of course with large contexts, either impossible or slower than using an API.
Comments
There's two types of built-in LLM's:
- The ones the user sees (like a sidepanel). These often use LLM API's like OpenAI.
- The browser API ones. These are indeed local, but are often very limited smaller models (for Chrome this is Gemini Nano). Results from these would be lower quality, and of course with large contexts, either impossible or slower than using an API.