Skip to main content

Custom Models

BrowseWiz supports adding any model that exposes an OpenAI API-compatible chat completions endpoint. This means you can use models from providers like Google, Anthropic, Mistral, open-source models, or your own self-hosted models.

Setup​

To add a custom model, provide:

  1. Model Name: the model identifier included in the request payload (e.g., gemini-2.5-pro, claude-sonnet-4-20250514). Required.
  2. Endpoint Base URL: the endpoint URL; /chat/completions is appended automatically. Required.
  3. API Key (optional): sent as the bearer token and stored encrypted in your local extension storage
  4. Headers (optional): extra request headers sent with every request to the endpoint, for example a provider's workspace ID. Headers added with the lock on (Store securely) are stored encrypted. A header named Authorization replaces the bearer token built from the API key.

Configure custom models in Options → Models. See Options → Models for step-by-step instructions.

Custom models stay in this browser only: they aren't backed up or synced to your other browsers, they stay when you sign out, and they aren't in the data export.

Using a custom model​

Pick it under Custom models in the model menu at the left of the chat input. The menu's Add model item opens Options → Models.

  • Chat requests go straight from your browser to your endpoint, not through BrowseWiz servers.
  • BrowseWiz streams replies, so the endpoint must support streamed chat completions.
  • Every tool you have turned on is sent to the model. Tools, such as the Browser Use Agent, work only if the model supports tool (function) calling, so turn tools off in the Tools menu for a model that doesn't. The Browser Use Agent itself runs on BrowseWiz's own models.
  • Scheduled tasks and monitors always run on BrowseWiz's own models.
  • Whether the model can read attached images and screenshots depends on the model: they are sent to it either way.
  • Thumbs-up and thumbs-down ratings appear only on replies from BrowseWiz models, not on custom-model replies.

When the endpoint returns an error​

If a request to your endpoint fails, the chat says the error came from your model's endpoint, with the endpoint's own error text under Technical Details:

  • "<model>'s endpoint returned an error (HTTP <status>)." for an HTTP error.
  • On HTTP 404 it adds: "Check that the base URL ends at the API version (for example https://integrate.api.nvidia.com/v1, without /chat/completions) and that the model name is exact."
  • On HTTP 401 or 403 it adds: "Check the model's API key."
  • "Couldn't reach <model>'s endpoint. Check the base URL." when the endpoint doesn't answer at all.
Last updated on