Configure the LLM model

Open Settings → LLM Model and pick a guide:

• OpenAI
• Anthropic
• OpenRouter
• Google/Gemini via OpenRouter
• Local Ollama
• LM Studio via Custom

Fields

Field	How to fill it in
Provider	Select one of the five available options. Each provider can keep its own saved configuration.
Model name	Use the exact identifier available in the API or on the local server. The extension’s suggestions do not guarantee your account has access.
Endpoint	OpenAI, Anthropic and OpenRouter use fixed, non-editable URLs. Ollama and Custom are editable.
API key	Required for all except Ollama, including Custom. The eye button shows/hides the typed value.

In Custom, provide the full URL compatible with OpenAI Chat Completions, including /v1/chat/completions. Do not assume Browser Mate adds that path. An endpoint that only implements the Responses API does not satisfy this route.

Save and test

1. Fill in the model, the endpoint when editable and the key when required.
2. Click Save and confirm the success message. A key already saved for the same provider is preserved if you leave the field empty.
3. Click Test connection and read the result and the capability report.
4. Start with a text-only request, no actions.

The test can cost money

“Test connection” sends real requests to the provider to verify the response and capabilities. It can consume tokens and balance; it is not a simple offline verification of the key.

Capabilities

The report distinguishes text, images, files, tools, streaming and web search. “Unknown” does not mean “supported”. Some results use heuristics and the public OpenRouter catalog; real behaviour depends on the model and the endpoint. For automation, prefer models with structured tools.

Common problems

• 401/403: check the key, permissions and account restrictions.
• Model not found: copy the identifier from the API, not the commercial name.
• 429: check balance, quotas and call limits.
• Incompatible response: confirm Chat Completions and tool support.
• Local server unreachable: check the address, port, running server, firewall and origin/CORS rules. Do not disable those protections globally.

These guides were checked against the extension code; they do not represent authenticated tests made with your accounts.