Skip to Content
Customize and connectModels, API keys and web search

Models, API keys and web search

Use a Starizon-hosted model or connect your own model provider. Web search has its own provider setting, so switching models leaves your search setup as it is.

Use a Starizon model

Sign in, choose Starizon as the model source in Settings and pick a model from the side panel’s model menu. Hosted model requests use your Starizon credits.

The model menu shows the current choices. Advanced models require Basic, Pro or Lifetime. Pick a model that supports the inputs and tools your task needs. Tasks that inspect screenshots, for example, need image support.

The /research command uses Starizon’s hosted research service. Its planning step can use your selected custom model, but the report itself requires a signed-in Starizon account and uses Starizon credits. You can still research pages and use web search in AI Chat with a compatible custom model.

Bring your own API key

  1. Open Settings and choose Custom API Keys under the model setting.
  2. Enable the provider you want and enter its API key.
  3. Set up the model list. For an OpenAI-compatible provider, enter its Base URL and the model identifiers that provider supplies.
  4. Choose the configured model in the side panel and send a short request to test it.

The settings support providers such as OpenAI, Gemini, Anthropic and Grok, plus compatible endpoints. Use Add OpenAI-compatible provider for a separate endpoint with its own key and models.

Enter the provider’s API endpoint, including any version path it requires. The address of a provider’s chat website will not work as a Base URL. A display alias helps you recognize a model, but the provider has to accept the underlying model identifier.

Changes apply to new tasks. A task already running keeps the model it started with.

Set language and response preferences

In Settings, UI language changes the interface labels and Response language sets the language you prefer for answers.

Custom Prompts holds separate instructions for AI Chat, PageChat and Agent S6. Edit the field for the mode you want to change. The other modes keep their own. For instructions you choose task by task, create a Skill.

Choose a web search provider

Starizon uses its built-in web search service by default. To use your own search account:

  1. Open Settings → Web Search.
  2. Set up a provider and enter its search API key. Fill in Endpoint when the provider needs one.
  3. Select Test and check that results come back.
  4. Choose the Default provider and turn on Use custom web search as default.
  5. Keep Web enabled in the chat input for tasks that need searches.

The settings include Tavily, Brave Search, Serper, Exa, Parallel, Linkup, Perplexity and SearXNG. These are search connections, separate from the model providers above. For SearXNG, use a reachable instance set up to return JSON search results.

Test sends the provider a small search request. A passing test confirms the connection. The services still differ in search coverage and supported filters. New tasks pick up the updated search setting, while running tasks keep the one they started with.

To go back to built-in search, turn off Use custom web search as default. MCP search tools have their own connection and access settings.

Understand the charges

RequestWhere usage is charged
Starizon-hosted modelStarizon credits
Model connected with your own API keyYour model provider
Starizon’s built-in web search, app services and email alertsStarizon credits
Your configured search providerThat provider, under its own terms

Your own model key pays for model requests only. Connected Starizon services still use credits, and your plan’s task limits still apply. See Plans and credits.

If a custom connection fails

Check the key, endpoint and model identifier first. Then confirm that the provider account has access and usage left. To use Starizon’s tools, choose a model that supports tool calling.

For search problems, run Test in Web Search settings and read the error it returns. A search API key and a model key are separate, so fixing one leaves the other unchanged.

Keep keys in Settings and redact them from screenshots and support messages. Troubleshooting covers other connection and task problems.

Last updated on