How Validation Works

Validate Key and Load Models does two things: it proves the key works, and it loads the list of models you can pick from.

Hosted providers

  1. Ping. The module sends the prompt ping to a small fixed test model:

    ProviderTest model
    ChatGPT (OpenAI)gpt-3.5-turbo
    Google Geminigemini-3-flash-preview
    Anthropic Claudeclaude-3-5-sonnet-20240620
    Mistral AIdevstral-medium-latest
    OpenRouteropenai/gpt-oss-20b
    Cerebrasgpt-oss-120b
    Coherecommand-r-plus-08-2024
    DeepSeekdeepseek-chat

    Any reply means the key is valid. A rate-limit or quota error also counts as valid — the key was accepted, the account is just out of credit or throttled.

  2. Load models. The module reads the provider's model catalogue and keeps only text-in / text-out chat models. For OpenAI, Gemini, Anthropic, Mistral, OpenRouter, and Cerebras it also checks the provider's live model list so retired models are dropped.

Ollama

The module calls GET {endpoint}/api/tags (15-second timeout) and lists every model your server has pulled. There is no ping.

After validating

  • The button reads Validated.
  • AI Model becomes a dropdown.
  • Changing the provider, key, or endpoint clears the model and shows Provider or key changed. Please validate again.

On save

For hosted providers the module fetches the model list once more on the server and rejects a model that is not in it. See the error list in General Settings. Ollama skips this check.

Quota-limited keys pass validation

Because rate-limit errors count as success, a key with no credit validates and saves, then fails at generation time with API request rate limit exceeded. Please try again after some time. Check your provider billing if summaries never appear.

Next: Ollama (Self-Hosted).