Configure AI models
Insights requires a default Agent model that is ready to use. A default multimodal model is also required when a session needs to process images, scanned PDFs, or images embedded in documents.
Who configures models
AI models are managed by a platform administrator in the platform operations console, which is separate from the organization console. Contact the platform administrator if the New Chat page reports that a required model is unavailable.
Administrator workflow
- Open the platform AI model settings.
- Add a model and choose the provider.
- Enter the display name, provider model ID, API base URL, and credential.
- Test connectivity and save the configuration.
- Keep the model active and set an appropriate model as the default Agent model.
- When visual understanding is required, set a compatible active model as the default multimodal model.
- Create a small test session and verify both the answer and its citations.
API keys and provider credentials are secrets. Keep them out of screenshots, GitHub issues, terminal history, and documentation. If a screenshot is unavoidable, leave credential fields empty or cover each value completely with an opaque block; do not rely on blurring to hide a secret.
User-visible readiness
Organization members can confirm that the models are ready from the product interface:
- New Chat opens without a missing-model warning;
- the intended analysis type is available;
- Start Chat becomes available after all required fields are selected;
- preparation reaches Ready and a test question returns an answer.
Before replacing or disabling a default model, configure its replacement and verify it with a test session. The provider's location, terms, and log-retention policy remain part of the organization's data-processing decision, even when a Private Data Gateway is used.