AI Providers and Usage
Audience: Company Admin, Group Admin and Super Admin
Settings → AI Provider (/platform/ai-provider) manages company models and shared group models. Choosing a model and deciding which records it can use are separate: configuring a provider does not expand a user's record access.
Provider scopes
| Scope | Management |
|---|---|
| Company models | An authorized company administrator adds, edits, tests, orders and deletes providers for the selected company. |
| Group models | Managed by Group Admins and the Super Admin; shared between companies. A Company Admin can hide a group model in their own company. |
| Platform default | Managed by the Super Admin under Administration → AI Provider Settings. It can serve as the default when company/group models are unavailable. |
Automatic selection considers company models, then non-hidden group models, then the platform default in priority order. You can also explicitly choose an eligible model offered by the chat model picker. Hiding a group model in your company does not delete it for other companies.
Add a company or group model
- Select the correct company and use Add Provider in the relevant card. The group action appears only when you can manage it.
- Enter provider type and display name. Available types/models depend on active model-registry entries.
- Enter the required API key and select a model. Leaving the key blank while editing preserves its stored value; the saved key is not returned in plaintext.
- For a custom OpenAI-compatible provider, enter model name and base URL; leave the key blank if the service requires none.
- Save, then use Test on the row to inspect the connection result. If it fails, review the message, model, destination address and provider credentials.
- Use up/down actions to change priority and the active switch to enable/disable use. Deletion requires confirmation.
A standard provider's base URL comes from the selected model-registry entry; a custom provider's URL is editable. Never paste secret keys into chat or support records.
SaaS monthly AI pool
The top card shows availability, monthly pool and usage. The group pool covers a calendar month, calculated as active users × monthly share per user; unused capacity does not roll over. Only messages and web research served by the COS model consume the pool. Company/group providers do not consume it; your provider's pricing and agreement apply.
If the pool is full, new COS-model requests are blocked; choose your own eligible provider if available. SaaS tool rate limits, chat retention and record permissions still apply. Access to AI Usage also requires administration authority.
On-prem installations have no AI pool/rate limits; the installation administrator configures a working provider. If none is configured or a connection fails, correct configuration; an absent/unreachable provider does not imply a successful response.
Data sharing
Messages and relevant tool-result content can be sent to the selected provider. Check your organization's data-sharing and privacy rules before use. Review the proposed records and values before approving a chat write. See AI Assistant for the chat flow and Platform Settings for retention and web-research configuration.