AI Connections
Open Settings > AI, then AI providers and models to use your own provider API keys or add models from a compatible endpoint. Bring Your Own Key is available on Professional, Team, Business, Enterprise, and Custom plans.
These connections configure model access. Agent Connections configure the sources and tools an Agent can use.
Personal and Organization Keys
Individual accounts manage personal keys. Organization members use the organization's keys and key preference. Organization administrators and members with permission to manage settings can change them; other members see the organization's configuration.
Saved API keys are encrypted. Clear Ideas displays a masked hint after saving, and the full saved key is not returned to the interface. Provider requests use the credential on the server.
Add and Test a Provider Key
- Select Add provider key.
- Choose the provider and enter its API key. Supported providers include OpenAI, Anthropic, Google Gemini, xAI, Groq, Cerebras, and Cohere.
- Test the key, then save it.
- Choose a Key preference and select Save preference.
A saved key can serve matching models already in the Clear Ideas catalog. You do not need to add each catalog model again.
Key Preference
| Setting | Behavior |
|---|---|
Use managed provider access (managed) | Use Clear Ideas-managed access for catalog models, even when a provider key is saved. This is the default. |
Prefer saved API keys (preferred) | Use a matching saved key when configured. Managed access remains available when no matching saved key is configured. |
Require saved API keys (required) | Reject a foundational-model request when it has no matching saved key. Managed access is not used as a fallback. |
Custom models use their own saved connection. Model and organization policy still apply to the selected model. Key preference controls credentials; it does not grant access to Sites, tools, or otherwise restricted models.
Add a Custom Model
Use a custom model when its endpoint or model identifier is not already covered by the catalog.
- Select a saved provider key, or add an OpenAI-compatible endpoint with its HTTPS base URL and key.
- Use model discovery where the provider supports it, or enter the provider's model identifier manually.
- Set a display name and the supported API protocol. Responses is available for OpenAI, xAI, and compatible endpoints; use the protocol your endpoint actually supports.
- Enter the context window and maximum output tokens. Maximum output must be smaller than the context window.
- Enable tool calling, structured outputs, and streaming only when the model supports them.
- Test the model before using it in Chat or an Agent.
Discovery lists candidate model identifiers; it does not establish every model's capabilities or token limits. Private-network endpoint addresses are rejected, and production endpoints must use HTTPS.
Custom models appear in applicable model selectors when enabled and supported by that feature. Features requiring tools or structured outputs may exclude incompatible models. See Models and Model Policy.
Billing and Changes
Inference routed through your saved key is billed by that provider. Clear Ideas does not charge its model-token credits for that customer-routed inference. Separate tool, operation, and compute charges can still use Clear Ideas credits.
Disabling or removing a connection makes its custom models unavailable. With Require saved API keys, removing the last matching key also prevents requests to the affected catalog models. Update dependent chats and Agents before removing a connection they use.
See AI Credits, AI Settings, and Agent Runs.