AI providers

Ciele uses one provider abstraction for Anthropic, OpenAI, and Google models. An Assistant selects its provider connection and model.

Provider connection types

Type	Use
Platform plan	Uses models supplied by the managed Ciele plan.
Subscription	Uses a member subscription for that member's own chats.
API key	Uses an Organization-managed provider key.
AI Gateway	Uses Anthropic, OpenAI, and Google models through AI Gateway.

A member subscription runs only that member's own chats: the Assistant Preview, and their chats with an AI Teammate. It does not serve a published widget, and it does not run another member's chat. Select a Platform plan or API-key connection for published traffic.

Add an API-key connection

Open Settings.
Select AI Provider.
Add a provider connection.
Select Anthropic, OpenAI, Google, AI Gateway, or ElevenLabs.
Enter the required provider credential.
Save the connection.

AI Gateway

An AI Gateway key is one key for Anthropic, OpenAI, and Google models. Ciele sends these calls through your own Vercel AI Gateway account, and that account pays for them.

If the Organization has no AI Gateway key, Ciele can use the platform AI Gateway key when the deployment has one. The platform plan pays for these calls.

AI Gateway does not serve every model. A model that AI Gateway does not serve does not show AI Gateway as a source.

Other connection types

OpenAI-compatible endpoints connect a server such as Ollama, vLLM, or LM Studio. Enter its base URL, models, and Context window in tokens. Use the context size configured on that server. The API key is optional. The base URL must use HTTPS and point to a public host. Ciele refuses addresses on a private network, such as localhost, 10.x.x.x, or 192.168.x.x. To use an endpoint on a private network, set it in the deployment environment. See Configuration.
Keyless enterprise auth connects Google Vertex, Anthropic, or Azure OpenAI through workload identity federation. Ciele stores no static key for these connections.

Ciele encrypts the credential before storage. Do not reuse a personal key that has unrelated permissions.

ElevenLabs connections supply voice models. Select these models in the Assistant's General settings under Voice mode. Google and OpenAI connections can also supply voice models. Available models depend on the connected account and its API permissions. ElevenLabs does not appear in the text model or knowledge embedding selectors.

Select an Assistant model

Open the Assistant.
Select General.
Select the provider connection.
Select an available model.
Optionally, select the model source.
Test the Assistant in Preview.
Publish the Assistant.

Model sources

One model can be available from more than one source. For example, Claude Sonnet 5 can use your Anthropic API key and also AI Gateway. The source selector shows when a model has two or more sources.

Automatic uses the first available source. The order is your API key, keyless enterprise auth, the platform plan, your AI Gateway key, and the platform AI Gateway key.
Platform plan, API key, Keyless, AI Gateway · your key, and AI Gateway · platform use only that source.

AI Gateway · your key uses the Organization's AI Gateway connection, and your Gateway account pays. AI Gateway · platform uses the platform AI Gateway key, and the platform plan pays.

If the selected source cannot serve the model, Ciele answers with another available model, as it does when a provider has no credential.

The models that visitors or colleagues can choose can also use one source each. The list shows a model once for each source, with the source name after the model name.

A member's own subscription is not a source that you select. It runs that member's own chats, whichever model or source the Assistant uses.

The published widget uses the provider and model in its publication snapshot. Later General changes require another publication.

Provider accounts are separate

Ciele API keys control Ciele administration. Provider API keys authorize model use. One credential cannot replace the other.