Choosing AI providers and models

Choose how Cedros connects to AI services, then decide whether Cedros should select models for you or use a model you choose. A provider supplies the AI service; a model is the particular AI you use through that provider.

These are shared site settings. If you only want to start a conversation, see Getting started with the Cedros assistant.

Choose who manages your models

Open Settings → Assistant → Overview and find Models. If your site offers Provider source, choose the arrangement you intend to use:

• Cedros Providers: Cedros manages model selection. The individual default-model and backup controls are hidden, and Processing loop explains the managed flow.
• Your API keys: your site uses its own provider connections. You can choose a default model, configure backups, and customize models for individual steps or jobs.

The source switch appears when Cedros Providers is active on your site. If it is absent, use the provider controls available to you. If you see Assistant setup needs attention instead of the Models card, choose Open Intelligence Settings to finish the connection first.

Before switching to your own keys, make sure the providers you plan to use are connected. Changing the source does not create credentials or give your account access to a model.

Connect your own provider

1. Open Settings → Providers → Intelligence.
2. Find the provider you want to use and enable its card.
3. Add its API key in the credential field, or follow the connection option offered on that card. Keep credentials out of chat messages and page content.
4. Wait for the save status to confirm the change. If saving is blocked because an enabled provider has no key, finish that connection or turn the unused provider off.
5. Review the available models. Use Refresh models when you need to update the list from the provider.

Stored keys are masked. Use Replace to enter a new key; the saved value is not displayed again. Refreshing the model list does not select a new default model for the assistant.

Your provider account must allow the model you choose. A model appearing in a dropdown is not proof that your account can run it. Managed sites may show a different provider workspace; follow its connection and status controls instead of looking for site-level key fields.

Set the default model

1. Return to Settings → Assistant → Overview → Models with Your API keys selected, where that source option is available.
2. Choose a connected Default provider.
3. In Default model, choose Follow Cedros defaults to receive approved model updates, or select a specific model to keep that choice pinned.
4. If you need a model that is not listed, choose Custom model… and enter its exact provider model ID. Confirm that your provider account supports it.
5. Wait for the change to save, then reopen the settings to confirm the selection.

If the card says automatic model selection is on, Cedros selects an approved model for each request and hides the default-model dropdown. See the next section to review that setting.

When comparing models, use work that represents your needs: a short customer reply, a summary of a document, or a more involved planning request. Compare accuracy, usefulness, response time, and usage cost. Use the provider's current account information for model access and pricing.

Review automatic selection and individual steps

On Overview, open Processing loop → Steps & models. In the mode that uses your own keys, Choose the best model automatically controls automatic selection. Turn it off if you want to return to choosing the default model yourself.

If automatic selection is unavailable, finish connecting the required providers first. The readiness message explains why Cedros is using the default model instead.

You can also review model choices for individual processing steps. Use recommended models applies the recommended choices to those steps; review existing custom choices before using it. This button remains unavailable until the required provider setup is ready.

With Cedros Providers, this area describes the managed processing flow. You do not need to assign each step yourself.

Add a backup model

When using your own keys, expand Backup model in the Models card.

1. Choose Add backup model.
2. Select its provider and enter the model ID.
3. Resolve any warning about a disabled provider, missing key, or model that is not allowed.
4. If you add several backups, use the arrows to put them in the order you want Cedros to try. Wait for the changes to save.

A backup gives Cedros another model to try when the default is unavailable. It does not fix an incomplete connection. Review each backup's readiness before relying on it, and remove entries you no longer want used.

Check models used by background jobs

When your site manages its own routing, Settings → Assistant → Jobs shows which models background jobs use. Review On defaults, Customized, and Missing a provider key to find jobs that need attention. A customized job can keep a different model after you change the assistant default; use its Use the default action if you want to remove that customization.

Live chat uses the controls on Overview. The Jobs settings are hidden while Cedros Providers manages model selection for you.

Verify a change before relying on it

1. Confirm that the provider source, connection, and model selection saved successfully.
2. If you want to test the connection, use the provider card's test control and read the result. Connection tests and assistant requests can use paid provider services; run them only when you intend that usage.
3. Try one small, non-sensitive request that asks for an answer in chat. Use the same request when comparing models, following Writing useful requests for the assistant.
4. Review the answer and any reported usage or errors before using the setup for larger work.

A saved configuration or a readiness label does not establish that a live request will succeed. A successful connection test also does not prove that every model or task will work.

Resolve common problems

• Controls are read-only: ask your administrator for access to edit shared assistant settings.
• The Models card is missing: follow Open Intelligence Settings and check the selected provider's enabled state and credentials.
• Your chosen model is not being used: check automatic selection, custom processing-step models, and the model assigned to the particular job or task.
• A model is unavailable: check its exact ID, your provider account's access, and the configured model list. Refresh the list if needed.
• Requests fail after a change: read the error, check the connection and account limits, and return to a previously working model if appropriate. If the request may have changed a record or sent something, check the result before retrying.

Changing models does not grant the assistant additional permissions. Review what it can read or change separately in Settings → Assistant → Tools.