> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.runtype.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.runtype.com/_mcp/server.

# Managing AI models

Use **Settings > Models** to choose the AI models that appear in Runtype model
pickers, change the credential that each model uses, and set the default model.

## Before you begin

* Open the organization whose models you want to manage. All members of an organization
  share its model list, model settings, and default model. Runtype records the member who
  makes each change.
* Check your role. A **Member** can add and remove models and set the default model. Only
  an **Admin** can change which provider credential the organization uses. For role
  details, see [Team members and permissions](/user-guide/settings/team-members-and-permissions).
* To compare models by task, capability, cost, and provider, see
  [Available AI models](/user-guide/settings/available-ai-models).
* To decide between the Runtype platform key and your own provider keys, see
  [Platform keys vs. bring your own key (BYOK)](/user-guide/getting-started/platform-keys-vs-byok).

## Require your own model keys

An organization admin can enable **Require your own model keys** in **Settings > Models > Advanced settings**.
Select and test your organization connections on the **Providers** tab first, then enable
the setting and confirm **Require BYOK**. You can also enable it without any connections;
model requests then fail until you configure the required connection.

The dashboard assistant, flow generator, and product generator use system
credentials and are unavailable while this restriction is enabled.

This restriction covers model calls for Flows, Agents, assistant features, embeddings,
evaluations, images, and voice. It refuses Runtype platform keys, including platform
gateway keys and Workers AI. A missing, revoked, or unavailable organization connection
causes an error, never platform fallback. Personal keys that are not selected as an
organization connection do not qualify.

Custom OpenAI-compatible providers count as BYOK too. Select an organization connection
with your provider's base URL and API key before using its models in this mode.

**Applying** means previously admitted work is still active. New calls already follow
the restriction. Active voice sessions close; background sessions can require cancellation
or archiving before the status becomes **Enforced**. Contact support if it remains
**Applying** after all work has stopped.

Cloudflare voice and ElevenLabs voice are unavailable in this mode. OpenAI realtime voice
requires an eligible organization connection. With ZDR enabled too, use an eligible model
through your organization's Vercel AI Gateway connection. This setting governs which
credentials Runtype uses; your gateway account controls its own downstream provider routing.

The restriction does not delete models, keys, or saved provider choices. Turning it off
restores those choices, including configured platform fallback. It does not change
non-model integrations or replace Runtype's subscription and execution limits.

If the setting is unavailable on your deployment, enforcement rollout is not yet complete.

## Require zero data retention for inference

An organization admin can enable **Zero data retention (ZDR)** in **Settings > Models > Advanced settings**.
The policy applies to the workspace, including API-triggered executions. Personal accounts
cannot enable this workspace setting.

With ZDR enabled, language model and embedding calls use eligible Vercel AI Gateway routes.
Runtype passes the Gateway's zero data retention option on each call. Vercel refuses routes
that cannot satisfy that requirement. For the provider agreement, see
[Vercel's ZDR documentation](https://vercel.com/docs/ai-gateway/security-and-compliance/zdr).

Turning on ZDR does not delete saved models, change their enabled flags, or replace the
default model. Unsupported models remain saved but are unavailable in model pickers.
Existing Flows and Agents that reference an unsupported route fail when they use it.
Choose a supported routed model or Vercel route, and change the default explicitly if
necessary. Turning ZDR off restores the models' previous availability.

Supported routed models use Vercel even if the workspace has native provider credentials.
Explicit native-provider routes, custom endpoints, and other gateways are unavailable.
This also blocks native Claude Managed inference, the Anthropic proxy, sandbox Workers AI,
unsupported realtime or image models, and Weaviate server-side vectorization. Provider-native
tools that require a direct connection cannot run under this policy. Google Search grounding
can remain on a supported Vercel route.

GPT-Live voice is the one exception. It runs under ZDR only through a direct OpenAI
connection whose OpenAI organization Runtype detected as ZDR when the key was connected or
rotated. Without such a connection, GPT-Live calls are refused.

Stop or restart running executions when changing the policy. An execution that already
loaded its credentials retains that policy snapshot. For storage across conversations,
Records, logs, memory, files, and connected services, see
[Data retention](/user-guide/settings/data-retention).

For routing details, see [Model routing and data flow](/user-guide/settings/model-routing-and-data-flow).

## Add a model

Only the models that you add appear in model pickers throughout Runtype. Add only the models
that your team uses, so that the pickers stay short.

To add a model, follow these steps:

1. Go to **Settings > Models** and open the **Models** tab.
2. Click **Add model**.
3. In the **Add Model** dialog, filter the catalog by **Model Creator**, **Capabilities**,
   **Context Length**, or **Price**, or search by name.
4. On the model card, click **Add**.

**Add** configures the model with the Runtype platform key. If a model is available from
more than one provider, open the menu next to **Add** to choose the provider.

If a model has no platform key, its **Add** button is unavailable. To use that model,
connect your provider credentials on the **Providers** tab. For details, see
[Connecting AI model providers](/user-guide/integrations/connecting-ai-model-providers).

## Change a model's credential

A configured model uses either the Runtype platform key or one of your provider keys. To
change the credential of a model that uses your own key or shows **Needs setup**, follow
these steps:

1. On the **Models** tab, open the model card's actions menu.
2. Select **Edit Configuration**.
3. To use a saved key, open the **Use Existing Key** tab. Under **Select PROVIDER Key**,
   choose **Platform Key** or one of your saved provider keys.
4. To enter a new key, open the **Add New Key** tab and enter your **API Key**. For Amazon
   Bedrock, enter the **AWS Region**, **AWS Access Key ID**, and **AWS Secret Access Key**.
5. Click **Update Configuration**.

In step 3, `PROVIDER` is the provider name, such as OpenAI or Anthropic.

A model that uses the platform key doesn't show **Edit Configuration**. To move it to your
own key, connect the key for that provider on the **Providers** tab.

In an organization, a provider key that you select becomes the credential for every run of
that provider in the organization. This includes runs that any member starts, scheduled
runs, and runs with no signed-in user. Runtype asks you to confirm. To save the change,
click **Use for the organization**.

Your own provider keys require the Team plan or higher. On the Launch plan, the dialog
shows **Add a payment method** instead of the key options. For plan details, see
[Billing and plans](/user-guide/settings/billing-and-plans).

## Set a default model

To set the default model, follow these steps:

1. On the **Models** tab, open the model card's actions menu.
2. Select **Set as Default**.

When you add a prompt step to a Flow, Runtype preselects the default model.

## Remove a model

To remove a model, follow these steps:

1. On the **Models** tab, open the model card's actions menu.
2. Select **Delete Configuration**.
3. In the **Delete Model Configuration** dialog, click **Delete Configuration**.

You can't undo this action. To use the model again, add it again and re-enter your API key
if the model used one. The actions menu shows **Delete Configuration** only when you have more
than one configured model.

You can also remove a model from the **Add Model** dialog. On a model that you already
added, click **Added**. Runtype removes the model without asking you to confirm.

## Read model card badges

Each configured model card shows badges that describe its state. The following table lists
the badges:

| Badge                   | Meaning                                                                           |
| ----------------------- | --------------------------------------------------------------------------------- |
| **Default**             | The model is the default model.                                                   |
| **Platform Key**        | The model uses the Runtype platform key.                                          |
| **BYOK**                | The model uses your own key or the organization's selected provider key.          |
| **Needs setup**         | The model has no credential. Select **Edit Configuration** to choose one.         |
| **Custom**              | The model is a custom model that you configured.                                  |
| **Needs configuration** | Teammates saved conflicting settings. The model can't run until you resolve them. |

A card can also show a credit promotion badge next to the model name when a promotion
applies to that model.

## Review cost and usage

When Runtype has pricing or usage data for a model, its card shows the cost per 1K tokens
and the number of requests this month. The **Usage Snapshot** card on the same page shows
this month's total requests and cost, and the model with the highest cost. For details,
click **View detailed usage**.

## Manage models from the CLI, API, or MCP server

You can manage model configurations without the dashboard:

* **CLI**: Run `runtype models list` to see your configured models and their
  configuration IDs. Run `runtype models default <config-id>` to set the default model,
  or `runtype models disable <config-id>` to turn off a model without removing it. To
  add a model, use the dashboard, the API, or the MCP server.
* **API**: Use the `/v1/model-configs` endpoints to list, create, update, enable, disable,
  delete, and set a default model configuration. An API key needs the `MODELS:WRITE`
  permission to make changes, and `MODELS:DELETE` to remove a model.
* **Runtype MCP server**: Call `list_model_configs`, `create_model_config`,
  `update_model_config`, `toggle_model_config`, `set_default_model_config`, or
  `delete_model_config`.

In these commands, `<config-id>` is the configuration ID that `runtype models list` shows.

## Troubleshooting

### A model shows Needs configuration

A red **Needs configuration** badge means that teammates saved conflicting settings for
the model. The card says **Teammates saved conflicting settings**. The model is disabled,
and you can't set it as the default, until you resolve the conflict.

To resolve the conflict, follow these steps:

1. On the **Models** tab, open the model card's actions menu.
2. Select **Edit Configuration**.
3. In **Organization model settings**, edit the settings into the one version that every
   member uses.
4. Click **Resolve and Re-enable**. Runtype replaces the conflicting versions and turns the
   model back on.

### Adding a custom inference endpoint model fails

The **Custom inference endpoint** provider uses your own endpoint and key, so it requires the
Team plan or higher. On the Launch plan, the request fails with the `BYOK_NOT_AVAILABLE`
error. To add the model, upgrade your plan in **Settings > Billing**.

## Next steps

Continue with these guides:

* [Connecting AI model providers](/user-guide/integrations/connecting-ai-model-providers): connect provider credentials
* [Model routing and data flow](/user-guide/settings/model-routing-and-data-flow): learn where requests go for each credential
* [Prompt caching](/user-guide/settings/prompt-caching): reduce model cost for repeated prompts
* [Using prompt steps](/user-guide/flows/using-prompt-steps): select models in Flows
* [Usage data](/user-guide/settings/usage-data): review model costs and execution activity