The Custom AI Providers feature enables a Bring Your Own Model (BYOM) approach, letting you connect your own large language model (LLM) backend to Cribl AI features instead of using the default Cribl-managed LLMs. This topic describes how to add, edit, or delete a custom AI provider.

Prerequisites ​

Before you start, make sure that:

• You have Owner or Admin permissions for the on-prem deployment or Cribl.Cloud Workspace. Only Owners and Admins can add, edit, or delete a custom AI provider. Users in other roles cannot view this configuration and receive a permission error if they call the underlying API.
• Cribl AI is configured correctly for your deployment (Cribl.Cloud or on-prem).
• For on-prem deployments, you are configuring your LLM on the Leader Node (not on a Worker/Edge Node), and the Leader Node has network access to your LLM endpoints.
• You have at least one supported LLM account, with:
  • A deployment identifier (such as the Deployment URL). Not required for Amazon Bedrock or Google Gemini.
  • Credentials (for example, an API key or bearer token) with permission to invoke the model.

Open AI Settings ​

1. Sign in to Cribl:
   • Cribl.Cloud: Open the Workspace where you want to configure Cribl AI.
   • On-prem: Open the Leader.
2. Select Settings (⚙).
3. If available for your deployment, open the Global tab.
4. Select AI Settings.

On the AI Settings page you can:

• Review Cribl AI availability and the model that powers it.
• See whether a custom AI provider is configured for the deployment.
• Open the custom AI provider configuration wizard to add, edit, or delete your own LLM.

Add a Custom AI Provider ​

To start the configuration, go to the AI Model Providers section of the AI Settings page and select Use Custom AI Provider.

The New Custom AI Provider wizard guides you through three focused steps.

Step 1: Select a Provider Type ​

Choose the type of LLM backend you want to connect:

• AWS Bedrock: Anthropic Claude models accessed through AWS Bedrock.
• Microsoft Foundry: OpenAI models accessed through an Azure OpenAI / Microsoft Foundry deployment.
• Google Gemini: Gemini models accessed through Google’s OpenAI-compatible developer API.
• LiteLLM: A self-hosted LiteLLM proxy that fronts one or more model backends.
• OpenAI Compatible: Any endpoint that implements the OpenAI Chat Completions API, including self-hosted models and third-party proxies.

Select a tile and then select Continue to move to the next step.

Step 2: Enter Connection Details ​

Provide the information Cribl needs to reach and authenticate with your provider.

URL Requirements ​

When you provide a Deployment URL or URL, Cribl validates the value before testing or saving the provider. The URL must:

• Use the http: or https: scheme. Other protocols are rejected.
• Not end with a trailing slash.
• Not include the request-path segment that Cribl appends automatically, such as /chat/completions.

Some provider types require a specific base-path suffix on the URL. For example, Microsoft Foundry requires the URL to end in /openai/v1. Bedrock and Gemini do not use a URL field: Bedrock reaches its own endpoint, and Gemini reaches Google’s default OpenAI-compatible endpoint. See the provider-specific configuration page for exact format requirements:

• Anthropic via Amazon Bedrock (no URL field)
• OpenAI via Microsoft Foundry
• Google Gemini (no URL field)
• LiteLLM
• OpenAI-compatible endpoints

In Cribl.Cloud, Cribl applies additional validation to prevent custom providers from probing internal cloud infrastructure:

• The hostname must not be localhost.
• The hostname must not be a literal private IP address (for example, anything in 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) or a link-local address (for example, 169.254.0.0/16).

If validation fails, the wizard returns an error and the provider is not saved. On-prem deployments do not enforce these address restrictions, because the Leader Node typically reaches LLM endpoints over a private network.

Select Continue to proceed to the model tier assignment step.

Step 3: Assign Models to Tiers and Test ​

Cribl routes AI requests to one of three model tiers based on the complexity of the task. In this step, assign specific model IDs to each tier for your chosen provider.

For AWS Bedrock, Microsoft Foundry, and Google Gemini, the wizard pre-fills suggested models from a supported model registry. You can select different models from the dropdown, but selection is restricted to the models Cribl validates for that provider.

For LiteLLM and OpenAI Compatible, there is no dropdown restriction. Enter the exact model IDs your endpoint accepts for each tier. Your endpoint must be able to handle every model ID you specify.

For providers that support suggested defaults (AWS Bedrock, Microsoft Foundry, and Google Gemini), select Reset to Defaults to restore the suggested model for every tier. See Reset Model Tiers to Suggested Defaults.

When all tiers are assigned, select Test and Save:

• Cribl runs a connectivity check for each assigned model, sending a basic request through your configured endpoint to confirm that the model is reachable and responds. This does not test model performance or feature compatibility.
• If every model passes, the configuration is saved automatically.
• If any model fails, the results panel shows the outcome (for example, 2 of 3 connections failed) and lists the failing models with their error messages. When more than a few models fail, select Show more to view the full list. The tier fields for failing models are flagged so you can correct them.
• After correcting the configuration, select Retry. The modal remains open so you can adjust values without starting over.

After you save:

• The configuration is stored in the Cribl secret store.
• A Custom AI provider card appears at the top of AI Settings showing the provider name, type, and the number of configured models.
• Supported AI capabilities in that deployment begin using your custom provider for new requests.

Reset Model Tiers to Suggested Defaults ​

For providers that offer suggested defaults (AWS Bedrock, Microsoft Foundry, and Google Gemini), the model tier assignment step includes a Reset to Defaults button. Select it to restore every tier to the same models Cribl suggests for a newly added provider of that type.

This is useful when editing a provider and you want to discard your custom tier choices and return to Cribl’s recommendations. Note the following:

• Reset to Defaults applies only to provider types that support suggestions. It does not appear for LiteLLM or OpenAI Compatible, which have no suggested defaults.
• The button is disabled when your current tier selections already match the suggested defaults.
• There is no separate confirmation. The Test and Save step serves as confirmation, so no changes take effect until you test and save.

Edit an Existing Custom AI Provider ​

You can update a configured provider if any details change (for example, keys rotate or you want to switch models).

1. Open Settings > AI Settings.
2. Locate the Custom AI provider card at the top of the page.
3. Select the gear (⚙️) on the card, then select Edit.
4. In the wizard, navigate through the steps to update the relevant fields.
5. On the final step, select Test and Save to validate and apply the changes.

When you reopen the wizard, the API Key and Virtual key fields are blank. This is by design: Cribl never returns stored credentials to the UI or API after you save them. To keep using the existing key, leave the field blank and update any other fields as needed. To rotate the credential, enter a new value, which replaces the stored key on save.

New AI requests use the updated configuration immediately after saving.

Delete a Custom AI Provider ​

Deleting a custom AI provider switches the deployment back to Cribl-managed LLMs for all supported capabilities.

1. Open Settings > AI Settings.
2. Locate the Custom AI provider card at the top of the page.
3. Select the gear (⚙️) on the card, then select Delete.
4. Confirm the deletion in the dialog.

After confirmation, the provider card is removed and Cribl AI begins using the default Cribl-managed models again.

If you want to use your own LLM again later, repeat the Add a Custom AI Provider steps.

Troubleshoot Common Errors ​

If there is a problem with your custom AI provider, you may see:

• Model test failures in the Test and Save step.
• Provider-specific errors surfaced in Cribl AI features.

Typical issues include:

• Invalid or expired credentials - Rotate or correct the API key, then retry.
• Wrong Deployment URL - Ensure the endpoint is reachable and correctly formatted (no trailing path suffixes). See URL Requirements for additional validation rules in Cribl.Cloud.
• Incorrect model ID - Confirm each model ID exactly matches what your provider or proxy expects.
• Rate limits or quotas exceeded - Check usage and raise limits or use a different deployment.

If you cannot resolve the issue quickly, delete the custom provider to fall back to Cribl-managed LLMs while you investigate.

When Your Custom Provider Is Unavailable ​

If your configured LLM endpoint is unreachable, Cribl AI features that depend on the provider return an error rather than fall back to Cribl-managed models. Common causes include a network outage, an expired credential, or a provider-side incident. Expect the following behavior:

• When you open AI Settings, Cribl automatically tests the configured provider. If the provider is unreachable, the page shows a connectivity alert in place of the provider card. Select Try Again on the alert to test the connection again after you resolve the issue.
• In-product Copilot features surface an error message indicating that the AI request failed. The error includes the provider ID so you can identify which custom provider returned the failure.
• Cribl does not automatically remove or disable a provider that is returning errors.
• Cribl does not retry against its own managed models when a custom provider is configured. This preserves the data-residency and governance guarantees described in Why Use Custom AI Providers?.

To restore the connection to your custom AI provider:

1. Confirm the provider is reachable from the Leader Node (for on-prem deployments) or from Cribl.Cloud (for cloud deployments).
2. Open AI Settings, select the gear (⚙️) on the provider card, then select Edit.
3. Run Test and Save to validate the connection. If the test still fails, correct the failing field (URL, API key, model ID) and retest.
4. If you need AI features immediately and cannot restore the provider, delete the custom provider. Cribl AI then routes requests to Cribl-managed models until you reconfigure your custom provider.