How Do I Fix a Broken AI Provider in Dave by voolama?

Summary

The steps

1. 1

   Identify the failure mode from the Fetch Models result

   Six failure modes are documented in the admin-providers help documentation for Dave by voolama (hellodave.ai), DEV v0.11.0, confirmed against content_orchestrator main at commit ddb945f, 2026-08-30. The Fetch Models button distinguishes the first three without requiring any field edits. Click it before touching anything else. The error strings it returns are the raw responses from the AI service itself, surfaced in the Dave by voolama UI. No vendor trust required.

   Fetch Models result or symptom	Failure mode	Field to fix
   Authentication error	API key expired, revoked, or incorrect	API Key field
   Resource not found	Base URI is wrong (most common on Azure OpenAI, OpenAI-Compatible, Custom)	Base URI field
   Connection timeout	AI service is unreachable	Check the AI provider's own status page
   Success but zero models listed	Key lacks list-models permission, or no models deployed (Azure)	Switch to JSON mode and enter models manually
   Provider absent from agent dropdown	Provider deleted or belongs to a different tenant	Recreate the provider
   Agents fail after provider deletion	Agents still reference the deleted provider	Update agent configurations to point to a new provider

   Do this

   1. Navigate to Admin, then API Management, then Providers.
   2. Identify the suspect provider from the agent error message.
   3. Click the pencil icon to open the edit form.
   4. Click Fetch Models without changing any field first.
   5. Read the result. Match it to the failure mode table above.
   6. Fix only the field the table names. Do not edit other fields.

   Example

   A workflow instance fails. The run timeline shows an agent_interaction node in error state with the message 'invalid_api_key'. The Admin navigates to Admin, then API Management, then Providers. The 'Production OpenAI' provider card is present. The Admin clicks the pencil icon to edit, then clicks Fetch Models without changing any field. The fetch returns 'Authentication failed: invalid API key.' Failure mode confirmed: expired or revoked key. The Admin does not touch the base URI or model catalog. The fix is in the API Key field only.

   Best practice

   The admin-providers help documentation states: "After creating or editing a provider, use Fetch Models to verify that the API key and base URI are correct. A successful fetch confirms the provider can reach the AI service." Always run Fetch Models as the first diagnostic step before editing any field. The fetch result names the exact problem. No vendor trust required.

2. 2

   Fix an invalid or expired API key

   The admin-providers help documentation for Dave by voolama (hellodave.ai) states: "The API key stored in the provider is expired, revoked, or incorrect. Fix: Navigate to Admin, then API Management, then Providers, edit the provider, and enter a valid API key. Click Fetch Models to confirm the new key works before saving." Sourced from the admin-providers help documentation, DEV v0.11.0, confirmed 2026-08-30.

   API keys in Dave by voolama are stored with AES-256-GCM authenticated encryption. Stored credential values carry a v2: version prefix that identifies them as GCM-encrypted, as distinct from values written before the GCM migration. The admin-providers help documentation states: "Keys are transmitted only over HTTPS and are never logged in plain text." This implements NIST SP 800-53 Rev 5 SC-28 (section 28a) and IA-5 (section 5b). Because providers are centralised in Dave by voolama, rotating the key in one provider record updates every agent that references it.

   Do this

   1. Generate a new API key in the AI provider's own console.
   2. In Dave by voolama, navigate to Admin, then API Management, then Providers.
   3. Click the pencil icon on the provider to edit.
   4. Clear the API Key field and paste the new key. Click the eye icon to confirm the new key is visible.
   5. Click Fetch Models. Confirm the fetch succeeds.
   6. Click Update Provider to save.
   7. Verify: run a workflow instance that uses an agent referencing this provider. The agent_interaction node completes without error.

   Example

   The Admin generates a new API key in the OpenAI console. In Dave by voolama, the Admin navigates to Admin, then API Management, then Providers, and clicks the pencil icon on 'Production OpenAI'. The Admin clicks the eye icon on the API Key field to confirm the old masked key is present. The Admin clears the field and pastes the new key. Before clicking Update Provider, the Admin clicks Fetch Models. The fetch succeeds and returns 47 models. The Admin clicks Update Provider. Verify: run a workflow instance that uses an agent referencing this provider. The agent_interaction node completes without error. No vendor trust required.

   Best practice

   The admin-providers help documentation states: "Rotate API keys regularly: Update the API key field periodically to follow security best practices." Always click Fetch Models after entering the new key and before clicking Update Provider. A failed fetch after entering the new key means the new key is also invalid: do not save it. NIST SP 800-53 Rev 5 IA-5 (section 5b) requires establishing lifetime restrictions for authenticators: regular key rotation in the provider record satisfies this requirement for AI service credentials.

3. 3

   Fix a wrong base URI or failed model fetch

   The admin-providers help documentation for Dave by voolama (hellodave.ai) states: "The API key is missing, the base URI is incorrect, or the AI service is unreachable. Fix: Verify the API Key field is filled in. Check the Base URI matches the service's expected endpoint. If using Azure OpenAI or a custom endpoint, confirm the URL is correct and the service is online." Sourced from the admin-providers help documentation, DEV v0.11.0, confirmed 2026-08-30.

   For known provider types (OpenAI, Anthropic, Google AI), the base URI is auto-filled and rarely the cause of a failed fetch. For Azure OpenAI, OpenAI-Compatible, and Custom providers, the base URI must be supplied manually and is the most common cause of a failed fetch. The seven provider types and their default base URIs:

   Provider type	Default base URI	Must supply manually?
   OpenAI	https://api.openai.com/v1	No
   Anthropic	https://api.anthropic.com/v1	No
   Azure OpenAI	(your Azure endpoint)	Yes
   Google AI	https://generativelanguage.googleapis.com/v1beta	No
   OpenAI-Compatible	(your endpoint)	Yes
   Local (Ollama)	http://localhost:11434/v1	No
   Custom	(your endpoint)	Yes, no model fetch available

   Do this

   1. Edit the provider. Check the API Key field: it must be filled in. Click the eye icon to confirm.
   2. Check the Base URI against the table above. For Azure OpenAI, OpenAI-Compatible, and Custom providers, confirm the URL matches your deployment.
   3. Click Fetch Models. Read the error message: authentication failure (key issue), resource not found (base URI issue), or connection timeout (service unreachable).
   4. Fix the named field. Click Fetch Models again to confirm.
   5. Click Update Provider to save.

   Example

   An Admin configures an Azure OpenAI provider. The base URI field was left at the auto-filled placeholder instead of the tenant's Azure deployment endpoint. Fetch Models fails with 'resource not found'. The Admin checks the Azure portal, copies the correct deployment endpoint, and pastes it into the Base URI field. The admin-providers help documentation recommends: 'Include the deployment name in the provider Name field for clarity (e.g., Azure OpenAI - eastus2 gpt-4o).' The Admin clicks Fetch Models. The fetch succeeds. The Admin clicks Update Provider.

   Best practice

   For Custom providers, model fetch is not available. The admin-providers help documentation states: "For Custom providers, model fetch is not available - you must supply the model catalog manually using the JSON editor." Use JSON mode and enter model objects in the format [{"id": "model-id", "name": "Model Name"}]. Verify the model ID against the custom service's own documentation before entering it.

4. 4

   Fix an empty model catalog or a missing provider

   Two additional failure modes are documented in the admin-providers help documentation for Dave by voolama (hellodave.ai), DEV v0.11.0, confirmed 2026-08-30.

   Empty model catalog after Fetch Models. The admin-providers help documentation states: "The API key has restricted permissions, or the service returned an empty model list. Fix: Verify the API key has permission to list models. For Azure OpenAI, ensure at least one model is deployed in your Azure resource. Try switching to JSON mode and entering models manually." The JSON mode workaround is the fastest verification: switch to JSON mode and enter [{"id": "gpt-4o", "name": "GPT-4o"}]. Save the provider. Configure an agent to use this provider and model. Run a test workflow instance. If the agent call succeeds, the key is valid but lacks list-models permission. No vendor trust required.

   Provider not appearing in agent configuration dropdown. The admin-providers help documentation states: "The provider belongs to a different tenant, or it was deleted. Fix: Confirm you are in the correct tenant. Check Admin, then API Management, then Providers to verify the provider exists. If it was deleted, recreate it." Verify: navigate to Admin, then API Management, then Providers. The provider list shows all providers for the current tenant. See Add an AI Provider to recreate it.

   Do this

   1. Empty model catalog: switch to JSON mode. Enter a known model ID: [{"id": "model-id", "name": "Model Name"}]. Save the provider. Test an agent that uses this provider and model.
   2. Missing provider in agent dropdown: navigate to Admin, then API Management, then Providers. Confirm the provider is listed. If absent, recreate it using Add an AI Provider.
   3. After recreating a provider: update every agent that referenced the old provider to point to the new one. See Create an Agent.

   Example

   An agent configuration dropdown shows no providers. The Admin navigates to Admin, then API Management, then Providers. The list is empty. The Admin confirms the correct tenant name in the top navigation bar. The provider was deleted. The Admin clicks Add Provider, selects Anthropic, enters the API key, clicks Fetch Models, selects the required models, and clicks Create Provider. The provider now appears in the agent configuration dropdown. Verify: open the agent configuration and confirm the provider is selectable. No vendor trust required.

   Best practice

   The admin-providers help documentation states: "Keep at least one active provider: If all providers are deleted or misconfigured, no agent can execute AI calls and all workflows that rely on agents will fail." Before deleting any provider, check which agents reference it and update them to point to a different provider first.

What this guide does not cover