An AI node can fail because the provider is unavailable, a key is missing or invalid, the selected model is unavailable, the prompt is empty, or the provider request returns an error. Use the saved node message as the starting point instead of guessing which provider-specific issue occurred.
Check the provider and node configuration
- Open Agent Workflows → Workflows and edit the affected workflow.
- Select the AI action and confirm its API key configuration, Model, and required prompt fields. The chat action nodes expose an API key, Model, optional System prompt, and required User prompt.
- Check any
{{...}}variable tokens in the prompt and confirm that the trigger or earlier node actually supplies those fields. - For the AI Agent node, check AI provider (fallback when no Chat Model is attached), API key (fallback provider), AI model (fallback), and its prompt configuration. If a Chat Model is attached, confirm that model’s provider and settings as well.
- Confirm the selected provider is one of the available providers in this version: OpenAI, Google Gemini, Anthropic Claude, OpenRouter, Groq, or DeepSeek.
Screenshot to add: [AI action provider and prompt settings]
Check credentials and provider availability
The plugin uses the WordPress AI Client stack, with a compatibility path for WordPress versions before 7.0. The editor links to the appropriate credentials screen for the installed environment. Use that link when configuring a site-wide provider key, or supply the key through the node’s supported API key field.
- If the saved message says no API key is configured, add a key for the provider selected in that node.
- If the editor cannot load models or the provider is not registered, confirm the AI Client is available and the selected provider is supported by this installation.
- If saving a site-wide key fails validation, re-enter the key and verify that the site can reach the provider. The key-validation path returns a generic connection/key error rather than a detailed provider response.
- Check the model value against the models available for the selected provider. Changing the provider without reviewing the model can leave an incompatible configuration.
Inspect the failed run
- Open Agent Workflows → Runs, select the failed or partial run, and find the AI node in Node log.
- Read the node’s Message first. The execution service records an action error message, or a generic message if the action did not provide one.
- Choose View in the Details column to inspect the saved Input and Output. The saved Input is the node configuration snapshot, not a complete record of every runtime-resolved variable value. The Output contains the recorded action result or error.
- If the message indicates a missing prompt, correct the prompt. If it indicates missing credentials, configure the selected provider. If the message contains a provider or transport error, check the provider service status, site outbound connectivity, and the actual request configuration before changing the workflow.
When a retry may occur
Background runs that finish as Failed or Partial may be retried when the recorded error matches a recognized temporary-failure signal, such as a timeout, rate limit, upstream server error, or dropped connection. The retry is a new run and starts the workflow again from the beginning. It does not resume at the AI node, so earlier actions can run again. A missing key or incorrect model configuration is unlikely to be fixed by retrying.
Test a corrected AI node with a short, non-sensitive prompt before running the complete production workflow. Never paste API keys or private customer data into support requests.