Skip to main content

​
Overview

Choose either connection method:
  • AI Puffer Cloud: connect without an API key. Eligible verified accounts receive 25 monthly free credits; optional top-ups pay for more hosted usage.
  • Your own provider: enter the provider’s API key and pay that provider directly.
You can keep both connected and choose the provider/model separately in each module. Settings > AI manages connections and defaults; the provider dialog does not choose a module’s model.
Pro unlocks plugin features. Cloud credits, your direct provider’s API billing, and the visitor credits you may sell on your WordPress site are separate from the Pro license.

AI Puffer Cloud

Hosted models without an API key, with monthly free credits.

OpenAI

Text, images, embeddings, speech, web, and realtime.

Google

Gemini, images, video, embeddings, TTS, and grounding.

Anthropic

Text, web search, and supported image analysis.

OpenRouter

Access many text, image, embedding, and web-capable models.

Azure

Azure OpenAI deployments for text, images, embeddings, and speech.

DeepSeek

Text generation for chat, writing, forms, and automations.

xAI

Grok text models, web search, supported image analysis, and image generation.

Ollama

Local or self-hosted text and embedding models.
AI providers power text, image, audio, , and retrieval features. ✓ means supported. - means not supported in the current provider strategy. Support also depends on the model and module. Cloud images means generation; image editing and video use supported direct providers. Cloud image analysis requires a text model with image-input capability. Realtime voice uses a separate OpenAI connection. AI Puffer > Settings is the central area for managing your connections to different AI providers.
AI Settings provider connection cards

​
Connect your own provider

  1. Open AI Puffer > Settings > AI.
  2. Click Connect on the provider card, or Manage if it is already connected.
  3. Enter or update that provider’s key and any required endpoint fields. Use Show/Hide to inspect the key.
  4. Click Connect to validate and save it. A rejected replacement is not saved.
  5. Refresh or sync the provider’s models when needed, then choose a model in your module.
A connected provider card indicates a saved connection. It does not guarantee every model is available to that account. New modules use your configured defaults where supported; changing a default does not replace every existing chatbot or saved template.

​
OpenAI

OpenAI supports the widest set of AI Puffer features.
  1. Open AI Puffer > Settings > AI.
  2. Click Connect or Manage on the OpenAI card.
  3. Paste your OpenAI API key.
  4. Keep the default base URL unless you use a compatible custom endpoint.
  5. Click Sync models beside the default model.
  6. Select the model you want to use by default.
For knowledge features, OpenAI embedding models are available in embedding model selectors. OpenAI Vector Stores are created and managed in Knowledge Base > Stores. OpenAI moderation is also available under Advanced settings. Set Moderation to Yes to check OpenAI chatbot input with OpenAI moderation, then set Moderation Message if you want to customize the message shown when input is blocked.
OpenAI API key settings

​
Google

Google provides Gemini, embeddings, image/video models, text to speech, and grounding.
  1. Open AI Puffer > Settings > AI.
  2. Click Connect or Manage on the Google card.
  3. Paste your Google API key.
  4. Click Sync models beside the default model.
  5. Select the default model.
  6. Review Google safety settings if you need to change blocking thresholds.
Google image and video model settings are configured in Images.
Google safety settings can block some responses. Change them only when the default behavior is too restrictive for your site.
Google API key settings

​
Anthropic

Anthropic is available for text workflows, web search, and supported Claude image analysis models.
  1. Open AI Puffer > Settings > AI.
  2. Click Connect or Manage on the Anthropic card.
  3. Paste your Anthropic API key.
  4. Click Sync models beside the default model.
  5. Select the default model.
Anthropic does not provide embeddings through the current AI Puffer provider strategy.
Anthropic API key settings

​
OpenRouter

OpenRouter gives access to models available in your OpenRouter account.
  1. Open AI Puffer > Settings > AI.
  2. Click Connect or Manage on the OpenRouter card.
  3. Paste your OpenRouter API key.
  4. Click Sync models beside the default model.
  5. Select the default model.
Some capabilities depend on the selected model.
OpenRouter support depends on the model you select. Check the model capabilities before using image, embedding, or web features.
OpenRouter API key settings

​
Azure

Azure uses your Azure OpenAI .
  1. Open AI Puffer > Settings > AI.
  2. Click Connect or Manage on the Azure card.
  3. Paste your Azure API key.
  4. Enter the Azure endpoint URL for your resource.
  5. Open Advanced settings.
  6. Check the API versions if your Azure resource requires different versions.
  7. Click Sync models to load deployments.
  8. Select the deployment to use as the default model.
Azure uses deployment names. If a deployment does not appear, confirm it exists in Azure and that the endpoint and API key belong to the same resource.
Azure API key settings

​
DeepSeek

DeepSeek is available for text workflows.
  1. Open AI Puffer > Settings > AI.
  2. Click Connect or Manage on the DeepSeek card.
  3. Paste your DeepSeek API key.
  4. Click Sync models beside the default model.
  5. Select the default model.
DeepSeek embeddings are not supported by the current provider strategy. Use OpenAI, Google, Azure, OpenRouter, or Ollama for embedding workflows.
DeepSeek API key settings

​
xAI

xAI is available for text workflows, Grok models, web search, supported image analysis models, and image generation.
  1. Open AI Puffer > Settings > AI.
  2. Click Connect or Manage on the xAI card.
  3. Paste your xAI API key.
  4. Keep the default base URL unless you use a compatible xAI endpoint.
  5. Keep the API version as v1.
  6. Click Sync models beside the default model.
  7. Select the default text model.
AI Puffer uses xAI’s Responses API for text generation and streaming. Legacy completions and chat completions are not used by this provider integration. xAI image generation and image editing use xAI image models such as grok-imagine-image. xAI image understanding uses Grok language models that support image input. xAI is not available as an embedding provider, vector store provider, video provider, speech provider, or realtime voice provider in the current integration.
xAI API key settings

​
Ollama

Ollama connects local or self-hosted models.
  1. Install Ollama from ollama.com/download.
  2. Run Ollama on the computer or server you want to use as the AI server.
  3. Pull a model:
  1. Open AI Puffer > Settings > AI.
  2. Click Connect or Manage on the Ollama card.
  3. Enter the Ollama base URL. The default is http://localhost:11434.
  4. Click Sync models.
  5. Select the default model.
Install notes: You can pull more than one model. AI Puffer shows synced Ollama models after Sync models runs.
If WordPress and Ollama are on different servers, do not use localhost unless Ollama is running on the same server as WordPress. Use the reachable server URL instead.
Do not expose Ollama publicly without access controls. Anyone who can reach the Ollama server can send model requests to it.
Ollama API key settings

​
WordPress AI Connectors

WordPress 7.0 includes a built-in AI Client and a Settings > Connectors screen. AI Puffer can manage those WordPress AI connectors so WordPress AI features, themes, and plugins that call wp_ai_client_prompt() use your AI Puffer provider setup. This does not replace the provider setup above. Add your provider keys in AI Puffer > Settings > AI first, then enable connector management when you want WordPress AI Client requests to route through AI Puffer. AI Puffer exposes OpenAI, Google, Anthropic, OpenRouter, Azure OpenAI, DeepSeek, xAI, and Ollama to the WordPress AI Client. Replicate remains available in AI Puffer’s Images module, but it is not exposed as a WordPress AI connector. See WordPress AI Connectors.

​
Troubleshooting

The provider account is out of credits, has reached a spend limit, or is blocked by billing settings.Check these items:
  1. Open the provider billing page.
  2. Check credits, usage limits, and monthly spend limits.
  3. Add credits or update billing if needed.
  4. If you just changed billing, wait a few minutes and try again.
For OpenAI, see error codes.
The saved key is wrong, old, revoked, copied with extra spaces, or belongs to a different account or organization.Check these items:
  1. Create or copy a fresh API key from the provider dashboard.
  2. Paste it again in AI Puffer > Settings > AI.
  3. Save the setting.
  4. Sync models again.
For OpenAI, see Incorrect API key provided.
AI Puffer uses saved model catalogs so opening a module does not wait for each provider. Connect or sync the provider to update its catalog.Check these items:
  1. Open AI Puffer > Settings > AI.
  2. Click Manage on the provider card.
  3. Click Sync models beside the default model and wait for it to finish.
  4. Reopen the model picker in the module.
  5. For frontend Images, check any restriction in Images > Settings > Frontend Models.
If the model still does not appear, confirm the provider account has access to that model.
Some OpenAI models require a verified API organization.Check these items:
  1. Open OpenAI Platform settings.
  2. Go to Organization > General.
  3. Complete organization verification.
  4. Wait up to 30 minutes.
  5. Generate a new API key if the error continues.
  6. Make sure AI Puffer is using a key from the verified organization.
For OpenAI, see API Organization Verification.
A WordPress security plugin or firewall may be blocking AI Puffer settings requests.Check these items:
  1. Check your security plugin or firewall logs.
  2. Whitelist AI Puffer admin requests if your tool supports allow rules.
  3. If using Wordfence or a similar firewall, switch to learning mode.
  4. Save AI Puffer settings and use the affected module a few times.
  5. Switch the firewall back to normal mode after it learns the requests.