AI provider
Optional. Connect your own AI provider account: you pay your provider directly, and your API key stays encrypted on this server. Rule-based recommendations never depend on AI.
The tab has three sub-tabs: Provider (the AI Provider and Connection cards), Limits & Usage (Requests & Limits and Usage (last 7 days)) and Ranking & Privacy (AI Ranking and Data Sent to the Provider).
Provider
The AI Provider card: pick the service that ranks your products. Each card shows that provider's own status (Connected, Not connected, Not tested, Invalid credentials, Rate limited or Error), and each provider keeps its own key.
| Card | Default model | Suggested models |
|---|---|---|
| OpenAI (default) · GPT models | gpt-4.1-mini | gpt-4.1, gpt-4o-mini, gpt-4o |
| Anthropic · Claude models | claude-haiku-4-5-20251001 | claude-sonnet-5-5, claude-opus-5-5 |
| Google Gemini · Google Gemini models | gemini-2.5-flash | gemini-2.5-flash-lite, gemini-2.5-pro |
| Groq · Fast open models | llama-3.3-70b-versatile | llama-3.1-8b-instant |
| OpenRouter · Many models, one key | openai/gpt-4.1-mini | anthropic/claude-haiku-4.5, google/gemini-2.5-flash, meta-llama/llama-3.3-70b-instruct |
| Mistral · Mistral models | mistral-small-latest | mistral-medium-latest, mistral-large-latest |
| Custom endpoint · Any OpenAI-compatible endpoint | None; choose a model | Read from the endpoint's /models list when it has one |
| WordPress AI · Connectors set up in WordPress | Set in WordPress | Only listed on WordPress 7.0+ with its AI client; uses the connectors set up in WordPress, no key stored here |
The sidebar's How to connect guide explains getting a key and billing for the selected provider, with links. This plugin keeps its own keys, separate from Siteryn AI Product Content.
Connection
Settings for the provider selected above. Save Changes keeps the provider, endpoint and model; keys are saved with their own button.
- Endpoint Base URL
Shown for Custom endpoint only: an HTTPS base URL such as
https://gateway.example.com/v1. Plain HTTP is refused. Enter it before the key: the key is tied to that host, and is withheld if the URL later points somewhere else. - API Key
Paste it and click Save Key. It's encrypted on this server and never shown again; Remove Key deletes it. The note under the field says whether a key is saved, defined in wp-config.php, or can no longer be decrypted.
- Model
Provider default (gpt-4.1-mini), a model from the list (Refresh models loads the models your key can use), or Other model ID... to type one. Then Save Changes.
- Status
Test Connection tests the provider and model shown above, with that provider's saved key, by sending one small request: "Connected to OpenAI using gpt-4.1-mini."
SITERYN_AI_PRODUCT_RECOMMENDATIONS_{PROVIDER}_API_KEY (for example SITERYN_AI_PRODUCT_RECOMMENDATIONS_OPENAI_API_KEY) or SITERYN_AI_PRODUCT_RECOMMENDATIONS_API_KEY for every built-in provider. A constant always wins and is never stored. The custom endpoint only uses its own SITERYN_AI_PRODUCT_RECOMMENDATIONS_CUSTOM_API_KEY.Limits & usage
The Requests & Limits card: guard-rails for your AI provider bill. Cached answers are free and don't count.
| Setting | Default | Notes |
|---|---|---|
| Request Timeout (seconds) | 30 | 5–120 |
| Max Response Tokens | 400 | 64–2048. Upper bound on each AI answer's length; a ranking needs very little. Raise it for reasoning models |
| Response Cache (minutes) | 1440 (1 day) | 0–10080. Identical requests reuse the earlier answer; 0 disables |
| Max AI Requests per Day | 500 | 0–100000. Site-wide, reset at midnight UTC; 0 means no limit |
There's also a fixed limit of 60 AI requests per hour. Usage (last 7 days) lists Requests, Cache hits, Errors and Tokens in / out per day (UTC); your provider's dashboard is the source of truth for charges.
- Requests & Limits card: timeout, max response tokens, response cache and daily limit
- Usage (last 7 days) table: requests, cache hits, errors, tokens in / out
AI ranking
The AI Ranking card. AI rankings are generated in the background and refreshed weekly.
Go Live Automatically: shown as soon as they are generated. Require My Approval: review each product's picks first. They wait for you in the product editor, and rule-based recommendations show until you approve them. AI only ever re-orders products that already exist in your store.
Data sent to the provider
The Data Sent to the Provider card. Only catalog data is ever sent, never customer, order or payment data.
Short description (trimmed to 200 characters), Tags, Attributes (visible attributes, up to 6 values each) and Price (with the store currency). Product id, name and category names are always sent; they are the minimum the AI needs to rank anything.