AI provider & limits

Connect your own AI account. You pay your provider directly, your key stays encrypted on your server, and limits plus a response cache keep costs predictable.

FreeSiteryn → AI Support Agent → AI Provider

Supported providers

ProviderDefault modelNotes
OpenAIgpt-4.1-miniThe default. A ChatGPT subscription doesn't include API access.
Anthropicclaude-haiku-4-5-20251001Claude models
Google Geminigemini-2.5-flash
Groqllama-3.3-70b-versatileFast open models
OpenRouteropenai/gpt-4.1-miniMany models, one key
Mistralmistral-small-latest
Custom endpoint—Any OpenAI-compatible HTTPS endpoint
WordPress AIautoConnectors set up in WordPress itself; no key stored by the plugin. Listed only when WordPress's AI Client is available (WordPress 7.0+).

Connecting

  1. Pick a provider card

    Each card shows its status: Connected, Not connected, Not tested, Invalid credentials, Rate limited or Error. The sidebar shows a step-by-step "How to connect" guide.

  2. Save the API key

    Paste it and click Save Key: "Key saved for OpenAI. Loading its models…". Keys are saved separately from the form. Remove Key deletes it.

  3. Choose a model

    "Provider default (…)", one from your account's list (Refresh models; without a saved key you see suggested models), or "Other model ID...".

  4. Save and test

    Save Changes (it keeps the provider, endpoint and model), then Test Connection under Status: "Connected to OpenAI using gpt-4.1-mini." It tests the provider and model shown, with that provider's saved key.

Keys are encrypted with libsodium using your site's security salts, stored on your server, never shown again and never sent to the browser. Listing models and testing send the key only to the provider you picked. If your site's salts change, enter the key again.

You can also define the key in wp-config.php; it then always wins:

define( 'SITERYN_AI_SUPPORT_AGENT_OPENAI_API_KEY', 'sk-…' );
// Or one key for every built-in provider:
define( 'SITERYN_AI_SUPPORT_AGENT_API_KEY', 'sk-…' );

For a Custom endpoint, enter the HTTPS base URL (e.g. https://gateway.example.com/v1) before saving its key; the key is tied to that endpoint.

The Connection card follows the provider you picked: Endpoint Base URL appears only for Custom endpoint, and WordPress AI has no API Key row.

Requests

Request Timeout (seconds)
Default 305–120

Visitors wait for each reply, so keep it short. Busy or failing providers are retried once.

Maximum Response Tokens
Default 600128–4096

Upper limit on each reply's length. Lower is cheaper; too low may cut answers short.

Answer Style
Default Precise

Precise, Balanced or Conversational. Precise keeps answers closest to your information, recommended for support.

Usage & cost controls

Messages per Shopper per Hour
Default 200–500, 0 = no limit

Per signed-in user or per anonymous visitor. Basic answers count too.

Messages per Day (whole site)
Default 5000–100000, 0 = no limit

When reached, basic answers take over (if on).

Response Cache (minutes)
Default 1200–10080

Reuses the reply to an identical question with identical context. Cache hits are free and don't count toward limits.

Log Provider Errors
Default On

Error type and HTTP status only (never messages or keys), to WooCommerce → Status → Logs. Requires WooCommerce.

Each visitor can also have at most 2 requests in flight, and the whole site 20. Daily token counts appear on System Status; your provider's dashboard is the authoritative record of charges.