First release of Leaf Concierge: a site assistant that answers from your own published pages, names the pages it used, and declines when your site has no answer.
Added
- Answers drawn only from your published, public pages. Drafts, private pages and password-protected pages are never indexed.
- Checked citations: the model must number its sources, and only pages the reply actually cites are shown to the visitor.
- Decline mode: when no page covers a question, no model is called and the question is recorded under Content gaps.
- Content gaps report: declined and not-helpful questions, counted and ranked, with a button to write an approved answer.
- Approved answers: your exact wording for a question, served with no model call.
- Page awareness: questions such as “summarise this page” use the page the visitor is reading.
- Team requests: visitors can send a question to you, with a consent tick, by email and in the dashboard.
- Four providers with failover inside a fixed time limit: OpenAI, Claude, Google Gemini and OpenRouter. You use your own account and key.
- Limits per visitor per hour, and per day and per 30 days for the whole site.
- Test panel on the Overview screen showing matched pages, match scores and whether a model was called.
- Licence activation and licensed updates through the Leaf licence server.
Security and privacy
- Provider keys are encrypted in the database, or can be kept out of it with a wp-config.php constant.
- No IP address, name or account is stored with a question. A one-way hash of the address is kept for about an hour to apply the hourly limit.
- Records expire after a number of days you set. WordPress personal data export and erase are supported for team requests.
- Reply text is rendered as text, never as HTML, in the visitor’s browser.
Known limits
- Only content stored in WordPress is indexed. Text built dynamically at display time is not.
- Answers are not streamed; the visitor sees the reply when it is complete.