Your site can answer questions now.
Visitors ask in plain language. The answer comes from your own pages and PDFs, with citations.
Wyregraf indexes your posts, pages, custom post types, PDFs, and public URLs, then answers questions from them in a chat on your site. Every answer cites the page it came from. When the answer is not in your content, it says so instead of guessing. It runs inside WordPress with your own OpenAI or Anthropic key, and your vectors and conversations stay in your own database.
Index what you already have. Answer from it, with proof. Keep control of what it costs.
Hybrid keyword and vector retrieval over your own content. Every answer links back to the page, post, or PDF it came from, so a reader can check it.
When the answer is not in your content, the assistant returns your fallback message instead of making something up. Set a persona and the allowed topics per chat.
Posts, pages, custom post types, PDFs, public URLs, and content you paste in. Generated summaries and questions make the right passage easier to find.
Drop it on a page with a shortcode, or turn on a floating launcher for one page or the whole site. Your accent color, logo, avatars, welcome message, and starter questions.
Per-role and guest limits by questions, tokens, or spend over hourly, daily, weekly, or monthly windows. Blocked words and IP bans stop a question before it reaches the model.
Transcripts with search and retention, usage analytics, thumbs up and down on every reply, and a Knowledge Gaps screen that shows what to write next.
Self-hosted, with your own key. No vector database to rent, no AI service in the middle.
Vectors and conversations live in your own WordPress database. You bring an OpenAI or Anthropic key and pay the provider directly at their rates; nothing routes through Artifex. Keys are stored server-side and redacted from debug output. Answers stream token by token, with an all-at-once fallback on hosts that buffer output. A REST API, WP-CLI, JSON import and export, and a clean uninstall are included.
One license, one site, every feature. It renews yearly, cancel anytime. Your AI provider bills you separately for the API calls your site makes, and the quotas let you cap that spend. Keys are issued at checkout and activate under Wyregraf → Options → Licenses in your WordPress admin. Not the right fit? Every license has a full refund within 30 days, no questions asked.
Setup, providers and keys, sources and indexing, the chat and the launcher, usage limits, accuracy tuning, the Retrieval Lab, and agents. Written for the person who runs the site.
OPEN THE DOCS →Every version, what changed, and whether you need to do anything about it.
SEE WHAT'S NEW →Every license includes direct support from the person who maintains the plugin, answered within 48 business hours.
ASK A QUESTION →Yes. Wyregraf is bring-your-own-key: you supply an OpenAI or Anthropic key, or both. The plugin uses your key to create embeddings and chat answers, so you have a direct relationship with the provider and no Artifex-hosted service sits in between.
Embeddings are generated with OpenAI (text-embedding-3-small, text-embedding-3-large, and ada-002) or Voyage. Chat answers can use OpenAI or Anthropic (Claude), and you pick the exact model per profile. You can mix providers, for example embed with OpenAI and answer with Claude.
Beyond the license, your only running cost is what your AI provider charges for the API calls you make: a one-time embedding cost when you index content, plus a per-question cost when visitors chat. You pay the provider directly at their rates, can pick cheaper models, and can cap usage with quotas.
Your embeddings and conversations are stored in your own WordPress MySQL database. There is no external vector database and no AI middleman. The one place text leaves your server is your chosen AI provider: to create an embedding or an answer, the relevant text is sent to the OpenAI or Anthropic account whose key you configured.
Yes. Turn on the grounded-answer guardrail and, when a question is not covered by your content, the assistant returns your fallback message instead of guessing. You can also set a persona and restrict the allowed topics per chat.
The built-in MySQL vector store performs well up to roughly 50,000 to 100,000 stored chunks, depending on your hosting. Past that, plan on a dedicated vector backend; pgvector, Qdrant, and Weaviate support is on the roadmap.
Yes. The chat is plain, well-classed HTML styled by a stylesheet that only loads on pages that show a chat. Paste the shortcode into any text or shortcode widget, or turn on the floating launcher. Accent color, logo, avatars, and welcome message are settings, not custom CSS.
Ask for a refund within 30 days of purchase through the refund form and it is issued in full, no questions asked. Keep the plugin installed while you decide; nothing about the trial period is limited.
Ask your archive.
Twenty years of newsletters, minutes, and PDFs become something that answers questions, with citations, on your own site. Every license includes a year of updates and direct support, and a full refund within 30 days if it is not the right fit.