Frequently asked questions
Do I need MySQL 9?
No. MySQL 9 unlocks native VECTOR storage and fast cosine-distance search. On older MySQL or MariaDB, the plugin automatically falls back to storing embeddings as JSON and scoring them in PHP. The Settings → Vector Database screen shows which mode is active.
Do I need to pay for anything?
You need an account with at least one LLM/embedding provider. OpenAI, Anthropic, Groq, and Cloudflare are paid (with free tiers); Ollama is fully free and runs locally. Cloudflare Workers AI includes a generous free tier and can serve both embeddings and chat.
Can I run fully local / offline?
Yes. Point both the LLM and embedding providers at a local Ollama instance and use the MySQL vector store. No external requests are made.
Can I use Cloudflare for everything?
Yes. Cloudflare Workers AI can serve both the embeddings and the chat completion, and Cloudflare Vectorize can host the vector index.
Is my data sent to the plugin authors?
No. All requests go directly from your WordPress server to the providers you configure. No telemetry, no call-home, no analytics. See Privacy.
How are citations built?
Each retrieved chunk stores the source URL (the permalink for posts/pages, the imported URL for documents). Citations are deduplicated by URL and listed under the answer. The LLM is also instructed to reference source numbers in its text.
Can I customize the chatbot’s appearance?
Yes — go to Settings → Appearance. Choose one of four preset themes (Light, Dark, Ocean, Sunset) or override individual colors. You can also set a logo, bot avatar, bot name, welcome message, and launcher position.
Does it work on multisite?
The plugin activates per-site. Network activation is supported but each site maintains its own knowledge base.
What are the limits of the JSON fallback?
The JSON fallback loads up to 5,000 chunks per query and scores them in PHP. For larger knowledge bases, upgrade to MySQL 9 (native VECTOR) or use Cloudflare Vectorize.