Knowledge Base
Use Chatbot → Knowledge Base to teach the assistant from your real content: WordPress posts and pages, single URLs, uploads, sitemap-discovered pages, curated Q&A, or manual notes. Each addition is a source you configure; then you run Learn so the model can use it in replies. You can enable Playground next to the Knowledge Base heading to chat in the admin and verify answers without loading the front end.
Settings
Under Knowledge Base → Settings you configure embedding models, vector storage, and search behavior—do this before ingesting sources if the UI requires it (for example, when the chat provider is not OpenAI or Gemini).
Embedding models
Choose one of the two providers that support embeddings for your content: OpenAI or Google Gemini. After you choose a provider, the API key and model fields appear.
This works like Chatbot → AI Settings, but AI Settings lists more providers overall. Embeddings turn text and other content into vectors (numeric representations) so the system can run similarity search. Only OpenAI and Gemini are supported for embeddings in the knowledge base.
If AI Settings already uses OpenAI or Gemini as the chat provider, you are not asked for a separate embedding provider here. If you use any other provider for chat, you must finish embedding model setup (provider, key, and model) under Knowledge Base before you can add and ingest knowledge.
Vector storage
Supported options: Local (default), Pinecone, and OpenAI. OpenAI vector storage fits best when OpenAI is already your chat provider in AI Settings.
- Local (default): Vectors are stored in the WordPress database. No extra setup. Best for smaller knowledge bases—typically up to about 10,000 items.
- Pinecone: Hosted vector database, common at larger scale. Add your Pinecone API key in the Pinecone section.
- OpenAI: Optional hosted vector storage from OpenAI; pairs cleanly when your chat provider in AI Settings is OpenAI.
When to scale up: Local storage is a strong default under roughly 10,000 knowledge items. Beyond that, Pinecone is usually the better fit.
Search settings
- Similarity threshold: Minimum similarity score (0–100%) for a retrieved snippet to count as a match.
- Minimum similar results: Minimum number of similar knowledge items to retrieve for the model when building a reply.
- Show knowledge sources: When enabled, lists the sources referenced under each assistant message.
Adding knowledge
- Click Add Knowledge.
- Choose a source type (see Supported sources below).
- Set options and visibility for that source.
- Click Learn and wait for ingestion to finish.
- Turn on Playground beside the Knowledge Base heading and ask questions that should use the new knowledge (see Learning and verification below).
You can add many sources over time; repeat the same flow for each new one.
Supported sources
WordPress
- Pick a post type (posts, pages, or custom post types).
- Narrow what to include: select specific items manually or filter by taxonomies, custom fields (meta), date range, or author.
- Choose which fields are sent to the knowledge index. The free tier includes title, main content, excerpt, post date, author, and taxonomies. PRO adds ACF fields, JetEngine fields, and native post meta.
URL
- Paste a single public page URL. The source is fetched and learned from that address.
File
- Choose one or more files from the WordPress Media Library.
- Supported types include PDF, TXT, DOCX, and CSV.
Sitemap
- Paste your XML sitemap URL.
- Click Fetch URLs, then choose which discovered links to learn.
Q/A
- Create question-and-answer pairs you want the assistant to follow.
- One answer can be tied to several similar phrasings of the same question.
Manual
- Use the rich text editor to paste or write content with or without formatting.
Visibility (roles)
Every source has a visibility control: limit which user roles receive that knowledge in chat. That gives you role-based access to different parts of what the model knows.
Learning and verification
When a source is configured, click Learn. An interactive progress view shows ingestion status.
Where to test: On the same Knowledge Base screen, switch on Playground next to the page heading. The playground opens there so you can prompt the assistant and confirm it uses what you just learned.
