Knowledge base
The knowledge base holds what your AI should know about your business: pages and posts, questions and answers, notes and files. Chatbots, AI forms, the content writer and automations can all read from it.
For a single chatbot, it’s usually simpler to teach it from the chatbot screen. Use the knowledge base when several tools share the same knowledge, or you want to see everything in one place.
Where it is
Section titled “Where it is”AI Puffer › Knowledge Base. It has four tabs: Data, Stores, Search and Settings.
Stores
Section titled “Stores”A store is where knowledge is kept. Create one under Stores › Create store, or while adding a source.
| Store | Kept where | You need |
|---|---|---|
| Local (called Built-in in the chatbot) | Your WordPress database | An embedding model, such as AI Puffer Cloud’s. Up to 10,000 pieces per store and 25,000 per site. |
| OpenAI | OpenAI’s vector stores | Your OpenAI key. OpenAI does the embedding. |
| Google’s file search stores | Your Google key. Only used by tools running on a Google model. | |
| Pinecone, Qdrant, Chroma | Your own vector database | Connect it under Settings › Connections, plus an embedding model. |
Add knowledge
Section titled “Add knowledge”On Data, select Add source:
- Q&A: a question visitors may ask and the answer to give.
- Text: anything you paste.
- Website: every published item of the content types you tick, such as posts, pages and products. Select Sync.
- Files Pro: PDF, Word, TXT, Markdown, CSV or JSON.
Under Destination, pick the store and, where needed, the embedding model. Then Add source.
You can also add posts from the WordPress lists: on Posts, Pages or Products, tick items and select Add to knowledge base. An Index Status column shows what’s already in.
Keep it current
Section titled “Keep it current”- Update on a website source reads it again; Edit on a text or Q&A source changes it and indexes it again.
- To keep everything current automatically, create a Build a knowledge base automation.
- Retry failed and Delete work on several sources at once.
Search on your site
Section titled “Search on your site”Search sets up a search box for visitors that finds content by meaning, not just matching words. Choose the store, the embedding model, how many results and the “no results” text, then add the shortcode:
[aipkit_semantic_search]It works with Local, Pinecone, Qdrant and Chroma stores.
Settings
Section titled “Settings”- Chunking Pro: how text is cut into pieces before it’s stored.
- Batches Pro: how many pieces are embedded per request.
- Content rules Pro: which fields, custom fields, taxonomies and product data go into each content type’s knowledge.
- Visibility: whether files visitors share in chats appear under Data.