Create a knowledge base

Create a knowledge base, choose its embedding model, and understand what it costs and why that choice is fixed.

What you will achieve

You will have an empty knowledge base ready for documents.

Step-by-step setup

Open Knowledge in the sidebar.
Choose New base (or Create your first base on an empty page).
In New knowledge base, enter a Name, for example Product documentation, and an optional Description such as "Refunds, shipping and support hours".
Choose an Embedding model. The list is grouped by provider and shows the price per million tokens.
Confirm. The base opens, ready for documents.

Configuration details

What embedding is. Phynxa turns each passage of your documents into numbers that capture its meaning. At call time the caller's question is turned into numbers the same way and matched against the passages.
The embedding model cannot be changed once anything is indexed, because questions and passages must come from the same model. To switch, create a new base and add the documents again. The model shows in the Embedding column and under Settings.
Choices. OpenAI's text-embedding-3-small (the cheapest and a strong default), text-embedding-3-large (most accurate OpenAI model) and text-embedding-ada-002 (older), and Google's gemini-embedding-2. Prices are shown in the list.
Cost. Each document is embedded once when it is added, and each question a caller asks is embedded during the call. The Total spent column on the Knowledge page shows the running total per base.

Cards show Knowledge bases, Documents, Used by agents and Failed documents. The table lists each base with Status, Embedding, Total spent, Documents, Passages, Agents and Updated.

Testing

Add a document, then use Test retrieval (see Search and settings).

Expected result

A new base appears in the list with 0 documents. Its page says "Nothing indexed yet".

Troubleshooting

A notice says a key is missing for the chosen model. That embedding provider is not available on the platform; pick another model or contact us.
A base shows a "retired model". Older bases made with removed local models keep working, but new bases cannot use them.

Something here does not match what you see in the dashboard? Let us know.