Knowledge bases
A knowledge base is your documents in which AI searches for answers: a price list, delivery terms, instructions, answers to frequently asked questions. The model knows nothing about your company until you give it this information. With a knowledge base, it first finds the relevant fragments of documents and then answers based on them.
Only the fragments found go into the request to the model, not the whole documents. That is why a base can be large while the responses stay inexpensive.
The section is in the project menu — the "Knowledge bases" item. Answers are searched in it by the AI chat, the "AI Agent" block and third-party services.
How the section works
The "Knowledge bases" list has all the project's bases. Open a base, and it has its own menu:
| Item | What it contains |
|---|---|
| "Documents" | The base's documents, their state, token usage and the paid-until date |
| "Search testing" | Which fragments the base finds for a query |
| "Statistics" | Token usage, the number of requests and the consumption log |
| "Settings" | The name, the plan and the auto-update schedule |
| "Connecting to third-party services" | The address of the base's MCP server |
| "AI help" | The AI chat where you can ask to add documents to the base or delete them — see Connect a base to AI |
What you need
- A paid base plan. The base is paid monthly from the account balance. Prices are on the Prices of additional services page.
- A Pro subscription is not needed.
- The project owner creates the base. Team members work with a ready base within their role.
Create a base
- In the project menu, open "Knowledge bases" and click "Create knowledge base".
- Fill in the form:
- "Name" — what the base will be called in the list and in the AI agent settings.
- "Primary language" — the language of your documents. Word search depends on it: the platform takes into account the word forms of this language. If the language is not in the list, choose "Other".
- "Embedding model" — the model that turns text into numbers for search by meaning. Under the field is the "What is embedding" hint describing each model, including where the text is sent during processing.
- "Plan" — the volume: 10, 30 or 50 million tokens per month.
- Click "Create". The plan is paid immediately and in full.
If there is not enough money on the balance, the "Insufficient funds" window opens with the top-up amount.
Important
The primary language and the embedding model are chosen once — at creation. They cannot be changed for an existing base.
Next, add documents to the base and test the search.
Limits
- One project can have up to 100 knowledge bases, and no more than 24 can be created per day.
- Scheduled auto-update can be turned on for no more than 12 documents in one base.
If something does not work
- The "Create knowledge base" button is unavailable. The project already has 100 bases: "A project can have no more than 100 knowledge bases". If there is no button at all, you are not the project owner.
- "No more than 24 knowledge bases can be created per day." The daily limit has been reached. Create the base the next day.
Pages of the section
- Documents — document types, file formats, processing parameters, document states and scheduled auto-update.
- Plan and tokens — what tokens are spent on, changing the plan, renewal and an inactive base.
- Search testing and statistics — which fragments the base finds, search parameters, statistics and the consumption log.
- Connect a base to AI — the "AI Agent" block, searching and changing the base from the AI chat, third-party services via MCP.
- Settings, list of bases and access — base settings, the order of bases in the list, team member permissions.