Knowledge Bases
Knowledge Bases are vector stores in your workspace. You index documents and records, then agents retrieve them for grounded answers (RAG), or you search and manage vectors from the knowledge base detail page and from workflows.
Use them when assistants need accurate answers from product docs, FAQs, policies, or tagged records — without pasting everything into the prompt.
When to use
- Ground an agent on docs or FAQs with semantic search
- Keep one vector store that agents, workflows, and MCP can reuse
- Enforce metadata on every upsert so retrieval can stay filtered
Providers
When you create a knowledge base, pick a Service — the vector-store provider behind it:
| Provider | What happens |
|---|---|
| Your own | Connect Pinecone (or another vector store) under Integrations → Third Party Services, then choose it. You set the Namespace yourself. |
| Built-in | Use Vettero’s managed vector store. It is billed against your plan. The namespace is created automatically — you do not enter one. |
Service (and namespace) are fixed after create.
Create a knowledge base
Open Data → Knowledge Bases and create a knowledge base. Fill in:
| Setting | What it does |
|---|---|
| Name / Description | How it appears in lists and when attaching to agents |
| Service | Your own connected vector store, or the built-in (billed) provider |
| Namespace | Required for your own provider; auto-created for built-in |
| Read Only | You can search and read vectors, but not upsert (or empty the store) |
| Strict | Every upsert must include metadata that matches your schema and is validated |
With Strict on, define Meta Fields (name, type as text / number / boolean, optional values and required). Structured datasources only are available when Strict is enabled.
Manage vectors in the detail page
Open a knowledge base to work with its data:
| Tab / action | What you do |
|---|---|
| Content | Add, refresh, or delete datasources (files, URLs, scrapes, YouTube). |
| Query | Search the store (by text and optional metadata filters). |
| Upsert | Add or update a single vector by id and text (with metadata). Hidden or blocked when Read Only. |
| Settings | Export Vectors (JSONL), or Delete All Vectors to empty the store (not available when Read Only). |
Content sources
| Kind | What you add |
|---|---|
| Structured | CSV, JSONL, Excel, or text — or a URL — with rows and metadata |
| Unstructured | Documents (PDF, DOCX, Markdown, and more), URLs, YouTube transcripts, or a website scrape |
Uploads are limited to 25MB per file. Strict knowledge bases only accept structured content.
From the Content tab you can also:
- Refresh a datasource — re-process it and update its vectors in the store
- Delete a datasource — removes the datasource and deletes its vectors from the knowledge base (other datasources are unchanged)
Strict and Read Only
- Strict — metadata is required on upsert and must pass validation against your meta fields. Use this when every record must carry consistent tags (product, locale, and so on).
- Read Only — search and query only; upsert and delete-all are not allowed.
Attach to an agent (RAG)
In the agent builder, open the RAG panel:
- Create New or Attach Existing knowledge base.
- Choose a RAG Strategy:
- Agentic — the model calls the knowledge base as a tool when it needs context (default)
- 2-Step — Vettero retrieves matching chunks before the turn and adds them to context
- Hybrid — both
- Set Top K — how many chunks to retrieve (default 3).
You can also mention the knowledge base in the prompt with @kb:..., or expose it as a tool on the Tools tab.
Use in workflows
In the workflow builder, use knowledge base blocks under the kb folder — Query, Upsert, Delete, and datasource helpers — to search or update the same store from automation.
Limits
- Built-in providers consume vector-store plan credits; your own provider uses your connected service.
- Service and namespace cannot change after create.
- Strict requires meta fields and validates metadata on upsert.
- Read Only allows read/search only — no upsert or delete-all.
- File uploads max 25MB.
Where knowledge bases are used
| Place | How |
|---|---|
| Agents | RAG panel (strategy + Top K) or a knowledge base tool |
| Workflows | Knowledge base blocks (query, upsert, delete, datasources) |
| MCP Servers | Expose knowledge bases on an MCP server as query tools |