Getting Started with Knowledge Bases

Sema4.ai Knowledge Bases allow agents to work with enterprise knowledge that lives outside of traditional databases — like documents, internal guides, FAQs, and chat transcripts. This page gives you a high-level view of how to go from zero to a fully functional agent that can reason over that knowledge.

How It Works: Knowledge Base Lifecycle

Create a vector storage target (e.g., pgvector)

Set up a vector-capable database to store embeddings. This is where your Knowledge Base will live.

Define your Knowledge Base (Using SDK)

Choose models, map columns, connect to storage

Insert data into the Knowledge Base

This triggers embedding and semantic indexing

Query your Knowledge Base

Use semantic + metadata filtering to retrieve relevant info

Package your queries into Action Package

Publish queries as Actions to Studio

Use the Knowledge Base in an agent

Add actions with queries to your agent and use them in the runbooks

Deploy your agent with the Knowledge Base

Fully integrated and ready for Control Room

How It All Comes Together

Here’s the lifecycle of working with a Knowledge Base:

This documentation is organized around this lifecycle:

Build a Knowledge Base→Define, add data, and test your Knowledge Base using the Sema4.ai SDK Use Knowledge Base in an Agent→Connect Knowledge Base to your agent and test Deploy Agent with Knowledge Base→Make Agents that use Knowledge Base accessible by your team

Prerequisites

Before you start, make sure you have this installed these extensions in your VS Code or Cursor.
- Sema4.ai SDK
- Sema4.ai Data Access
You must have access to Embedding models and API keys for those models.
A vector-capable database (e.g., PostgreSQL with pgvector extension) must be set up and accessible.

Knowledge Base Build a Knowledge Base