Build agents
A short orientation for builders. These pages explain how building agents in Sema4.ai works — enough to understand each feature, not a step-by-step tutorial. Here are the building blocks you'll work with, and how they fit together:
- Agent types — a Conversational or Worker agent does the work. Everything below configures it.
- Runbooks — the agent's instructions in plain English: its objective, how to use its tools and data, and the steps to follow.
- Versioning & lifecycle — edit a draft, test it, publish it live, and roll back.
- Tools with MCP — how the agent takes action, via MCP servers (vendor, Sema4.ai gallery, or your own).
- Data connections — link the workspace to your databases and files so agents can use them.
- Semantic data models — let the agent query that data in natural language.
- Data frames — tabular data the agent works with, from files or query results.
- Document Intelligence — extract structured data from PDFs and images.
- Files — the agent's file space for inputs, working files, and outputs.
- Conversation guide — starter prompts that show users what to ask.
- Evals — measure whether the agent does its job correctly, and catch regressions.
- Advanced settings — tune the model, reasoning, and execution limits when the defaults aren't enough.
Once agents are live, Analytics and the Audit log show usage and a full evidence trail.