Build Overview

Everything you build in Agent Studio is assembled from a small set of blocks.

Block	What it does	Start here
Tools	Actions grounded in your catalog, data products, and databases	Tools overview
Agents	LLM + prompt + tools, combined to complete a task	Agents overview
Flows	Multi-step processes chaining agents and tools	Flows overview
Schedules	Flows that run on a cadence, without anyone asking	Scheduling flows
Evals	Q&A sets that measure agent accuracy before and after launch	SQL evaluations
Models	The LLM behind your agents: OpenAI, Anthropic, or your own	LLM configuration

A typical build path

1. Check whether the built-in tools cover your data sources; create custom tools for anything they don’t.
2. Clone the closest built-in agent and adjust its prompt and tools.
3. Build an evaluation set from real user questions and iterate until accuracy holds.
4. If the task is multi-step or recurring, wrap the agent in a flow and schedule it.
5. Publish, then pick a channel in Integrations. Every built-in tool call is metered — see usage and metering.