Agent Builder
Create intelligent agents, your way.
Agent Builder is a development environment for designing AI agents and workflows. You define an agent's purpose, tools, information sources, behavior and workflow, then publish it for Agent Run to execute.
What an agent is
An agent combines a language model with an execution loop and tools. It receives a task, retrieves information, calls APIs, evaluates the results and continues until it produces an answer or reaches a stopping condition.
Agent Builder makes these elements easy to configure. You connect components visually, set their parameters, test the result and expose it to the rest of the platform. It serves developers and technical business teams who understand their processes and want to spend less time on repetitive implementation work.
What you configure
| Capability | What you configure | Business value |
|---|---|---|
| Agent instructions | Purpose, responsibilities, constraints and response style | Makes intended behavior explicit |
| Model configuration | Model provider, model and supported generation settings | Matches model capabilities to the task |
| Tool connections | APIs, database queries, business functions and MCP servers | Lets agents work with business systems |
| Knowledge retrieval | Documents, search indexes and retrieval rules | Grounds answers in relevant information |
| Workflow design | Steps, branches, loops, parallel paths and dependencies | Represents complex business processes |
| Multi-agent composition | Specialist agents, responsibilities and handoffs | Divides work across focused capabilities |
| Human review | Approval points and requests for missing information | Keeps people involved in consequential decisions |
| Testing and publishing | Test cases, input examples, versions and release controls | Makes changes reviewable and repeatable |
Connected to your systems
Agents reach business systems through APIs, database queries, business functions and Model Context Protocol (MCP) servers. MCP standardizes how an AI application connects to external tools and context sources, while the platform manages credentials, permissions, connection lifecycle and tool-specific validation around it.
Example: an operations investigation agent
A user configures an agent that:
- Retrieves operational records.
- Compares current performance against a reference period.
- Investigates anomalies.
- Prepares a report.
Agent Builder captures the instructions and workflow that govern this process, so the same investigation can run again with consistent behavior.
What you publish
The result is a versioned agent or workflow definition. It records the selected execution framework, the configuration, tool references, input and output contracts, and execution limits.
Agent Run supports two frameworks: LangGraph / LangChain and Strands Agents. Agent Builder shows which features each framework supports, because a configuration that works in one framework may require different behavior in the other.
Why Agent Builder
- Design agents and workflows visually.
- Connect business tools and knowledge.
- Reuse configurable components.
- Test and version every change.
- Choose the supported execution framework.
Turn business processes into reusable AI applications with less repetitive development work.