Capabilities
Capabilities explain platform behavior and the decisions available to you. Use Guides for outcome-oriented procedures and the TypeScript SDK or REST API Reference for exact contracts.
An Agent is configuration, a Turn is one execution, and a Session is stored conversation state.
Choose what you want to do
Configure your Agent
Set Agent behavior, models, reusable input, and lifecycle.
Run your Agent
Work with Sessions, Turns, generation, approvals, and Tasks.
Extend your Agent
Add Tools, Skills, MCP connections, and Memory.
Work with files and outputs
Use Sandboxes, file operations, and Artifacts.
Apply platform controls
Manage tenancy, usage, credentials, limits, and reliability.
Configure your Agent
Configure your Agent covers persisted Agent identity and behavior, model credentials, reusable Prompt input, and configuration history.
Run your Agent
Run your Agent explains stateful and stateless execution, streaming, structured output, Tool approvals, and asynchronous work.
Extend your Agent
Extend your Agent covers the capabilities an Agent can load or call while it works.
Work with files and outputs
Files and outputs explains durable Sandboxes, file operations, and deliberately published Artifacts.
Apply platform controls
Platform controls covers Tenant isolation, end-user Attribution, usage, quotas, credentials, and operational boundaries.
Related guides
Use Guides when you want a complete procedure for an integration or operational outcome.
Reference
Use the TypeScript SDK Reference for exported client methods and types, or the REST API Reference for endpoints and wire contracts.