Choose what to build next

You have completed a successful Turn. Add only the Capabilities your product needs, using the choices below to move from an outcome to its canonical explanation and first practical Guide.

Customize Agent behavior

Models and Providers

Choose this when you need a different model or your own Provider.

Prompts

Choose this when named, reusable message input should replace literal text.

Bring your own Provider

Configure and verify a Provider, then attach it to an Agent.

Choose this path when the Agent's model selection or reusable input needs to change.

Add tools and knowledge

Tools

Choose built-in operations that the Agent may call during a Turn.

Skills

Add a progressively loaded multi-file capability.

MCP connections

Expose tools from an external MCP server.

Memory

Persist Agent-owned notes across Sessions.

Create and attach a Skill

Build the first reusable Skill extension.

Connect an MCP server

Configure an external tool source.

Add durable Memory

Store and retrieve durable Agent recall.

Build a coding Agent

Apply built-in Tools to a bounded file task in a Sandbox.

Choose this path when the Agent must take actions, load reusable instructions, reach external services, or recall notes across Sessions.

Work with files

Sandboxes

Choose a durable execution environment for file-backed work.

File operations

Enable the supported file and shell operations.

Artifacts

Publish selected immutable files from a Turn.

Configure file operations

Attach a Sandbox and verify files persist across Turns.

Publish and download Artifacts

Deliver a generated file through your backend.

Choose this path when work must survive beyond one Turn or a generated file must become a product output.

Produce typed results

Structured output

Choose this when downstream code needs data constrained by JSON Schema.

Generate structured output

Build and verify one typed generation flow.

Choose this path when application code needs structured data instead of conversational text.

Run work in the background

Tasks and schedules

Choose this for on-demand or scheduled asynchronous work.

Run a background Task

Create an on-demand Task and inspect its result.

Schedule recurring work

Configure and verify recurring execution.

Choose this path when work should continue independently of an interactive application request.

Operate in production

Tenancy and end-user Attribution

Separate tenant ownership from end-user reporting dimensions.

Usage and quotas

Measure Turns and apply tenant-set consumption ceilings.

Security and credentials

Protect credentials and understand trust boundaries.

Limits and reliability

Plan for supported limits and failure behavior.

Build a multi-tenant application

Apply tenant isolation and end-user Attribution in an application.

Monitor usage and quotas

Query usage and verify quota settings.

Troubleshoot a failed integration

Diagnose stable errors and operational failures.

Choose this path before exposing the integration to production traffic or when adding operational controls.

Integration: Build a chat endpoint
Extensions: Create and attach a Skill
Files: Configure file operations
Automation: Run a background Task

Reference

Use Reference when you need exact methods, inputs, or endpoints: