Agents

Build a single agent: give it a model, roles and actions, then make it reliable. Read the pages in this order.

Build

Create an agent: AgentBuilder or an annotated class, then chat.
Principal and authority: the two settings every agent needs.
Roles and actions: what the agent can do.
Actions and tools: choose between an action, a @Tool POJO, a built-in toolkit and an MCP tool.
Capabilities: reuse a set of actions across roles.

Shape the output

Memory: conversation history and shared stores.
Structured output: parse replies into records.
Streaming and events: watch a turn as it runs.
Prompt strategies: build a prompt with reasoning instructions and send it yourself.
Variants: quality, speed and cost tiers that the framework stores but does not act on.

Make it reliable

Contracts: check an action's input and result.
Resilience: retries, timeouts and circuit breakers.
Exceptions: what an action call can throw.
Long runs: runs that outlive one request.
Schema, identity and enums: tool parameter schemas, identity and style, DIDs and core enums.

Multi-agent: when one agent is not enough.
Knowledge: ground chat and actions in your files.