Agents

Agents wrap a completion model with reusable behavior and a bounded runtime loop. They can combine instructions, tools, memory, knowledge, hooks, and observability across one or more model turns.

Explore agents

Page	Learn how to
Build an agent	Create an agent and send its first prompt.
Stable behavior	Decide what belongs on the builder or in a scoped factory.
Instructions	Define durable policy, role, tone, and workflow rules.
Context	Supply static facts, retrieved knowledge, and request state.
Per-run controls	Configure and execute one prompt request.
Runtime lifecycle	Understand the model, tool, memory, and event sequence.
Errors and limits	Bound runs and map failures into safe product responses.

Agent lifecycle

agent defaults → prompt request → model turn → optional tools → final response

The builder defines behavior shared by every run. The prompt request supplies one input and any tighter run-specific controls. Anvia owns the model-and-tool loop; the application still owns authentication, permissions, services, persistence policy, and the response exposed to users.

Use a direct completion when the workflow needs only one model call and no agent orchestration.