# Operade Agents

An Operade agent is a versioned, text-based AI configuration that answers conversations inside one workspace. Start in the dashboard for a guided setup, or use the public API when your own system needs to create and inspect agents.

## Choose a starting path

- Follow the [dashboard quickstart](/docs/agents/quickstart) to create, test, and publish an agent in Agent Studio.
- Follow the [API quickstart](/docs/agents/quickstart-api) to create an agent with a workspace API key and read it back.
- Read [core concepts](/docs/agents/core-concepts) before designing a versioning or knowledge workflow.

Each new agent starts with a Main branch and an immutable initial version. Editing creates a draft; publishing creates another immutable version. Publishing Main updates the version used as the agent's active configuration.

## What you can configure

Agent Studio exposes the system prompt, first message, language, model, temperature when supported, maximum conversation duration, default personality setting, and the built-in `end_call` tool. A model can appear in the selector without having an active production adapter; check [models](/docs/agents/models) before testing. Other configuration fields can exist in the API even when the dashboard has no dedicated control.

Knowledge is managed at workspace level and then attached to an agent. Attached documents must be ready before the agent can publish. See [Knowledge overview](/docs/knowledge) for the full flow.

## A safe operating loop

1. Create an agent and edit its draft.
2. Use [test chat](/docs/agents/test-chat) with the draft selected.
3. Review the changes and publish Main.
4. Start a fresh test conversation without the draft to verify the published version.
5. Archive the agent when it should no longer be used.

Publishing and archiving have different effects. Publishing advances an immutable history. Archiving hides the agent from the normal list while preserving its versions and conversations. Read [agent lifecycle](/docs/agents/agent-lifecycle) before archiving.