Concepts

Agents are autonomous systems that understand natural language and use tools and reasoning to accomplish tasks. They maintain conversation context through sessions and can handle multi-turn interactions.

To build your first agent, see the Agent quick start. You can also try our Agent API tutorial notebook. For more information about our Agent APIs see the API Reference.

How agents work

An agent consists of these primary components that define its behavior and capabilities.

• Instructions serve as the system prompt for the agent and define its behavior, how to respond to user input, and which tools to use.
• Tools provide agents with capabilities to interact with data and external systems including corpora search, web search, artifacts (uploaded files), and other built-in tools.
• Sub-agents enable you to create modular agent systems. The parent agent can delegate domain-specific tasks to an existing agent or agents. This sub-agent then executes its own reasoning loop and returns the final result to the parent agent.
• Sessions preserve context throughout a conversation so the agent can consider all prior information when responding to a query.
• Artifacts define uploaded documents and files within a session for the agent to analyze without indexing them into a permanent corpus. You can also generate new documents from these artifacts and upload to a corpus for future use.

To chat with your agent, read on about Sessions.