OpenAI Agent

OpenAI agent is developed from previous prototype Swarm.

• Key concepts OpenAI Agent framework is LLM centric, with memeory and tools.

• Agent

• Handoffs

• Guardrails

• Tracing

• General Agent

• Voice Agent

• Comparison with other agent frameworks

SDK

• Runner

result = await Runner.run(agent, input)

• Agent

agent = Agent.from_tools(tools)

Internal workflow of Runner

LLM centric approact,

In the center of the agent, we have a loop. It runs the LLM, check the output:

• Is it a stop signal? a final output (i.e. the agent produces something of type agent.output_type, the loop terminates.
• Is it a tool call? call tools, and submit the tool call result to the LLM.
• Is it a handoff? if yes, run agent b

Hooks

Telemetry

• workflow
• Tasks
• agents
• tools
• guardrails
• handoffs