Agents & the Runner Loop

Understanding the agent loop is key to building effective agents with the OpenAI Agents SDK. The Runner orchestrates a cycle of model calls, tool executions, and handoffs until the agent produces a final output.

The Agent Loop

User Message → Model Call → [Tool Call?] → [Handoff?] → Final Output
                    ↑            |
                    └────────────┘ (loop back if tool was called)

Each iteration:

1. The model receives the conversation history and decides what to do
2. If it calls a tool, the Runner executes the function and feeds the result back
3. If it triggers a handoff, control passes to another agent
4. If it produces a text response (no tool calls), the loop ends

Configuring an Agent

from agents import Agent
 
agent = Agent(
    name="Research Assistant",
    instructions="""You are a research assistant. When asked about a topic:
    1. Search for relevant information using your tools
    2. Synthesize the findings
    3. Present a clear, structured summary""",
    model="gpt-4o",
)

Agent Properties

Runner Options

from agents import Runner
 
result = Runner.run_sync(
    agent,
    "Analyze the latest AI trends",
    max_turns=10,
)

• max_turns — limits how many loop iterations the agent can take (prevents infinite loops)
• The runner automatically manages conversation state across turns

Inspecting Results

result = Runner.run_sync(agent, "Hello")
 
print(result.final_output)     # The agent's final text response
print(result.last_agent)       # The agent that produced the final output (useful with handoffs)
 
for item in result.new_items:
    print(item)                # All messages, tool calls, and tool results

Key Takeaways

• The Runner manages the agent loop: model call → tool execution → repeat
• The loop ends when the model produces a text response without tool calls
• max_turns prevents runaway loops
• Results include the full trace of messages and tool calls