09_human_in_the_loop.py

# Copyright (c) 2025 Agentspan
# Licensed under the MIT License. See LICENSE file in the project root for details.

"""Human-in-the-Loop — approval workflows.

Demonstrates how tools with approval_required=True pause the workflow
until a human approves or rejects the action.  Uses interactive streaming
with schema-driven console prompts to handle the HITL pause.

Requirements:
    - Conductor server with LLM support
    - AGENTSPAN_SERVER_URL=http://localhost:8080/api as environment variable
    - AGENTSPAN_LLM_MODEL=openai/gpt-4o-mini as environment variable
"""

from conductor.ai.agents import Agent, AgentRuntime, EventType, tool
from settings import settings


@tool
def check_balance(account_id: str) -> dict:
    """Check the balance of an account."""
    return {"account_id": account_id, "balance": 15000.00}


@tool(approval_required=True)
def transfer_funds(from_acct: str, to_acct: str, amount: float) -> dict:
    """Request a funds transfer; runtime pauses for human approval before execution."""
    return {"status": "completed", "from": from_acct, "to": to_acct, "amount": amount}


agent = Agent(
    name="banker",
    model=settings.llm_model,
    tools=[check_balance, transfer_funds],
    instructions=(
        "You are a banking assistant. Use check_balance for balance inquiries. "
        "When asked to transfer money, first check the balance, then call "
        "transfer_funds to request the transfer. The runtime will pause for "
        "human approval before the transfer executes."
    ),
)


if __name__ == "__main__":
    with AgentRuntime() as runtime:
        handle = runtime.start(agent, "Transfer $500 from ACC-789 to ACC-456. Check the balance first.")
        print(f"Started: {handle.execution_id}\n")

        for event in handle.stream():
            if event.type == EventType.THINKING:
                print(f"  [thinking] {event.content}")

            elif event.type == EventType.TOOL_CALL:
                print(f"  [tool_call] {event.tool_name}({event.args})")

            elif event.type == EventType.TOOL_RESULT:
                print(f"  [tool_result] {event.tool_name} -> {str(event.result)[:100]}")

            elif event.type == EventType.WAITING:
                status = handle.get_status()
                pt = status.pending_tool or {}
                schema = pt.get("response_schema", {})
                props = schema.get("properties", {})
                print("\n--- Human input required ---")
                response = {}
                for field, fs in props.items():
                    desc = fs.get("description") or fs.get("title", field)
                    if fs.get("type") == "boolean":
                        val = input(f"  {desc} (y/n): ").strip().lower()
                        response[field] = val in ("y", "yes")
                    else:
                        response[field] = input(f"  {desc}: ").strip()
                handle.respond(response)
                print()

            elif event.type == EventType.DONE:
                print(f"\nDone: {event.output}")

        # Non-interactive alternative (no HITL, will block on human tasks):
        # result = runtime.run(agent, "What's the balance on ACC-789?")
        # result.print_result()

        # Production pattern:
        # 1. Deploy once during CI/CD:
        # runtime.deploy(agent)
        #
        # 2. In a separate long-lived worker process:
        # runtime.serve(agent)