Human hand-off

Humans sometimes need to talk to an other human instead of an AI, yeah I know sounds weird right? But turns out this could be quite beneficial for customer satisfaction. Superagent enables this by offering a specific tool we call Human hand-off.

Step-by-step guide

Prerequisites: This tutorial expects you to have configured an LLM and created an Agent

Create a HAND_OFF tool. This tool will make sure to trigger and ask the user if they want to speak to a human.

1 tool = client.tool.create(
2     name="Human hand-off",
3     description="Useful when the user wants to speak to a human operator",
4     type="HAND_OFF"
5 )
6 
7 # Add the tool to your agent
8 client.agent.add_tool(agent_id=agent.data.id, tool_id=tool.data.id)

Invoke the Agent to test the hand-off

1 # Write a helper function that checks for human hand-off
2 def check_for_human_handoff(data: list) -> bool:
3     for item, _ in data:
4         if item['tool'] == 'human-hand-off':
5             return True
6     return False
7 
8 prediction = client.agent.invoke(
9     agent_id=agent.data.id,
10     enable_streaming=False,
11     input="I want to speak to a human",
12     session_id="my_session_id"
13 )
14 
15 output = prediction.data.get("output")
16 steps = prediction.data.get("intermediate_steps")
17 
18 if check_for_human_handoff(steps):
19     # Run any local code here
20     print("HUMAN HANDOFF DETECTED")

That’s it, you can now run arbitrary code whenever the human hand-off is triggered.

Full code

1 tool = client.tool.create(
2     name="Human hand-off",
3     description="Useful when the user wants to speak to a human operator",
4     type="HAND_OFF"
5 )
6 
7 # Add the tool to your agent
8 client.agent.add_tool(agent_id=agent.data.id, tool_id=tool.data.id)
9 
10 # Write a helper function that checks for human hand-off
11 def check_for_human_handoff(data: list) -> bool:
12     for item, _ in data:
13         if item['tool'] == 'human-hand-off':
14             return True
15     return False
16 
17 prediction = client.agent.invoke(
18     agent_id=agent.data.id,
19     enable_streaming=False,
20     input="I want to speak to a human",
21     session_id="my_session_id"
22 )
23 
24 output = prediction.data.get("output")
25 steps = prediction.data.get("intermediate_steps")
26 
27 if check_for_human_handoff(steps):
28     # Run any local code here
29     print("HUMAN HANDOFF DETECTED")

Overview

Getting Started

SAML

Installation

RAG & Retrieval

Logging

Vector Stores

Memory

Dependencies

Step-by-step guide

Full code

1	tool = client.tool.create(
2	name="Human hand-off",
3	description="Useful when the user wants to speak to a human operator",
4	type="HAND_OFF"
5	)
6
7	# Add the tool to your agent
8	client.agent.add_tool(agent_id=agent.data.id, tool_id=tool.data.id)

1	# Write a helper function that checks for human hand-off
2	def check_for_human_handoff(data: list) -> bool:
3	for item, _ in data:
4	if item['tool'] == 'human-hand-off':
5	return True
6	return False
7
8	prediction = client.agent.invoke(
9	agent_id=agent.data.id,
10	enable_streaming=False,
11	input="I want to speak to a human",
12	session_id="my_session_id"
13	)
14
15	output = prediction.data.get("output")
16	steps = prediction.data.get("intermediate_steps")
17
18	if check_for_human_handoff(steps):
19	# Run any local code here
20	print("HUMAN HANDOFF DETECTED")