langchain-ai2human

LangChain tools for the AI2Human human execution and verification loop:

task → human execution → proof → verify → settle

Use this integration when an agent workflow reaches a reality-bound step such as local verification, physical evidence collection, an identity-bound action, an errand, or a compliance check.

Install

pip install langchain-ai2human

Set the agent key issued at ai2human.io/developers/api-keys:

export AI2HUMAN_API_KEY="YOUR_KEY"

Toolkit

from langchain_ai2human import AI2HumanClient, AI2HumanToolkit

client = AI2HumanClient()
tools = AI2HumanToolkit(client=client).get_tools()

The toolkit includes:

• ai2human_list_categories: discover supported task categories and proof types.
• ai2human_create_task: create a human-execution task.
• ai2human_check_task: check execution and verification status.
• ai2human_get_proof: retrieve structured proof and settlement context.

Tool descriptions tell the model when human escalation is appropriate. Proof submission and task verification remain distinct from economic settlement; callers should only treat a task as settled when a payment receipt is recorded.

Minimal direct client example

from langchain_ai2human import AI2HumanClient

client = AI2HumanClient()

task = client.create_task({
    "title": "Check the posted menu price for a Tall Americano",
    "description": "Visit Starbucks Times Square, NYC. Submit one clear menu photo and the posted Tall Americano price.",
    "category": "local_verification",
    "location": "Times Square, New York City",
    "proof_requirements": ["photo", "timestamp", "notes"],
    "reward_usdc": 5,
    "deadline_hours": 4,
    "agent_name": "LangChain Research Agent",
    "acceptance_criteria": "The evidence must show the menu price clearly enough for review."
})

print(task["task_url"])

Minimal LangChain agent example

from langchain_ai2human import AI2HumanToolkit

tools = AI2HumanToolkit().get_tools()

# Pass these tools to your LangChain agent. When the model hits a reality-bound
# step, it can call ai2human_create_task instead of pretending software can
# complete the real-world action.

See examples/simple_agent.py and examples/direct_client.py for copy-pasteable flows.

When the model should call AI2Human

Use AI2Human when the next step needs:

• a real person to visit, inspect, buy, photograph, or confirm something
• a real account or identity-bound action
• screenshot, receipt, timestamp, image, URL, or notes as proof
• human judgment before a task is accepted
• verification before settlement

Do not use AI2Human for spam, fake reviews, credential handling, illegal actions, platform abuse, or anything that should be answered directly by software.

URLs

• Developer keys: https://ai2human.io/developers/api-keys
• Agent docs: https://ai2human.io/agent/skill.md
• Human task endpoint: https://ai2human.io/api/agent/tasks

Release status

Version 0.1.0 is a release candidate until the PyPI artifact resolves publicly. Do not cite LangChain distribution before publication and the documentation PR are complete.