Skip to content

README.md

Latest commit

 

History

History
228 lines (168 loc) · 5.8 KB

README.md

File metadata and controls

228 lines (168 loc) · 5.8 KB

Ask-a-Human Python SDK

Python SDK for integrating Ask-a-Human into your AI agents. Get human input when your agent is uncertain or needs subjective judgment.

Installation

pip install ask-a-human

Or install from source:

pip install -e path/to/sdk-python

Installing from Source with Virtual Environment

On modern systems (especially macOS with Homebrew), you'll need to use a virtual environment:

# Create and activate a virtual environment
python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install the SDK
pip install -e path/to/sdk-python

# Install with development dependencies (for running tests)
pip install -e "path/to/sdk-python[dev]"

Quick Start

from ask_a_human import AskHumanClient

# Create a client
client = AskHumanClient(agent_id="my-agent")

# Submit a question
result = client.submit_question(
    prompt="Should this error message apologize to the user or just state the facts?",
    type="text",
    audience=["product", "creative"],
    min_responses=5
)

print(f"Question submitted: {result.question_id}")

# Later, check for responses
response = client.get_question(result.question_id)

if response.status in ("CLOSED", "PARTIAL"):
    for r in response.responses:
        print(f"Human said: {r.answer} (confidence: {r.confidence})")

Using the Orchestrator

For more complex workflows, use the AskHumanOrchestrator which handles polling and timeouts:

from ask_a_human import AskHumanClient, AskHumanOrchestrator

client = AskHumanClient(agent_id="my-agent")
orchestrator = AskHumanOrchestrator(client, poll_interval=30.0)

# Submit a question
submission = orchestrator.submit(
    prompt="Which button label is clearer?",
    type="multiple_choice",
    options=["Submit", "Send", "Confirm", "Done"],
    min_responses=10
)

# Wait for responses (with timeout)
responses = orchestrator.await_responses(
    question_ids=[submission.question_id],
    min_responses=5,
    timeout=300  # 5 minutes
)

# Process responses
question = responses[submission.question_id]
print(f"Status: {question.status}")
print(f"Got {question.current_responses} responses")

if question.summary:
    print(f"Summary: {question.summary}")

Multiple Choice Questions

result = client.submit_question(
    prompt="What tone should this notification use?",
    type="multiple_choice",
    options=[
        "Formal and professional",
        "Friendly and casual",
        "Urgent and direct",
        "Neutral and informative"
    ],
    audience=["product"],
    min_responses=10
)

# Check responses
response = client.get_question(result.question_id)

# For multiple choice, responses have selected_option instead of answer
for r in response.responses:
    option_text = response.options[r.selected_option]
    print(f"Human chose: {option_text} (confidence: {r.confidence})")

# summary field shows vote counts
if response.summary:
    print(f"Vote distribution: {response.summary}")

Handling Timeouts and Partial Results

The orchestrator can return partial results when a timeout is reached:

responses = orchestrator.await_responses(
    question_ids=["q_abc123"],
    min_responses=10,
    timeout=60  # Short timeout
)

question = responses["q_abc123"]

if question.status == "PARTIAL":
    print(f"Got {question.current_responses} of {question.required_responses} responses")
    # Decide whether to proceed with partial results or wait longer
elif question.status == "EXPIRED":
    print("Question expired before getting enough responses")

Configuration

Environment Variables

  • ASK_A_HUMAN_BASE_URL - Override the API base URL (default: https://api.ask-a-human.com)
  • ASK_A_HUMAN_AGENT_ID - Default agent ID if not specified in constructor

Client Options

client = AskHumanClient(
    base_url="https://api.ask-a-human.com",  # API endpoint
    agent_id="my-agent",                      # Your agent identifier
    timeout=30.0                              # HTTP request timeout
)

Orchestrator Options

orchestrator = AskHumanOrchestrator(
    client=client,
    poll_interval=30.0  # Seconds between polls (uses exponential backoff)
)

Error Handling

from ask_a_human import AskHumanClient
from ask_a_human.exceptions import (
    AskHumanError,
    ValidationError,
    QuestionNotFoundError,
    RateLimitError,
    ServerError
)

client = AskHumanClient(agent_id="my-agent")

try:
    result = client.submit_question(prompt="...", type="text")
except ValidationError as e:
    print(f"Invalid request: {e}")
except RateLimitError as e:
    print(f"Rate limited: {e}")
except ServerError as e:
    print(f"Server error: {e}")
except AskHumanError as e:
    print(f"Unknown error: {e}")

API Reference

AskHumanClient

Low-level API client.

  • submit_question(prompt, type, options=None, audience=None, min_responses=5, timeout_seconds=3600, idempotency_key=None) - Submit a question
  • get_question(question_id) - Get question status and responses

AskHumanOrchestrator

High-level orchestration with polling and timeouts.

  • submit(prompt, **kwargs) - Submit a question (same args as client)
  • await_responses(question_ids, min_responses=1, timeout=3600) - Wait for responses
  • poll_once(question_ids) - Non-blocking status check

Types

  • QuestionType - "text" or "multiple_choice"
  • QuestionStatus - "OPEN", "PARTIAL", "CLOSED", or "EXPIRED"
  • QuestionSubmission - Response from submitting a question
  • QuestionResponse - Full question with status and responses
  • HumanResponse - Individual human response

Examples

See the examples/ directory for more usage examples:

  • basic_usage.py - Simple question submission and polling
  • multi_question.py - Managing multiple concurrent questions

License

MIT