Adstract SDK for Python

Adstract integrates ad-enhanced prompts into LLM applications and provides the acknowledgment flow required to close the ad cycle after the final model response is produced.

Official Documentation

Full documentation is available at Adstract Documentation.

Install

pip install adstractai

Authentication

Pass an API key when initializing the client, or set the ADSTRACT_API_KEY environment variable.

export ADSTRACT_API_KEY="adpk_live_123"

from adstractai import Adstract

client = Adstract()

Quickstart

from adstractai import Adstract, AdRequestContext

client = Adstract(api_key="adpk_live_123")

result = client.request_ad(
    prompt="How do I improve analytics in my LLM app?",
    context=AdRequestContext(
        session_id="sess-1",
        user_agent=(
            "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 "
            "(KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36"
        ),
        user_ip="203.0.113.24",
    ),
)

prompt_for_model = result.prompt
llm_response = "Your final LLM response here"

ack = client.acknowledge(
    enhancement_result=result,
    llm_response=llm_response,
)

if ack is not None:
    print(ack.ad_ack_id)
    print(ack.status)
    print(ack.success)

client.close()

Core Flow

The SDK integration flow has two main steps:

1. Call request_ad() or request_ad_async() to get an EnhancementResult.
2. After your LLM produces its final response, call acknowledge() or acknowledge_async() to close the ad cycle.

EnhancementResult.prompt always gives you the prompt your application should use next:

• enhanced prompt when ad injection succeeds;
• original prompt when the SDK falls back.

Required Request Context

request_ad() and request_ad_async() require an AdRequestContext with:

• session_id
• user_agent
• user_ip

from adstractai.models import AdRequestContext

context = AdRequestContext(
    session_id="sess-1",
    user_agent="Mozilla/5.0 (X11; Linux x86_64)",
    user_ip="203.0.113.24",
)

Missing required values raise MissingParameterError, or are captured in EnhancementResult.error when raise_exception=False.

Optional Context

Pass OptionalContext to include optional targeting signals.

from adstractai import Adstract, AdRequestContext, OptionalContext

client = Adstract(api_key="adpk_live_123")

result = client.request_ad(
    prompt="How do I improve analytics in my LLM app?",
    context=AdRequestContext(
        session_id="sess-1",
        user_agent="Mozilla/5.0 (X11; Linux x86_64)",
        user_ip="203.0.113.24",
    ),
    optional_context=OptionalContext(
        country="US",
        region="California",
        city="San Francisco",
        asn=15169,
        age=30,
        gender="female",
    ),
)

Validation rules:

Enhancement Results

request_ad() and request_ad_async() return EnhancementResult.

Important fields:

• prompt: the prompt your application should pass to the model
• session_id: the request session identifier
• ad_response: parsed backend response when available
• success: whether ad enhancement succeeded
• error: captured failure when raise_exception=False

if result.success:
    prompt_for_model = result.prompt
else:
    print(result.error)
    prompt_for_model = result.prompt

Acknowledgment Results

acknowledge() and acknowledge_async() return AdAckResponse on successful acknowledgment.

AdAckResponse includes:

• ad_ack_id
• status
• success

success means the acknowledgment itself completed successfully:

• status="ok" -> success=True
• status="no_ad_used" -> success=True
• status="recoverable_error" -> success=False

If enhancement_result.success is False, acknowledgment is skipped and the method returns None.

Error Handling

By default, SDK methods raise on failure.

• request_ad(..., raise_exception=True)
• acknowledge(..., raise_exception=True)

Set raise_exception=False if you want a fallback-first integration flow.

Enhancement exceptions

from adstractai.errors import (
    AdEnhancementError,
    AuthenticationError,
    DuplicateAdRequestError,
    NoFillError,
    PromptRejectedError,
)

result = client.request_ad(
    prompt="My prompt",
    context=context,
    raise_exception=False,
)

if result.success:
    print(result.prompt)
elif isinstance(result.error, PromptRejectedError):
    print("Prompt not suitable for ad injection")
elif isinstance(result.error, NoFillError):
    print("No ad inventory available")
elif isinstance(result.error, DuplicateAdRequestError):
    print("This message already has an ad request")
elif isinstance(result.error, AuthenticationError):
    print("Authentication failed")
elif isinstance(result.error, AdEnhancementError):
    print(result.error)

Acknowledgment exceptions

from adstractai.errors import (
    AdResponseNotFoundError,
    AuthenticationError,
    DuplicateAcknowledgmentError,
    UnsuccessfulAdResponseError,
)

try:
    ack = client.acknowledge(
        enhancement_result=result,
        llm_response="Final response",
    )
except AuthenticationError:
    print("Authentication failed")
except AdResponseNotFoundError:
    print("The ad response was not found")
except UnsuccessfulAdResponseError:
    print("The ad response was not created by a successful enhancement")
except DuplicateAcknowledgmentError:
    print("This response was already acknowledged")

Wrapping Type

Control how ads are wrapped in the enhanced prompt. The default is "xml".

client = Adstract(api_key="adpk_live_123", wrapping_type="markdown")

Supported values:

• "xml"
• "plain"
• "markdown"

Async Usage

import asyncio

from adstractai import Adstract, AdRequestContext


async def main() -> None:
    client = Adstract(api_key="adpk_live_123")

    result = await client.request_ad_async(
        prompt="Need performance tips",
        context=AdRequestContext(
            session_id="sess-99",
            user_agent=(
                "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 "
                "(KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36"
            ),
            user_ip="192.0.2.1",
        ),
    )

    llm_response = "Your final LLM response here"

    ack = await client.acknowledge_async(
        enhancement_result=result,
        llm_response=llm_response,
    )

    if ack is not None:
        print(ack.ad_ack_id)

    await client.aclose()


asyncio.run(main())

Public API

License

This SDK is distributed under the Adstract SDK Proprietary License. See LICENSE.