Microsoft Foundry¶

Survey Scribe can use a Microsoft Foundry model endpoint only when that endpoint implements the Azure OpenAI-compatible Chat Completions contract used by AsyncAzureOpenAI.

Microsoft Foundry and Palantir Foundry are different platforms. See the separate Palantir Foundry deployment assessment.

Supported boundary¶

The Azure adapter constructs this request shape:

{azure_endpoint}/openai/deployments/{deployment}/chat/completions
    ?api-version={api_version}

Supported configuration includes:

One HTTPS endpoint.
One deployment or route name.
One API version.
An Azure API key or a synchronous token callback.
Strict Instructor tool output.
Package-owned bounded retries and concurrency.

The package does not support a Responses-only endpoint, deploy a model, create a project, assign roles, configure network access, manage quota, discover managed identity, or call a health route.

All facade-generated model capabilities are configuration-only. They permit request validation but do not prove live endpoint access, model limits, or quality.

Install¶

Use an approved distribution with the openai extra:

python -m pip install "survey-scribe[openai]"

Survey Scribe 0.1.x is alpha software, with published releases available on PyPI. See Installation for package and pinned-wheel options.

API-key quick start¶

Keep credentials outside source and TOML:

import os
from pathlib import Path

from survey_scribe import ResultStatus, SurveyScribe

with SurveyScribe(
    provider="azure_openai",
    model="<approved-deployment-name>",
    base_url="https://resource.example/",
    api_version="<approved-api-version>",
    api_key=os.environ["AZURE_OPENAI_API_KEY"],
) as client:
    result = client.convert(Path("questionnaire.docx"))

if result.status is not ResultStatus.success:
    raise RuntimeError("Send questionnaire extraction to review")

The endpoint must use HTTPS and must not contain user information, a fragment, or a recognized secret query parameter.

Host-managed token quick start¶

The host application owns identity and token acquisition:

from pathlib import Path

from survey_scribe import SurveyScribe


def acquire_token() -> str:
    return application_token_source()


async def extract_questionnaire():
    async with SurveyScribe(
        provider="azure_openai",
        model="<approved-deployment-name>",
        base_url="https://resource.example/",
        api_version="<approved-api-version>",
        token_callback=acquire_token,
    ) as client:
        return await client.aconvert(Path("questionnaire.docx"))

application_token_source() is application code. Survey Scribe does not import Azure Identity, select a scope, or call the token callback during construction. The Azure SDK calls it when authentication is required.

Use a refreshable callback for managed runtimes. A configured bearer_token becomes a callback that always returns the same static value.

Direct provider construction¶

Construct AzureOpenAIProvider directly when an administrator must provide an exact ModelCapabilities row or approved gateway headers. This is required for the generic mAI Factory pattern.

Do not combine an injected provider with facade-level model, endpoint, credential, API-version, callback, or environment-resolution arguments. Use a non-secret SurveyScribeConfig only for generation, retry, confidence, concurrency, and artifact settings.

Production checklist¶

Confirm that the endpoint implements Azure Chat Completions for the selected API version.
Confirm the exact deployment name and strict structured-output behavior.
Supply an administrator-owned capability row when generic limits are not sufficient.
Use host-managed key or token storage.
Set endpoint egress and data-retention policy outside Survey Scribe.
Run an offline SDK contract test with the exact headers and route.
Run one authorized live test before you label a model row verified.
Inspect success, partial, and failed results before artifact publication.

Operational limits¶

SDK retries are disabled. Survey Scribe owns the retry budget.
Authentication and header-callback failures are not retried.
Rate limits, timeouts, connection failures, HTTP 408, and HTTP 5xx can be retried.
There is no public provider request-timeout setting.
There is no jitter, Retry-After handling, circuit breaker, or process-global limiter.
Configured provider and model names are reported. The adapter does not capture a dynamic backend model identity from the response.

See AI Providers, Configuration, and Provider Contracts.