Skip to content

API.md

Latest commit

 

History

History
114 lines (78 loc) · 2.9 KB

API.md

File metadata and controls

114 lines (78 loc) · 2.9 KB
Raw

Python API

Import

from openai_document_analyzer import DocumentAnalyzer

DocumentAnalyzer

load_text(filepath) -> str

Loads UTF-8 TXT or Markdown content, or extracts text from a PDF. Leading and trailing whitespace is removed.

Raises:

  • FileNotFoundError
  • DocumentReadError
  • DocumentTooLargeError
  • EmptyDocumentError

extract_text_from_pdf(filepath) -> str

Extracts nonempty PDF pages locally and separates them with a blank line. Image-only PDFs return EmptyDocumentError; run OCR before calling the package.

analyze_text(text, prompt=..., model=None) -> str

Sends text and an analysis request through the Responses API. The model argument overrides the configured default for one call.

Raises:

  • ConfigurationError for an empty prompt;
  • DocumentTooLargeError or EmptyDocumentError for invalid text;
  • AnalysisError when the OpenAI SDK raises an API error;
  • EmptyResponseError when the API provides no output text.

analyze_document(filepath, prompt=..., model=None) -> str

Combines load_text() and analyze_text().

ask_questions(prompt, example_prompt, example_response, text_to_analyze, model=None) -> str

Uses one example request and response as answer-shape guidance. Examples are not treated as evidence about the document. Both example values are required.

This method exists for version 1 migration. Prefer a direct, explicit prompt when an example is unnecessary.

Constants

DEFAULT_MODEL

The package default: gpt-5.6-sol.

MODEL_PROFILES

Documented Sol, Terra, and Luna model IDs with their intended tradeoff. It is not an allowlist.

Exception hierarchy

DocumentAnalyzerError
├── ConfigurationError
├── DocumentReadError
├── DocumentTooLargeError
├── EmptyDocumentError
└── AnalysisError
    └── EmptyResponseError

Catch the narrowest exception that the application can handle:

from openai_document_analyzer import (
    AnalysisError,
    DocumentAnalyzer,
    DocumentReadError,
)

analyzer = DocumentAnalyzer()

try:
    result = analyzer.analyze_document("report.pdf")
except DocumentReadError as exc:
    print(f"Document problem: {exc}")
except AnalysisError as exc:
    print(f"API problem: {exc}")

Programming errors and unexpected failures are intentionally not converted to plausible analysis text.

Client injection

from openai import OpenAI
from openai_document_analyzer import DocumentAnalyzer

client = OpenAI(timeout=45.0, max_retries=4)
analyzer = DocumentAnalyzer(client=client)

The injected object must provide client.responses.create(...) and return an object with output_text.

Sync behavior

Version 2 is synchronous. Applications requiring async execution should place the synchronous call in an appropriate worker or implement a dedicated AsyncOpenAI integration with equivalent request, privacy, and error behavior.