markitdown-pro 2.1.2__tar.gz → 2.2.0__tar.gz

markitdown_pro-2.2.0/.github/workflows/.github/workflows/issues_add_to_ksg_project.yaml

@@ -0,0 +1,32 @@
+name: Add new issues to VIVI project
+
+# When an issue is opened in this repo:
+#   1. Apply the `VIVI` label.
+#   2. Add the issue to the org project (Kinetic-Solutions-Group / project #1).
+#
+# Step 2 targets an *organization* project, which the default GITHUB_TOKEN
+# cannot write to. It requires a PAT stored as the `ADD_ISSUE_TO_PROJECT_PAT`
+# secret with project read/write scope. See the repo README / setup notes.
+
+on:
+  issues:
+    types: [opened]
+
+permissions:
+  issues: write
+
+jobs:
+  add-to-vivi-project:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Apply MKDWN label
+        env:
+          GH_TOKEN: ${{ github.token }}
+          GH_REPO: ${{ github.repository }}
+        run: gh issue edit "${{ github.event.issue.number }}" --add-label "MKDWNPRO"
+
+      - name: Add issue to VIVI project
+        uses: actions/add-to-project@5afcf98fcd03f1c2f92c3c83f58ae24323cc57fd # v2.0.0
+        with:
+          project-url: https://github.com/orgs/Kinetic-Solutions-Group/projects/1
+          github-token: ${{ secrets.ADD_ISSUE_TO_PROJECT_PAT }}

{markitdown_pro-2.1.2 → markitdown_pro-2.2.0}/.github/workflows/test.yml

@@ -36,10 +36,14 @@ jobs:
         run: uv run pytest tests/unit/ -v
 
   integration-tests:
-    name: Integration Tests (Azure + Gotenberg)
+    name: Integration Tests (${{ matrix.provider }} + Gotenberg)
     runs-on: ubuntu-latest
     environment: ci
     needs: unit-tests
+    strategy:
+      fail-fast: false
+      matrix:
+        provider: [azure_openai, openai]
     services:
       gotenberg:
         image: gotenberg/gotenberg:8
@@ -74,8 +78,10 @@ jobs:
           cat <<EOF > .test.env
           DEBUG=True
           LOG_LEVEL=10
-          AZURE_OPENAI_ENDPOINT=${{ secrets.AZURE_OPENAI_ENDPOINT }}
-          AZURE_OPENAI_API_KEY=${{ secrets.AZURE_OPENAI_API_KEY }}
+          OPEN_AI_PROVIDER=${{ matrix.provider }}
+          OPENAI_API_KEY=${{ secrets.OPENAI_API_KEY }}
+          AZURE_OPENAI_ENDPOINT=${{ secrets.AZURE_OPEN_AI_EASTUS2_ENDPOINT }}
+          AZURE_OPENAI_API_KEY=${{ secrets.AZURE_OPEN_AI_EASTUS2_KEY }}
           AZURE_OPENAI_API_VERSION=${{ secrets.AZURE_OPENAI_API_VERSION }}
           AZURE_DOCINTEL_ENDPOINT=${{ secrets.AZURE_DOCINTEL_ENDPOINT }}
           AZURE_DOCINTEL_KEY=${{ secrets.AZURE_DOCINTEL_KEY }}
@@ -91,8 +97,10 @@ jobs:
           cat <<EOF > .env
           DEBUG=True
           LOG_LEVEL=10
-          AZURE_OPENAI_ENDPOINT=${{ secrets.AZURE_OPENAI_ENDPOINT }}
-          AZURE_OPENAI_API_KEY=${{ secrets.AZURE_OPENAI_API_KEY }}
+          OPEN_AI_PROVIDER=${{ matrix.provider }}
+          OPENAI_API_KEY=${{ secrets.OPENAI_API_KEY }}
+          AZURE_OPENAI_ENDPOINT=${{ secrets.AZURE_OPEN_AI_EASTUS2_ENDPOINT }}
+          AZURE_OPENAI_API_KEY=${{ secrets.AZURE_OPEN_AI_EASTUS2_KEY }}
           AZURE_OPENAI_API_VERSION=${{ secrets.AZURE_OPENAI_API_VERSION }}
           AZURE_DOCINTEL_ENDPOINT=${{ secrets.AZURE_DOCINTEL_ENDPOINT }}
           AZURE_DOCINTEL_KEY=${{ secrets.AZURE_DOCINTEL_KEY }}
@@ -117,5 +125,5 @@ jobs:
         if: always()
         uses: actions/upload-artifact@v4
         with:
-          name: test-results
+          name: test-results-${{ matrix.provider }}
           path: tests/data/

{markitdown_pro-2.1.2 → markitdown_pro-2.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: markitdown-pro
-Version: 2.1.2
+Version: 2.2.0
 Summary: A package that converts almost any file format to Markdown.
 Author: Developer
 License-Expression: MIT
@@ -32,7 +32,7 @@ Description-Content-Type: text/markdown
 
 # MarkItDown-Pro
 
-**MarkItDown-Pro** is a Python library that converts **50+ document formats into Markdown**, built to power RAG (Retrieval-Augmented Generation) pipelines for semantic search. It extends [Microsoft MarkItDown](https://github.com/microsoft/markitdown) with Azure AI services, per-page OCR, and customizable converter pipelines.
+**MarkItDown-Pro** is a Python library that converts **50+ document formats into Markdown**, built to power RAG (Retrieval-Augmented Generation) pipelines for semantic search. It extends [Microsoft MarkItDown](https://github.com/microsoft/markitdown) with Azure AI and OpenAI services, per-page OCR, and customizable converter pipelines.
 
 ---
 
@@ -41,7 +41,7 @@ Description-Content-Type: text/markdown
 - **Async-first API** -- all public methods are `async`, designed for concurrent document processing
 - **Per-page PDF routing** -- classifies each page as text or image, extracts text locally and OCRs only image pages
 - **Customizable pipelines** -- inject your own converter order per handler to optimize for quality, speed, or cost
-- **GPT Vision OCR** -- concurrent page-by-page OCR via Azure OpenAI (gpt-5.4-mini default)
+- **GPT Vision OCR** -- concurrent page-by-page OCR via Azure OpenAI (default) or OpenAI (gpt-5.4-mini default model)
 - **Gotenberg integration** -- convert Office files to PDF for full OCR via [Gotenberg](https://gotenberg.dev) HTTP API
 - **Azure Document Intelligence** -- layout-aware text extraction with the `prebuilt-layout` model
 - **Azure Speech-to-Text** -- audio transcription with automatic language detection
@@ -117,7 +117,10 @@ uv sync --dev
 Create a `.env` file in the project root:
 
 ```bash
-# Azure OpenAI (required for GPT Vision OCR)
+# GPT Vision OCR provider (optional; defaults to azure_openai)
+OPEN_AI_PROVIDER="azure_openai"
+
+# Azure OpenAI credentials (used only when the provider is azure_openai)
 AZURE_OPENAI_ENDPOINT="https://<resource>.openai.azure.com"
 AZURE_OPENAI_API_KEY="your-key"
 AZURE_OPENAI_API_VERSION="2024-12-01-preview"
@@ -144,7 +147,33 @@ MARKITDOWN_MIN_IMAGE_AREA="150000"
 LOG_LEVEL=20  # 10=DEBUG, 20=INFO, 30=WARNING
 ```
 
-All services are optional -- the library degrades gracefully when credentials are missing.
+To use OpenAI instead of Azure OpenAI for OCR, set:
+
+```bash
+OPEN_AI_PROVIDER="openai"
+OPENAI_API_KEY="your-openai-key"
+# Optional; defaults to the public OpenAI API
+OPENAI_BASE_URL="https://api.openai.com/v1"
+MARKITDOWN_OCR_MODEL="gpt-5.4-mini"
+MARKITDOWN_OCR_FALLBACK_MODEL="gpt-5.4"
+```
+
+`OPEN_AI_PROVIDER` accepts exactly `azure_openai` or `openai`. Omitting it preserves
+Azure behavior and existing `AZURE_OPENAI_*` names. Invalid values raise `ValueError`;
+credentials never implicitly switch the provider. OpenAI does not require Azure credentials
+or an Azure API version. Set these variables before creating converters; the OCR model
+variables are read when the library is imported.
+
+Direct OpenAI uses `OPENAI_API_KEY` and the optional `OPENAI_BASE_URL`.
+
+`MARKITDOWN_OCR_MODEL` and `MARKITDOWN_OCR_FALLBACK_MODEL` identify Azure deployments or
+OpenAI model IDs, depending on the selected provider. Choose vision-capable models available
+to your account. Both the primary and fallback model use the same selected provider.
+This applies to image OCR, scanned PDF pages, and Office files routed through PDF OCR.
+Azure Speech, Document Intelligence, and Gotenberg configuration are unchanged.
+
+All services are optional -- missing selected-provider credentials leave GPT Vision unavailable
+so other converters can still run; there is no automatic fallback to the other provider.
 
 ## Usage
 

{markitdown_pro-2.1.2 → markitdown_pro-2.2.0}/README.md

@@ -1,6 +1,6 @@
 # MarkItDown-Pro
 
-**MarkItDown-Pro** is a Python library that converts **50+ document formats into Markdown**, built to power RAG (Retrieval-Augmented Generation) pipelines for semantic search. It extends [Microsoft MarkItDown](https://github.com/microsoft/markitdown) with Azure AI services, per-page OCR, and customizable converter pipelines.
+**MarkItDown-Pro** is a Python library that converts **50+ document formats into Markdown**, built to power RAG (Retrieval-Augmented Generation) pipelines for semantic search. It extends [Microsoft MarkItDown](https://github.com/microsoft/markitdown) with Azure AI and OpenAI services, per-page OCR, and customizable converter pipelines.
 
 ---
 
@@ -9,7 +9,7 @@
 - **Async-first API** -- all public methods are `async`, designed for concurrent document processing
 - **Per-page PDF routing** -- classifies each page as text or image, extracts text locally and OCRs only image pages
 - **Customizable pipelines** -- inject your own converter order per handler to optimize for quality, speed, or cost
-- **GPT Vision OCR** -- concurrent page-by-page OCR via Azure OpenAI (gpt-5.4-mini default)
+- **GPT Vision OCR** -- concurrent page-by-page OCR via Azure OpenAI (default) or OpenAI (gpt-5.4-mini default model)
 - **Gotenberg integration** -- convert Office files to PDF for full OCR via [Gotenberg](https://gotenberg.dev) HTTP API
 - **Azure Document Intelligence** -- layout-aware text extraction with the `prebuilt-layout` model
 - **Azure Speech-to-Text** -- audio transcription with automatic language detection
@@ -85,7 +85,10 @@ uv sync --dev
 Create a `.env` file in the project root:
 
 ```bash
-# Azure OpenAI (required for GPT Vision OCR)
+# GPT Vision OCR provider (optional; defaults to azure_openai)
+OPEN_AI_PROVIDER="azure_openai"
+
+# Azure OpenAI credentials (used only when the provider is azure_openai)
 AZURE_OPENAI_ENDPOINT="https://<resource>.openai.azure.com"
 AZURE_OPENAI_API_KEY="your-key"
 AZURE_OPENAI_API_VERSION="2024-12-01-preview"
@@ -112,7 +115,33 @@ MARKITDOWN_MIN_IMAGE_AREA="150000"
 LOG_LEVEL=20  # 10=DEBUG, 20=INFO, 30=WARNING
 ```
 
-All services are optional -- the library degrades gracefully when credentials are missing.
+To use OpenAI instead of Azure OpenAI for OCR, set:
+
+```bash
+OPEN_AI_PROVIDER="openai"
+OPENAI_API_KEY="your-openai-key"
+# Optional; defaults to the public OpenAI API
+OPENAI_BASE_URL="https://api.openai.com/v1"
+MARKITDOWN_OCR_MODEL="gpt-5.4-mini"
+MARKITDOWN_OCR_FALLBACK_MODEL="gpt-5.4"
+```
+
+`OPEN_AI_PROVIDER` accepts exactly `azure_openai` or `openai`. Omitting it preserves
+Azure behavior and existing `AZURE_OPENAI_*` names. Invalid values raise `ValueError`;
+credentials never implicitly switch the provider. OpenAI does not require Azure credentials
+or an Azure API version. Set these variables before creating converters; the OCR model
+variables are read when the library is imported.
+
+Direct OpenAI uses `OPENAI_API_KEY` and the optional `OPENAI_BASE_URL`.
+
+`MARKITDOWN_OCR_MODEL` and `MARKITDOWN_OCR_FALLBACK_MODEL` identify Azure deployments or
+OpenAI model IDs, depending on the selected provider. Choose vision-capable models available
+to your account. Both the primary and fallback model use the same selected provider.
+This applies to image OCR, scanned PDF pages, and Office files routed through PDF OCR.
+Azure Speech, Document Intelligence, and Gotenberg configuration are unchanged.
+
+All services are optional -- missing selected-provider credentials leave GPT Vision unavailable
+so other converters can still run; there is no automatic fallback to the other provider.
 
 ## Usage
 

{markitdown_pro-2.1.2 → markitdown_pro-2.2.0}/markitdown_pro/services/openai_services.py

@@ -1,5 +1,5 @@
 """
-Asynchronous PDF → image → OCR pipeline using Azure OpenAI (Vision via Chat Completions).
+Asynchronous PDF → image → OCR pipeline using Azure OpenAI or OpenAI (Vision via Chat Completions).
 
 Key features
 ------------
@@ -37,10 +37,11 @@ from pathlib import Path
 from typing import TypeVar
 
 import fitz
+import httpx
 
 # from docling.datamodel.pipeline_options import PictureDescriptionApiOptions
 from langchain_core.messages import BaseMessage, HumanMessage, SystemMessage
-from langchain_openai import AzureChatOpenAI
+from langchain_openai import AzureChatOpenAI, ChatOpenAI
 from PIL import Image as PILImage
 from pydantic import SecretStr
 
@@ -54,13 +55,19 @@ T = TypeVar("T")
 _PNG_DPI = 150
 _DEFAULT_CONTENT_TYPE = "image/jpeg"
 _RETRY_HTTP_STATUSES = (408, 409, 429, 500, 502, 503, 504)
+_EXCEPTION_CHAIN_LIMIT = 5
+_OPENAI_HTTP_LIMITS = httpx.Limits(
+    max_connections=1000,
+    max_keepalive_connections=100,
+    keepalive_expiry=5.0,
+)
 
 
 class GPTVision:
     """
-    High-level async wrapper around AzureChatOpenAI for image OCR of PDF pages.
+    High-level async wrapper around Azure OpenAI or OpenAI for image OCR of PDF pages.
 
-    This class renders PDF pages to PNG, sends them to an Azure OpenAI vision model,
+    This class renders PDF pages to PNG, sends them to the selected provider's vision model,
     and returns markdown text per page. It is designed for high-throughput,
     robust processing with timeouts and retries.
     """
@@ -80,9 +87,9 @@ class GPTVision:
         Parameters
         ----------
         model_name : str
-            Azure OpenAI deployment name (e.g., "gpt-4.1-mini").
+            Azure OpenAI deployment name or OpenAI model ID (e.g., "gpt-4.1-mini").
         api_version : str
-            Azure OpenAI API version (e.g., "2024-09-01-preview").
+            Azure OpenAI API version; ignored for OpenAI.
         completion_tokens : int
             Max completion token budget per call (output-side).
         max_concurrency : int
@@ -94,28 +101,59 @@ class GPTVision:
         page_timeout_s : float
             Hard timeout for a single page end-to-end (wrapping process_image).
         """
-        azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT", "")
-        if not azure_endpoint:
-            logger.error("GPTVision: AZURE_OPENAI_ENDPOINT environment variable is not set.")
-            self.lang_client = None
-            return
+        self._llm = None
+        self.lang_client = None
+        self._http_client: httpx.Client | None = None
+        self._http_async_client: httpx.AsyncClient | None = None
+
+        provider = os.getenv("OPEN_AI_PROVIDER", "azure_openai")
+        if provider not in {"azure_openai", "openai"}:
+            raise ValueError("OPEN_AI_PROVIDER must be 'azure_openai' or 'openai'.")
+
+        if provider == "azure_openai":
+            azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT", "")
+            if not azure_endpoint:
+                logger.error("GPTVision: AZURE_OPENAI_ENDPOINT environment variable is not set.")
+                return
+            key_variable = "AZURE_OPENAI_API_KEY"
+            api_key = os.getenv(key_variable, "")
+        else:
+            key_variable = "OPENAI_API_KEY"
+            api_key = os.getenv(key_variable, "")
+            base_url = os.getenv("OPENAI_BASE_URL") or "https://api.openai.com/v1"
 
-        api_key = os.getenv("AZURE_OPENAI_API_KEY", "")
         if not api_key:
-            logger.error("GPTVision: AZURE_OPENAI_API_KEY environment variable is not set.")
-            self.lang_client = None
+            logger.error(f"GPTVision: {key_variable} environment variable is not set.")
             return
 
         try:
-            self._llm = AzureChatOpenAI(
-                azure_endpoint=azure_endpoint,
-                api_key=SecretStr(api_key),
-                api_version=api_version,
-                model=model_name,
-                model_kwargs={"max_completion_tokens": completion_tokens},
-                temperature=0,
-                cache=False,
-            )
+            http_client_options = {
+                "limits": _OPENAI_HTTP_LIMITS,
+                "follow_redirects": True,
+            }
+            self._http_client = httpx.Client(**http_client_options)
+            self._http_async_client = httpx.AsyncClient(**http_client_options)
+            client_options = {
+                "api_key": SecretStr(api_key),
+                "model": model_name,
+                "model_kwargs": {"max_completion_tokens": completion_tokens},
+                "temperature": 0,
+                "cache": False,
+                "http_client": self._http_client,
+                "http_async_client": self._http_async_client,
+                "openai_proxy": "",
+            }
+            if provider == "azure_openai":
+                self._llm = AzureChatOpenAI(
+                    azure_endpoint=azure_endpoint,
+                    api_version=api_version,
+                    **client_options,
+                )
+            else:
+                self._llm = ChatOpenAI(
+                    base_url=base_url,
+                    **client_options,
+                )
             # In this version we use the raw LLM response (string markdown),
             # not structured output. Keep a separate client if you later switch.
             self.lang_client = self._llm
@@ -134,13 +172,26 @@ class GPTVision:
 
             logger.info("GPTVision: GPT-vision client initialized successfully.")
         except Exception as e:
-            logger.warning(f"GPTVision: Failed to initialize GPT-vision client: {e}")
+            self._close_clients_after_initialization_failure()
+            logger.warning(
+                f"GPTVision: Failed to initialize {provider} client ({type(e).__name__})."
+            )
             self.lang_client = None
             self._semaphore = asyncio.Semaphore(1)
             self.request_timeout_s = 60.0
             self.acquire_timeout_s = 30.0
             self.page_timeout_s = 120.0
 
+    def _close_clients_after_initialization_failure(self) -> None:
+        """Close the sync transport and release the unused async transport reference."""
+        http_client = self._http_client
+        self._http_client = None
+        self._http_async_client = None
+
+        if http_client is not None:
+            with contextlib.suppress(Exception):
+                http_client.close()
+
     # -------------------------
     # Async offloading helpers
     # -------------------------
@@ -276,13 +327,35 @@ class GPTVision:
                 raise last_err
             cap = min(max_delay, base_delay * (2 ** (attempt - 1)))
             sleep_for = random.uniform(0, cap)
-            label = context_label or file_or_url
             logger.warning(
-                f"GPTVision: ainvoke transient error {label}: attempt {attempt}/{max_retries}: "
-                f"{type(last_err).__name__}: {last_err}. Retrying in {sleep_for:.2f}s"
+                f"GPTVision: ainvoke transient error: attempt {attempt}/{max_retries}: "
+                f"error_chain={self._format_exception_chain(last_err)}. "
+                f"Retrying in {sleep_for:.2f}s"
             )
             await asyncio.sleep(sleep_for)
 
+    @staticmethod
+    def _format_exception_chain(error: BaseException) -> str:
+        """Return bounded transport diagnostics without exception payload text."""
+        parts: list[str] = []
+        seen: set[int] = set()
+        current: BaseException | None = error
+
+        while current is not None and len(parts) < _EXCEPTION_CHAIN_LIMIT:
+            identity = id(current)
+            if identity in seen:
+                break
+            seen.add(identity)
+
+            name = type(current).__name__
+            errno = getattr(current, "errno", None)
+            if isinstance(errno, int) and not isinstance(errno, bool):
+                name = f"{name}(errno={errno})"
+            parts.append(name)
+            current = current.__cause__ or current.__context__
+
+        return " -> ".join(parts)
+
     @staticmethod
     def _is_transient_error(e: Exception, retry_http_statuses: tuple[int, ...]) -> bool:
         """
@@ -300,6 +373,7 @@ class GPTVision:
                 "temporarily unavailable",
                 "timeout",
                 "timed out",
+                "connection error",
                 "connection reset",
             )
         ):
@@ -394,7 +468,8 @@ class GPTVision:
             return None
         except Exception as e:
             logger.error(
-                f"GPTVision: process_image: Error during GPT-vision image OCR for {file_or_url}: {e}"
+                "GPTVision: process_image: Error during GPT-vision image OCR: "
+                f"error_chain={self._format_exception_chain(e)}"
             )
             return None
 
@@ -517,34 +592,21 @@ class GPTVision:
 
     async def aclose(self) -> None:
         """
-        Idempotently close any async/sync clients created under the hood so they
-        don't try to close after the event loop is gone (pytest teardown).
+        Idempotently close the async/sync transports owned by this GPTVision.
 
-        LangChain's AzureChatOpenAI lazily creates httpx AsyncClient instances.
-        We close them explicitly here to prevent 'Event loop is closed' errors
-        during garbage collection.
+        GPTVision creates these httpx clients explicitly and closes them before
+        the event loop is gone to prevent teardown errors during garbage collection.
         """
-        if not hasattr(self, "_llm") or self._llm is None:
-            return
-
-        # Close the async OpenAI client (wraps httpx.AsyncClient)
+        # Close only the transports created and owned by this GPTVision instance.
         with contextlib.suppress(Exception):
-            async_client = getattr(self._llm, "async_client", None)
-            if async_client and hasattr(async_client, "close"):
-                await async_client.close()
+            if self._http_async_client is not None:
+                await self._http_async_client.aclose()
 
-        # Close the sync OpenAI client
         with contextlib.suppress(Exception):
-            client = getattr(self._llm, "client", None)
-            if client and hasattr(client, "close"):
-                client.close()
+            if self._http_client is not None:
+                self._http_client.close()
 
-        # Close httpx clients buried inside the OpenAI SDK
-        with contextlib.suppress(Exception):
-            async_client = getattr(self._llm, "async_client", None)
-            http_client = getattr(async_client, "_client", None)
-            if http_client and hasattr(http_client, "aclose"):
-                await http_client.aclose()
+        self.lang_client = None
 
 
 # def azure_openai_vlm_options(