NanoNets · Siddhant011 · Oct 14, 2025 · Oct 14, 2025
diff --git a/.gitignore b/.gitignore
@@ -199,4 +199,6 @@ ENV/
 .vscode/ 
 .playwright-mcp/
 
-examples/
+examples/
+
+venv1/
diff --git a/CLAUDE.md b/CLAUDE.md
@@ -6,9 +6,8 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 DocStrange is a Python library for extracting and converting documents (PDFs, Word, Excel, PowerPoint, images, URLs) into multiple formats (Markdown, JSON, CSV, HTML) with intelligent content extraction and advanced OCR capabilities.
 
-The library offers three processing modes:
+The library offers two processing modes:
 - **Cloud Mode (default)**: Instant conversion using cloud API
-- **CPU Mode**: Local processing for privacy
 - **GPU Mode**: Local processing with GPU acceleration
 
 ## Commands
@@ -117,16 +116,10 @@ python -m twine upload dist/*
 - Authentication: `docstrange login` or API key
 - Best for: Quick processing without GPU
 
-### CPU Mode
-- Force with `cpu=True` parameter
-- Uses local neural models
-- Requires model downloads (~500MB first run)
-- Best for: Privacy-sensitive documents
-
 ### GPU Mode  
 - Force with `gpu=True` parameter
 - Requires CUDA-compatible GPU
-- Faster than CPU for large documents
+- Fastest local processing
 - Best for: Batch processing, high-volume workloads
 
 ## Authentication & Rate Limits
@@ -215,9 +208,6 @@ structured = result.extract_data(json_schema=schema)
 
 ### Force local processing
 ```python
-# CPU mode
-extractor = DocumentExtractor(cpu=True)
-
 # GPU mode (requires CUDA)
 extractor = DocumentExtractor(gpu=True)
 ```

diff --git a/README.md b/README.md
@@ -38,7 +38,7 @@ DocStrange converts documents to Markdown, JSON, CSV, and HTML quickly and accur
 > Extract documents data instantly with the cloud processing - no complex setup needed 
 
 > **🔒 Local Processing !**  
-> Use `cpu` or `gpu` mode for 100% local processing - no data sent anywhere, everything stays on your machine.
+> Use `gpu` mode for 100% local processing - no data sent anywhere, everything stays on your machine.
 
 
 ## **What's New**
@@ -56,7 +56,7 @@ Convert and extract data from PDF, DOCX, images, and more into clean Markdown an
 
 `DocStrange` is a Python library for converting a wide range of document formats—including **PDF**, **DOCX**, **PPTX**, **XLSX**, and **images** — into clean, usable data. It produces LLM-optimized **Markdown**, structured **JSON** (with schema support), **HTML**, and **CSV** outputs, making it an ideal tool for preparing content for RAG pipelines and other AI applications.
 
-The library offers both a powerful cloud API and a 100% private, offline mode that runs locally on your CPU or GPU. Developed by **Nanonets**, DocStrange is built on a powerful pipeline of OCR and layout detection models and currently requires **Python >=3.8**.
+The library offers both a powerful cloud API and a 100% private, offline mode that runs locally on your GPU. Developed by **Nanonets**, DocStrange is built on a powerful pipeline of OCR and layout detection models and currently requires **Python >=3.8**.
 
 **To report a bug or request a feature, [please file an issue](https://github.com/NanoNets/docstrange/issues). To ask a question or request assistance, please use the [discussions forum](https://github.com/NanoNets/docstrange/discussions).**
 
@@ -185,12 +185,9 @@ print(structured_data)
 
 **Local Processing**
 
-For complete privacy and offline capability, run DocStrange entirely on your own machine. You can specify whether to use your CPU or GPU for processing.
+For complete privacy and offline capability, run DocStrange entirely on your own machine using GPU processing.
 
 ```python
-# Force local CPU processing
-extractor = DocumentExtractor(cpu=True)
-
 # Force local GPU processing (requires CUDA)
 extractor = DocumentExtractor(gpu=True)
 ```
@@ -201,7 +198,7 @@ extractor = DocumentExtractor(gpu=True)
 
 💡 Want a GUI? Run the simple, drag-and-drop local web interface for private, offline document conversion.
 
-For users who prefer a graphical interface, DocStrange includes a powerful, self-hosted web UI. This allows for easy drag-and-drop conversion of PDF, DOCX, and other files directly in your browser, with 100% private, offline processing on your own CPU or GPU. The interface automatically downloads required models on its first run.
+For users who prefer a graphical interface, DocStrange includes a powerful, self-hosted web UI. This allows for easy drag-and-drop conversion of PDF, DOCX, and other files directly in your browser, with 100% private, offline processing on your own GPU. The interface automatically downloads required models on its first run.
 
 ### How to get started?
 
@@ -230,9 +227,9 @@ python -c "from docstrange.web_app import run_web_app; run_web_app()"
 
 - 🖱️ Drag & Drop Interface: Simply drag files onto the upload area.
 - 📁 Multiple File Types: Supports PDF, DOCX, XLSX, PPTX, images, and more.
-- ⚙️ Processing Modes: Choose between Local CPU and Local GPU processing.
+- ⚙️ Processing Modes: Choose between Cloud and Local GPU processing.
 - 📊 Multiple Output Formats: Get Markdown, HTML, JSON, CSV, and Flat JSON.
-- 🔒 100% Local Processing: No data leaves your machine.
+- 🔒 Privacy Options: Choose between cloud processing (default) or local GPU processing.
 - 📱 Responsive Design: Works on desktop, tablet, and mobile
 
 ### **Supported File Types:**
@@ -245,9 +242,8 @@ python -c "from docstrange.web_app import run_web_app; run_web_app()"
 
 ### **Processing Modes:**
 
-- **Local CPU**: Works offline, slower but private (default)
+- **Cloud processing:** For instant, zero-setup conversion, you can head over to [docstrange.nanonets.com](http://docstrange.nanonets.com/) **—** no setup (default)
 - **Local GPU**: Fastest local processing, requires CUDA support
-- **Cloud processing:** For instant, zero-setup conversion, you can head over to [docstrange.nanonets.com](http://docstrange.nanonets.com/) **—** no setup
 
 ### **Output Formats:**
 
@@ -295,7 +291,7 @@ docstrange web --port 8001
 
 - The interface automatically detects GPU availability
 - GPU option will be disabled if CUDA is not available
-- CPU mode will be selected automatically
+- Error will be thrown
 
 3. Model Download Issues:
 
@@ -367,13 +363,6 @@ csv_data = result.extract_csv()
 print(csv_data)
 ```
 
-**Requirements for enhanced JSON (if using cpu=True):**
-
-- Install: `pip install 'docstrange[local-llm]'`
-- [Install Ollama](https://ollama.ai/) and run: `ollama serve`
-- Pull a model: `ollama pull llama3.2`
-
-*If Ollama is not available, the library automatically falls back to the standard JSON parser.*
 
 **c. Extract Specific Fields & Structured Data**
 
@@ -484,11 +473,6 @@ contract_schema = {
 contract_data = contract.extract_data(json_schema=contract_schema)
 ```
 
-**Local extraction requirements (if using cpu=True):**
-
-- Install ollama package: `pip install 'docstrange[local-llm]'`
-- [Install Ollama](https://ollama.ai/) and run: `ollama serve`
-- Pull a model: `ollama pull llama3.2`
 
 **e. Chain with LLM**
 
@@ -591,7 +575,6 @@ docstrange document.pdf
 docstrange document.pdf --api-key YOUR_API_KEY
 
 # Local processing modes
-docstrange document.pdf --cpu-mode
 docstrange document.pdf --gpu-mode
 
 # Different output formats
@@ -629,8 +612,6 @@ docstrange document.pdf --output json --extract-fields title author date summary
 # Or use API key for 10k docs/month access (alternative to login)
 docstrange document.pdf --api-key YOUR_API_KEY --output json --extract-fields title author date summary
 
-# Force local processing with field extraction (requires Ollama)
-docstrange document.pdf --cpu-mode --output json --extract-fields key_points conclusions recommendations
 ```
 
 **Example schema.json file:**

diff --git a/docstrange/cli.py b/docstrange/cli.py
@@ -182,9 +182,6 @@ def main():
   # Convert with free API key with increased limits
   docstrange document.pdf --api-key YOUR_API_KEY
 
-  # Force local CPU processing
-  docstrange document.pdf --cpu-mode
-
   # Force local GPU processing  
   docstrange document.pdf --gpu-mode
 
@@ -207,10 +204,10 @@ def main():
   # Convert multiple files
   docstrange file1.pdf file2.docx file3.xlsx --output markdown
 
-  # Extract specific fields using Ollama (CPU mode only) or cloud
+  # Extract specific fields using cloud processing
   docstrange invoice.pdf --output json --extract-fields invoice_number total_amount vendor_name
 
-  # Extract using JSON schema (Ollama for CPU mode, cloud for default mode)
+  # Extract using JSON schema with cloud processing
   docstrange document.pdf --output json --json-schema schema.json
 
   # Save output to file
@@ -242,12 +239,6 @@ def main():
     )
 
     # Processing mode arguments
-    parser.add_argument(
-        "--cpu-mode",
-        action="store_true",
-        help="Force local CPU-only processing (disables cloud mode)"
-    )
-
     parser.add_argument(
         "--gpu-mode", 
         action="store_true",
@@ -280,12 +271,12 @@ def main():
     parser.add_argument(
         "--extract-fields",
         nargs="+",
-        help="Extract specific fields using Ollama (CPU mode) or cloud (default mode) (e.g., --extract-fields invoice_number total_amount)"
+        help="Extract specific fields using cloud processing (e.g., --extract-fields invoice_number total_amount)"
     )
 
     parser.add_argument(
         "--json-schema",
-        help="JSON schema file for structured extraction using Ollama (CPU mode) or cloud (default mode)"
+        help="JSON schema file for structured extraction using cloud processing"
     )
 
     parser.add_argument(
@@ -361,7 +352,6 @@ def main():
         extractor = DocumentExtractor(
             api_key=args.api_key,
             model=args.model,
-            cpu=args.cpu_mode,
             gpu=args.gpu_mode
         )
         print_supported_formats(extractor)
@@ -404,12 +394,11 @@ def main():
     extractor = DocumentExtractor(
         api_key=args.api_key,
         model=args.model,
-        cpu=args.cpu_mode,
         gpu=args.gpu_mode
     )
 
     if args.verbose:
-        mode = "local" if (args.cpu_mode or args.gpu_mode) else "cloud"
+        mode = "local" if args.gpu_mode else "cloud"
         print(f"Initialized extractor in {mode} mode:")
         print(f"  - Output format: {args.output}")
         if mode == "cloud":
@@ -418,8 +407,7 @@ def main():
             if args.model:
                 print(f"  - Model: {args.model}")
         else:
-            processor_type = "GPU" if args.gpu_mode else "CPU"
-            print(f"  - Local processing: {processor_type}")
+            print(f"  - Local processing: GPU")
         print()
 
     # Process inputs

diff --git a/docstrange/extractor.py b/docstrange/extractor.py
@@ -34,7 +34,6 @@ def __init__(
         ocr_enabled: bool = True,
         api_key: Optional[str] = None,
         model: Optional[str] = None,
-        cpu: bool = False,
         gpu: bool = False
     ):
         """Initialize the file extractor.
@@ -45,35 +44,28 @@ def __init__(
             ocr_enabled: Whether to enable OCR for image and PDF processing
             api_key: API key for cloud processing (optional). Prefer 'docstrange login' for 10k docs/month; API key from https://app.nanonets.com/#/keys is an alternative
             model: Model to use for cloud processing (gemini, openapi) - only for cloud mode
-            cpu: Force local CPU-only processing (disables cloud mode)
             gpu: Force local GPU processing (disables cloud mode, requires GPU)
 
         Note:
-            - Cloud mode is the default unless cpu or gpu is specified
+            - Cloud mode is the default unless gpu is specified
             - Without login or API key, limited calls per day
             - For 10k docs/month, run 'docstrange login' (recommended) or use an API key from https://app.nanonets.com/#/keys
         """
         self.preserve_layout = preserve_layout
         self.include_images = include_images
         self.api_key = api_key
         self.model = model
-        self.cpu = cpu
         self.gpu = gpu
 
         # Determine processing mode
-        # Cloud mode is default unless CPU/GPU preference is explicitly set
-        self.cloud_mode = not (self.cpu or self.gpu)
-
-        # Validate CPU/GPU preferences
-        if self.cpu and self.gpu:
-            raise ValueError("Cannot specify both cpu and gpu. Choose one or neither.")
+        # Cloud mode is default unless GPU preference is explicitly set
+        self.cloud_mode = not self.gpu
 
         # Check GPU availability if GPU preference is set
         if self.gpu and not should_use_gpu_processor():
             raise RuntimeError(
                 "GPU preference specified but no GPU is available. "
-                "Please ensure CUDA is installed and a compatible GPU is present, "
-                "or use cpu=True for CPU-only processing."
+                "Please ensure CUDA is installed and a compatible GPU is present."
             )
 
         # Default to True if not explicitly set
@@ -157,7 +149,7 @@ def authenticate(self, force_reauth: bool = False) -> bool:
             return False
 
     def _setup_local_processors(self):
-        """Setup local processors based on CPU/GPU preferences."""
+        """Setup local processors based on GPU preferences."""
         local_processors = [
             PDFProcessor(preserve_layout=self.preserve_layout, include_images=self.include_images, ocr_enabled=self.ocr_enabled),
             DOCXProcessor(preserve_layout=self.preserve_layout, include_images=self.include_images),
@@ -169,19 +161,11 @@ def _setup_local_processors(self):
             URLProcessor(preserve_layout=self.preserve_layout, include_images=self.include_images),
         ]
 
-        # Add GPU processor based on preferences and availability
-        gpu_available = should_use_gpu_processor()
-
-        if self.cpu:
-            logger.info("CPU preference specified - using CPU-based processors only")
-        elif self.gpu:
-            if gpu_available:
-                logger.info("GPU preference specified - adding GPU processor with Nanonets OCR")
-                gpu_processor = GPUProcessor(preserve_layout=self.preserve_layout, include_images=self.include_images, ocr_enabled=self.ocr_enabled)
-                local_processors.append(gpu_processor)
-            else:
-                # This should not happen due to validation in __init__, but just in case
-                raise RuntimeError("GPU preference specified but no GPU is available")
+        # Add GPU processor if GPU preference is specified
+        if self.gpu:
+            logger.info("GPU preference specified - adding GPU processor with Nanonets OCR")
+            gpu_processor = GPUProcessor(preserve_layout=self.preserve_layout, include_images=self.include_images, ocr_enabled=self.ocr_enabled)
+            local_processors.append(gpu_processor)
 
         self.processors.extend(local_processors)
 
@@ -312,14 +296,12 @@ def get_processing_mode(self) -> str:
         """
         if self.cloud_mode and self.api_key:
             return "cloud"
-        elif self.cpu:
-            return "cpu_forced"
         elif self.gpu:
             return "gpu_forced"
         elif should_use_gpu_processor():
             return "gpu_auto"
         else:
-            return "cpu_auto"
+            return "cloud"
 
     def _get_processor(self, file_path: str):
         """Get the appropriate processor for the file.
@@ -340,7 +322,7 @@ def _get_processor(self, file_path: str):
         gpu_available = should_use_gpu_processor()
 
         # Try GPU processor only if format is supported AND (gpu OR auto-gpu)
-        if not self.cpu and ext in gpu_supported_formats and (self.gpu or (gpu_available and not self.gpu)):
+        if ext in gpu_supported_formats and (self.gpu or (gpu_available and not self.gpu)):
             for processor in self.processors:
                 if isinstance(processor, GPUProcessor):
                     if self.gpu:
@@ -349,7 +331,7 @@ def _get_processor(self, file_path: str):
                         logger.info(f"Using GPU processor with Nanonets OCR for {file_path} (GPU available and format supported)")
                     return processor
 
-        # Fallback to normal processor selection (CPU processors)
+        # Fallback to normal processor selection
         for processor in self.processors:
             if processor.can_process(file_path):
                 # Skip GPU processor in fallback mode to avoid infinite loops

diff --git a/docstrange/pipeline/model_downloader.py b/docstrange/pipeline/model_downloader.py
@@ -76,7 +76,7 @@ def download_models(self, force: bool = False, progress: bool = True) -> Path:
         if gpu_available:
             logger.info("GPU detected - including Nanonets OCR model")
         else:
-            logger.info("No GPU detected - skipping Nanonets OCR model (CPU-only mode)")
+            logger.info("No GPU detected - skipping Nanonets OCR model (cloud mode)")
 
         models_to_download = [
             ("Layout Model", self.LAYOUT_MODEL),