{
"event": "PreToolUse",
"tool_name": "Read",
"tool_input": {
"file_path": "\/opt\/scripts\/pipeline\/vision.py",
"limit": 120
}
}
{
"tool_response": {
"type": "text",
"file": {
"filePath": "\/opt\/scripts\/pipeline\/vision.py",
"content": "#!\/usr\/bin\/env python3\n\"\"\"\nVision analysis module for KI-System Pipeline.\nExtracts PDF pages as images and analyzes them with vision models.\n\nUsage:\n python vision.py <pdf_path> # Analyze all pages\n python vision.py <pdf_path> --page 1 # Analyze specific page\n\"\"\"\n\nimport base64\nimport os\nimport sys\nfrom pathlib import Path\n\nimport requests\n\nfrom config import OLLAMA_HOST\nfrom db import db\n\n# Default vision model (can be overridden by pipeline config)\nDEFAULT_VISION_MODEL = \"minicpm-v:latest\"\n\n# Image settings\nIMAGE_DPI = 150 # Balance between quality and size\nIMAGE_FORMAT = \"png\"\nMAX_IMAGE_SIZE_MB = 10\n\n\ndef pdf_to_images(file_path, dpi=IMAGE_DPI):\n \"\"\"\n Convert PDF pages to images.\n\n Args:\n file_path: Path to PDF file\n dpi: Resolution for image extraction\n\n Returns:\n List of dicts with page_number, image_bytes, width, height\n \"\"\"\n import fitz # PyMuPDF\n\n doc = fitz.open(file_path)\n pages = []\n\n for page_num in range(len(doc)):\n page = doc[page_num]\n\n # Render page to image\n mat = fitz.Matrix(dpi \/ 72, dpi \/ 72) # 72 is default PDF DPI\n pix = page.get_pixmap(matrix=mat)\n\n # Convert to PNG bytes\n img_bytes = pix.tobytes(IMAGE_FORMAT)\n\n pages.append({\n \"page_number\": page_num + 1,\n \"image_bytes\": img_bytes,\n \"width\": pix.width,\n \"height\": pix.height,\n \"size_kb\": len(img_bytes) \/ 1024\n })\n\n doc.close()\n return pages\n\n\ndef analyze_image_ollama(image_bytes, model=DEFAULT_VISION_MODEL, prompt=None):\n \"\"\"\n Analyze an image using Ollama vision model.\n\n Args:\n image_bytes: PNG\/JPEG image as bytes\n model: Vision model name (e.g., minicpm-v:latest)\n prompt: Custom prompt (default: document analysis prompt)\n\n Returns:\n dict with analysis results\n \"\"\"\n if prompt is None:\n prompt = \"\"\"Analysiere diese Seite aus einem Schulungsdokument.\n\nBeschreibe strukturiert:\n1. **Überschriften\/Titel**: Welche Überschriften gibt es?\n2. **Hauptinhalt**: Worum geht es auf dieser Seite?\n3. **Visuelle Elemente**:\n - Gibt es Bilder\/Fotos? Was zeigen sie?\n - Gibt es Diagramme\/Charts? Was stellen sie dar?\n - Gibt es Tabellen? Was enthalten sie?\n4. **Layout**: Wie ist die Seite aufgebaut (Spalten, Boxen, etc.)?\n5. **Besonderheiten**: Gibt es Hervorhebungen, Zitate oder Callouts?\n\nAntworte auf Deutsch und sei präzise.\"\"\"\n\n # Encode image as base64\n image_base64 = base64.b64encode(image_bytes).decode(\"utf-8\")\n\n try:\n response = requests.post(\n f\"{OLLAMA_HOST}\/api\/generate\",\n json={\n \"model\": model,\n \"prompt\": prompt,\n \"images\": [image_base64],\n \"stream\": False,\n \"options\": {\n \"temperature\": 0.3,\n \"num_predict\": 1000\n }\n },\n timeout=120\n )\n response.raise_for_status()\n\n result = response.json()\n return {\n \"success\": True,\n \"analysis\": result.get(\"response\", \"\"),\n \"model\": model,\n \"eval_count\": result.get(\"eval_count\", 0),",
"numLines": 120,
"startLine": 1,
"totalLines": 371
}
}
}