[{"data":1,"prerenderedAt":283},["ShallowReactive",2],{"navigation":3,"\u002Fgetting-started":99,"\u002Fgetting-started-surround":280},[4,23,44,78],{"title":5,"path":6,"stem":7,"children":8,"icon":22},"Getting Started","\u002Fgetting-started","1.getting-started\u002F1.index",[9,12,17],{"title":10,"path":6,"stem":7,"icon":11},"Introduction","i-lucide-house",{"title":13,"path":14,"stem":15,"icon":16},"Installation","\u002Fgetting-started\u002Finstallation","1.getting-started\u002F2.installation","i-lucide-download",{"title":18,"path":19,"stem":20,"icon":21},"MCP Setup","\u002Fgetting-started\u002Fmcp-setup","1.getting-started\u002F3.mcp-setup","i-lucide-plug","i-lucide-rocket",{"title":24,"icon":25,"path":26,"stem":27,"children":28,"page":43},"CLI","i-lucide-terminal","\u002Fcli","2.cli",[29,33,38],{"title":30,"path":31,"stem":32,"icon":25},"Usage","\u002Fcli\u002Fusage","2.cli\u002F1.usage",{"title":34,"path":35,"stem":36,"icon":37},"Options","\u002Fcli\u002Foptions","2.cli\u002F2.options","i-lucide-sliders-horizontal",{"title":39,"path":40,"stem":41,"icon":42},"Batch & JSON","\u002Fcli\u002Fbatch-json","2.cli\u002F3.batch-json","i-lucide-package",false,{"title":45,"icon":46,"path":47,"stem":48,"children":49,"page":43},"Providers","i-lucide-cpu","\u002Fproviders","3.providers",[50,55,60,65,70,74],{"title":51,"path":52,"stem":53,"icon":54},"Claude (Area-Based)","\u002Fproviders\u002Fclaude","3.providers\u002F1.claude","i-lucide-square",{"title":56,"path":57,"stem":58,"icon":59},"GPT-4o & GPT-5 (Tiling)","\u002Fproviders\u002Fgpt","3.providers\u002F2.gpt","i-lucide-grid-2x2",{"title":61,"path":62,"stem":63,"icon":64},"Gemini (Large Tiles)","\u002Fproviders\u002Fgemini","3.providers\u002F3.gemini","i-lucide-grid-3x3",{"title":66,"path":67,"stem":68,"icon":69},"Llama Vision (Tiles)","\u002Fproviders\u002Fllama","3.providers\u002F4.llama","i-simple-icons-meta",{"title":71,"path":72,"stem":73,"icon":64},"Qwen-VL (Patch Grid)","\u002Fproviders\u002Fqwen","3.providers\u002F5.qwen",{"title":75,"path":76,"stem":77,"icon":59},"DeepSeek-VL (Open Weights)","\u002Fproviders\u002Fdeepseek","3.providers\u002F6.deepseek",{"title":79,"icon":80,"path":81,"stem":82,"children":83,"page":43},"Guides","i-lucide-book-open","\u002Fguides","4.guides",[84,89,94],{"title":85,"path":86,"stem":87,"icon":88},"Python Bindings","\u002Fguides\u002Fpython-bindings","4.guides\u002F1.python-bindings","i-lucide-file-code",{"title":90,"path":91,"stem":92,"icon":93},"Sandbox (Think in Code)","\u002Fguides\u002Fsandbox","4.guides\u002F2.sandbox","i-lucide-flask-conical",{"title":95,"path":96,"stem":97,"icon":98},"Crawler Integration","\u002Fguides\u002Fcrawler-integration","4.guides\u002F3.crawler-integration","i-lucide-globe",{"id":100,"title":10,"body":101,"description":273,"extension":274,"links":275,"meta":276,"navigation":277,"path":6,"seo":278,"stem":7,"__hash__":279},"docs\u002F1.getting-started\u002F1.index.md",{"type":102,"value":103,"toc":265},"minimark",[104,112,115,120,147,151,235,239],[105,106,107,111],"p",{},[108,109,110],"strong",{},"VisionSqueezer"," is a middleware that bridges the gap between human-centric images and LLM-native vision tokenomics. It pre-processes images to trigger the absolute minimum billable tiles across major AI providers — Claude, GPT-4o\u002FGPT-5, and Gemini.",[105,113,114],{},"Vision models do not \"see\" pixels the way you do. Each provider tokenizes images through its own internal grid math, and a handful of stray pixels can push an image into an extra tile row that doubles its cost. VisionSqueezer simulates that math and snaps your images to the cheapest valid boundary.",[116,117,119],"h2",{"id":118},"why-it-matters","Why it matters",[121,122,123,135,141],"ul",{},[124,125,126,129,130,134],"li",{},[108,127,128],{},"Padding is expensive."," Claude bills by pixel area ",[131,132,133],"code",{},"(W × H \u002F 750)",". Every pixel of solid border costs tokens.",[124,136,137,140],{},[108,138,139],{},"Tile spill-over is silent."," A 769px-wide image costs the same as a 1536px one on a 768px-tile grid.",[124,142,143,146],{},[108,144,145],{},"Format ≠ tokens."," Token cost is dimensional. AVIF\u002FWebP\u002FJPEG only affect file size and upload latency — not API token count.",[116,148,150],{"id":149},"key-features","Key Features",[121,152,153,159,169,175,185,195,208,214,220,226],{},[124,154,155,158],{},[108,156,157],{},"Provider-aware resizing"," — exact internal grid math for Claude (area-based), GPT-4o\u002FGPT-5 (512px tiles), and Gemini (768px tiles).",[124,160,161,164,165,168],{},[108,162,163],{},"Smart crop"," — saliency-based crop (Sobel-lite gradient energy) via ",[131,166,167],{},"--smart-crop",", plus default corner-tolerance padding strip.",[124,170,171,174],{},[108,172,173],{},"AVIF \u002F WebP \u002F JPEG output"," — AVIF is typically 20–50% smaller than WebP, ~3× smaller than JPEG at equal quality.",[124,176,177,180,181,184],{},[108,178,179],{},"Auto-quality"," — ",[131,182,183],{},"--auto-quality 0.95"," binary-searches quality to hit a perceptual SSIM target.",[124,186,187,190,191,194],{},[108,188,189],{},"Batch \u002F recursive mode"," — squeeze a whole directory tree, mirror structure with ",[131,192,193],{},"--output-dir",".",[124,196,197,180,200,203,204,207],{},[108,198,199],{},"Machine-readable output",[131,201,202],{},"--json"," and ",[131,205,206],{},"--dry-run"," for pipelines.",[124,209,210,213],{},[108,211,212],{},"Think in Code (Sandbox)"," — execute atomic image ops locally before sending to the LLM.",[124,215,216,219],{},[108,217,218],{},"Persistent analytics"," — cumulative token & USD savings in a local SQLite database.",[124,221,222,225],{},[108,223,224],{},"Universal MCP"," — one-liner integration for Claude Code, Cursor, Zed, VS Code, Windsurf, JetBrains.",[124,227,228,180,231,234],{},[108,229,230],{},"Python bindings",[131,232,233],{},"pip install vision-squeezer"," via pyo3 wheels for Linux\u002FmacOS\u002FWindows.",[116,236,238],{"id":237},"next-steps","Next steps",[240,241,242,248,253,259],"card-group",{},[243,244,245],"card",{"icon":16,"title":13,"to":14},[105,246,247],{},"Install via cargo, npm, or pip.",[243,249,250],{"icon":21,"title":18,"to":19},[105,251,252],{},"Wire VisionSqueezer into your AI editor or agent.",[243,254,256],{"icon":25,"title":255,"to":31},"CLI Usage",[105,257,258],{},"Optimize a single image or a whole tree.",[243,260,262],{"icon":46,"title":261,"to":52},"Provider Math",[105,263,264],{},"Understand how each model is billed.",{"title":266,"searchDepth":267,"depth":268,"links":269},"",1,2,[270,271,272],{"id":118,"depth":268,"text":119},{"id":149,"depth":268,"text":150},{"id":237,"depth":268,"text":238},"LLM-native image optimization that mathematically snaps images to provider grid boundaries to cut vision token usage.","md",null,{},{"icon":11},{"title":10,"description":273},"14Tup7BdRyI_2WxNM2_fArzzCtu4tn5KXYJYn_UYzxk",[275,281],{"title":13,"path":14,"stem":15,"description":282,"icon":16,"children":-1},"Install VisionSqueezer via cargo, npm, or pip.",1782053691569]