vision-squeezer 0.8.0

---
seo:
  title: VisionSqueezer — Stop Leaking Vision Tokens
  description: Fit images to a token budget before they reach Claude, GPT, or Gemini. For code, agents and pipelines, not for images pasted into a chat.
---

::u-page-hero{class="dark:bg-gradient-to-b from-neutral-900 to-neutral-950"}
---
orientation: horizontal
---
#top
:hero-background

#headline
:version-badge

#title
Stop Leaking [Vision Tokens]{.text-primary}.

#description
Fit images to a token budget before they reach the model. Give it a budget (1600 by default in the MCP server) and it downsizes images on Claude, GPT, and Gemini's exact grid boundaries: about 66% fewer Claude tokens on our sample photos, with composition, colour, and large text intact. It works on images your code, agents, and project files send. It cannot shrink an image you paste into a chat window.

#links
  :::u-button
  ---
  to: /getting-started
  size: xl
  trailing-icon: i-lucide-arrow-right
  ---
  Get started
  :::

  :::u-button
  ---
  icon: i-simple-icons-github
  color: neutral
  variant: outline
  size: xl
  to: https://github.com/eralpozcan/vision-squeezer
  target: _blank
  ---
  View on GitHub
  :::

#default
  ```bash [terminal]
  cargo run -- data/image.jpg --model gpt6

  // Squeezer simulating OpenAI's short-side algorithm...
  Input:  4096×3072  (2.2 MB)
  Output: 4095×2048  (1.2 MB)

  Tokens Saved: 5,595 (33.3% cheaper)
  ```
::

::u-page-section
#title
The Math Behind the Magic

#description
Every provider tokenizes images differently. Squeezer simulates each provider's internal grid math and snaps your images to the cheapest valid boundary.

#features
  :::u-page-feature
  ---
  icon: i-lucide-gauge
  to: /guides/token-budget
  ---
  #title
  Token budget

  #description
  `--max-tokens N` downscales until the target model's estimate fits. The MCP server defaults to 1600, about 66% fewer Claude tokens and 50% fewer GPT-6 tokens on the sample photos.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-square
  ---
  #title
  OpenAI GPT-6 / GPT-5.6

  #description
  Current OpenAI models count 32×32 patches with a 2500-patch high-detail budget. Squeezer fits that budget and removes partially used edge patches.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-grid-2x2
  ---
  #title
  Claude 4.7+ (Patch System)

  #description
  Claude counts 28×28 patches. Squeezer fits the high-resolution or standard tier and aligns dimensions to the patch grid.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-grid-3x3
  ---
  #title
  Gemini (Massive Tiles)

  #description
  Gemini uses huge 768×768 blocks. A slightly overlapping image costs you double. We snap images securely down to the nearest tile boundary.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-flask-conical
  ---
  #title
  Think in Code (Sandbox)

  #description
  Let your agent execute custom crops, binarization, or filters locally. Extract only the context you need to save up to 99.9% tokens.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-bar-chart-3
  ---
  #title
  Persistent Analytics

  #description
  Locally tracks every optimization in a SQLite database. View your cumulative USD savings directly from your terminal or AI agent.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-plug
  ---
  #title
  Universal MCP

  #description
  Works natively with Claude Code, Cursor, Zed, and VS Code. No complex setup — just plug it into your favorite AI tool.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-image
  ---
  #title
  AVIF Output

  #description
  `--format avif` encodes ~20–50% smaller than WebP at equal quality, ~3× smaller than JPEG. Same tokens, less bandwidth.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-crop
  ---
  #title
  Smart Crop & Auto-Quality

  #description
  `--smart-crop` uses edge-energy (Sobel-lite) to keep high-information regions. `--auto-quality 0.95` binary-searches quality to hit a perceptual SSIM target.
  :::

  :::u-page-feature
  ---
  icon: i-lucide-package
  ---
  #title
  Batch & JSON Output

  #description
  Pass a directory + `--recursive` to squeeze a whole tree at once. `--json` emits a structured record for pipelines. `--dry-run` reports without writing.
  :::
::

::u-page-section
#title
Interactive [Savings Calculator]{.text-primary}

#description
Real numbers from the Squeezer pipeline. Pick an image source and an optimization target to see token and file-size savings per provider.

#default
  :savings-calculator
::

::u-page-section
#title
Universal [MCP Integration]{.text-primary}

#description
Select your agent or editor. Thanks to `npx -y`, zero global installation is required — just paste the configuration.

#default
  :install-selector
::

::u-page-section
#title
Current OpenAI support

#description
Use `--model gpt6` for GPT-6 and GPT-5.6 patch tokenization. The `gpt4o` and `gpt5` profiles remain for legacy endpoints with 512px high-detail tiling.
::

::u-page-section{class="dark:bg-gradient-to-b from-neutral-950 to-neutral-900"}
  :::u-page-c-t-a
  ---
  links:
    - label: Get started
      to: '/getting-started'
      trailingIcon: i-lucide-arrow-right
    - label: View on GitHub
      to: 'https://github.com/eralpozcan/vision-squeezer'
      target: _blank
      variant: subtle
      icon: i-simple-icons-github
  title: Stop paying for padding.
  description: Install VisionSqueezer in one command and start cutting vision token costs across every provider.
  class: dark:bg-neutral-950
  ---

  :stars-bg
  :::
::