SlimSnap

Product Introduction

1. Definition: SlimSnap is a native macOS application and developer tool that functions as a screenshot-to-JSON converter. It captures screen regions and user annotations, performing on-device OCR and layout analysis to produce a structured, token-efficient JSON representation of the visual UI, specifically designed for consumption by CLI coding agents and text-based tools.

2. Core Value Proposition: SlimSnap solves the fundamental problem of giving vision to text-only AI agents (like Claude Code, Aider, Codex CLI). It translates the visual information in a screenshot into a deterministic, machine-readable JSON format that these tools can parse, allowing developers to point at UI elements and describe issues directly within their terminal workflow without switching contexts or writing verbose textual descriptions.

Main Features

1. Capture and Annotate: Utilizes native macOS keyboard shortcuts (⌘⇧S) for precise region selection. Users can immediately overlay semantic annotations such as arrows, callouts, and highlights directly on the capture. This process is handled locally with no additional software installation required beyond the free Mac app.

2. Export to Optimized JSON Schema: The core technical output is a structured JSON file that includes normalized bounding box coordinates (bbox in 0-1 range), unique element IDs, element types (e.g., button, input, label), text values extracted via OCR, and annotation metadata. This structured data representation is dramatically more efficient than raw image tokens.

3. Deterministic Layout Engine: Every UI element detected in the screenshot is assigned a precise bounding box in normalized coordinates. This eliminates guesswork for the AI agent, providing exact positional information ("bbox": [0.34, 0.34, 0.32, 0.07]) that can be directly used for code generation or layout analysis.

4. Built-in Local OCR: The application performs optical character recognition entirely on the user's machine to extract all visible text from labels, buttons, error messages, and input fields. This ensures the AI agent receives the exact textual content the user sees, improving accuracy over relying solely on visual interpretation.

5. Privacy-First, On-Device Architecture: All processing, including capture, OCR, and analysis, occurs locally on the Mac. No screenshots or data are ever uploaded to external servers, requiring no user account and ensuring complete data privacy for sensitive work environments and proprietary UIs.

Problems Solved

1. Pain Point: Terminal-based AI coding agents (Claude Code, Aider, Codex CLI) are blind to visual information. Describing a UI bug, layout issue, or desired feature in text is inefficient, ambiguous, and consumes significant context window space (e.g., 1,568 tokens per image on Sonnet). This creates a workflow friction point between visual software and text-based development tools.

2. Target Audience: The primary users are software developers, front-end engineers, and full-stack developers using CLI-based AI assistants. Secondary audiences include product managers and QA testers who need to communicate visual issues to developers using text-centric collaboration tools like Slack, Jira, or Git commits.

3. Use Cases: Debugging UI layout bugs in a web application by pointing an arrow at the misaligned element. Documenting a feature request by annotating a screenshot of a mockup. Reporting a visual error from a CI/CD pipeline log. Collaborating remotely where only terminal access (SSH) is available, but a visual context is needed.

Unique Advantages

1. Differentiation: Unlike pasting a screenshot directly into a vision-capable LLM (like ChatGPT), SlimSnap is built specifically for the constrained environment of terminal agents and text-only contexts. It offers significant token cost savings (55-85% fewer tokens than raw images) and provides structured, parseable data instead of requiring the AI to interpret pixels. Compared to manual description, it is faster and more accurate.

2. Key Innovation: The product creates a local, efficient pipeline from pixels to structured tokens for CLI tools. Its key innovations are the deterministic normalized coordinate system for precise element referencing and the tight integration with the Claude Code skill, which automatically loads the latest capture JSON into the agent's context via a simple config file (~/.slimsnap/config.json), seamlessly bridging the gap between macOS GUI actions and terminal-based AI workflows.

Frequently Asked Questions (FAQ)

1. What does SlimSnap do that pasting a screenshot into ChatGPT can't? SlimSnap is designed for CLI coding agents (Claude Code, Aider, Codex CLI) that cannot accept image inputs. It converts your screenshot into a token-efficient JSON format that these text-only tools can read, saving you from writing paragraphs to describe a UI. It is also more cost-effective in long coding sessions.

2. Is my screenshot data sent to a server when using SlimSnap? No. Privacy is a core design principle. All capture, OCR, and processing happen entirely on your local Mac. No data is uploaded, and no account is required. Your screenshots never leave your machine.

3. Does SlimSnap work on Windows or Linux? Currently, SlimSnap is a native macOS application. However, because the JSON schema is open-source (MIT license), developers on other platforms can use the format by building their own capture and OCR pipelines to generate a compatible JSON file.

4. Is SlimSnap open-source? The JSON schema and the Claude Code skill are fully open-source under the MIT license on GitHub. The Mac application itself is a free, closed-source download.

5. How do I use SlimSnap with Claude Code or Aider? After installing the Mac app, install the Claude Code skill. The skill reads a config file (~/.slimsnap/config.json) written by SlimSnap to find your capture folder. When you paste a capture (as JSON) into your terminal, the skill automatically loads the latest JSON file into the agent's context, giving it "eyes" on your UI.