Claude Overlay

Product Introduction

1. Definition: Claude Overlay is a Windows-native desktop application built with Python and Tkinter that functions as a frameless, always-on-top AI assistant interface. It is a technical overlay layer that integrates directly with the official Claude Code CLI (Command Line Interface) and the claude-agent-sdk.
2. Core Value Proposition: It exists to eliminate context-switching for technical and creative workflows by providing a screen-aware, autonomous Claude Code agent that floats over any application. Its primary value is enabling users to query and command a full AI agent about visual content on their desktop without using an API key, leveraging their existing Claude subscription for file editing, command execution, and real-time screen analysis.

Main Features

1. Multi-Monitor Screen Capture & Context: On every message sent, the overlay automatically captures a screenshot of every active monitor using the Python Pillow library's ImageGrab.grab() function. It labels each image as primary or secondary within the prompt, providing Claude Code with spatial context of the user's digital workspace. This allows for queries like "compare the graph on my left monitor to the data table on the right."
2. Frameless, Collapsible Orb UI: The interface is a borderless Tkinter window with Win32 API calls for true always-on-top behavior and DPI awareness. It can be collapsed into a small, draggable "orb" (a pixel sprite) that minimizes screen real estate. The orb displays a checkmark (✓) when a background task completes, and each instance can be individually named, allowing users to run multiple task-specific overlays simultaneously.
3. Native Claude Code Agent Integration: Instead of calling the Anthropic API directly, the overlay spawns and communicates with the user's locally installed claude CLI as a subprocess via the claude-agent-sdk. This grants the overlay the full capabilities of Claude Code (Opus 4.8), including the Read, Edit, Run, and Plan tools, operating within a user-defined WORKING_DIR. This enables autonomous file editing and shell command execution.
4. Direct Application Interaction via Automation: A key advanced feature is the agent's ability to act upon the very applications visible on screen. By combining screen vision with shell/scripting access (e.g., PowerShell, COM automation), Claude Code can be prompted to edit the open document in Microsoft Word, modify formulas in a visible Excel spreadsheet, or adjust a slide in PowerPoint, effectively using the host OS as an instrument.
5. Local & Subscription-Based Operation: The entire stack runs locally on the user's machine. Authentication and billing are handled through the user's official Claude Pro/Max subscription via the claude auth login flow, bypassing separate API key management and per-token metered billing. All data (screenshots, conversation context) remains local until sent to Anthropic's servers via the official CLI channel.

Problems Solved

1. Pain Point: Disruptive workflow interruption caused by constant alt-tabbing between a primary application (IDE, design tool, spreadsheet) and a separate AI chat window or browser tab.
2. Pain Point: Inefficiency and error in manually describing complex visual content (error dialogs, UI layouts, data visualizations) or copying text between applications to seek AI assistance.
3. Pain Point: Limited actionability of standard AI chatbots that can only discuss content, not directly modify source files, run build commands, or manipulate GUI applications based on visual input.
4. Target Audience: Windows-based Software Developers debugging code or parsing logs; Data Analysts & Financial Modelers working in Excel or BI tools; Content Creators & Technical Writers editing documents or presentations; Product Managers & Support Engineers analyzing UI/UX or error states; Multilingual Professionals needing real-time translation of on-screen text.
5. Use Cases: Live Code Debugging: Capturing an IDE with a runtime error and asking "How do I fix this?" with the agent potentially editing the source file. Presentation Assistance: Getting real-time fact-checks or translations while in full-screen slideshow mode. Data Model Generation: Instructing the agent to "build a three-statement financial model in this open Excel workbook" based on a verbal description. GUI Application Support: Taking a screenshot of a complex settings panel and asking "Which option enables feature X?"

Unique Advantages

1. Differentiation vs. Claude Desktop App/Browser: Unlike the official desktop app which is a separate window, Claude Overlay is a persistent, transparent layer that never requires window focus change. Unlike the web interface, it has direct filesystem and shell access as an agent and operates on your subscription without API costs.
2. Differentiation vs. Other Screen-Capture AI Tools: Many "screen-aware" AI tools are cloud-based SaaS that upload screenshots to their own servers. Claude Overlay is open-source (MIT license), runs the official Anthropic CLI locally, and uses your direct Anthropic account, offering greater transparency and data control.
3. Key Innovation: The integration of visual context (multi-monitor screenshots) with a fully autonomous agent (Claude Code) within a minimalist, non-modal UI (collapsible overlay). This creates a seamless "ambient intelligence" paradigm where the AI assistant is an operable layer atop the existing desktop environment, not a separate destination application.

Frequently Asked Questions (FAQ)

1. Is Claude Overlay free to use? Claude Overlay itself is free and open-source software, but it requires an active Claude Pro or Claude Max subscription to function, as it operates through your official claude CLI login. There are no additional API fees or overlay licensing costs.
2. How does Claude Overlay compare to using the Claude Code CLI directly in a terminal? The CLI requires manual prompting and lacks screen vision. Claude Overlay adds automatic visual context capture, a persistent GUI, conversation management, and a collapsible interface, transforming the CLI from a command tool into an interactive visual copilot.
3. Is Claude Overlay safe? Does it send my screenshots to Anthropic? The overlay captures screenshots locally and sends them to Anthropic's servers via the same pipeline as the official Claude Code CLI when you submit a message. The default "bypassPermissions" mode allows autonomous file edits and command execution, so users should exercise caution. For safer operation, change the PERMISSION_MODE setting to "acceptEdits" or "plan" in the configuration.
4. Does Claude Overlay work on macOS or Linux? Currently, Claude Overlay is built for Windows 10/11 only, as it relies on Win32 APIs for window management and the Pillow ImageGrab module which is fully featured on Windows. Cross-platform support for macOS and Linux would require significant re-engineering of the windowing and screen capture components.
5. Can I use Claude Overlay without the Claude Code CLI installed? No. The claude CLI is the essential backend engine. The overlay is solely a visual interface and controller for it. The included setup.cmd script can automatically download and install the CLI for you if it is missing.