ProtonSearch v1.1.0: Windows Launcher Integrates AI Agents

What changed

ProtonSearch, a fast and lightweight Windows launcher, has released version 1.1.0. This new version, formerly known as OmniSearch, offers a unified search experience for various user data and applications directly from a single keyboard shortcut. The launcher is built with Rust and Win32 for speed, utilizing SQLite FTS5 for its search capabilities. It is designed as a local-first application, meaning indexed data remains on the user's PC for privacy.

Key features of ProtonSearch include searching installed applications, files, folders, browser history (Chromium-based and Firefox), clipboard history (including images and text editing), Git activity, system settings, and local notes. It also supports searching Steam games and open application windows. A notable addition is the integration with AI agents, specifically mentioning Hermes agents, which allows users to execute autonomous tasks, run approved commands, and manage AI chats from within the launcher. Commands requiring user approval are prompted before execution, providing an agent layer without making the launcher primarily a chatbot.

ProtonSearch also incorporates several workflow enhancements. These include a "Circle to Search" feature using Google Lens for visual searches directly from the screen, a built-in calculator and unit converter, custom quicklinks for site searches, and snippets for text expansions. The clipboard functionality is robust, offering pinning, multiselect, image support, and editing capabilities. Users can also define focus categories to block distracting applications for distraction-free sessions.

Installation is facilitated via a PowerShell script that downloads the latest release and runs the Windows installer. Existing OmniSearch users can install ProtonSearch over their current installation, with data and settings migrating automatically. The launcher runs from the Windows system tray, allowing background operation without a persistent open window. Search prefixes are available to narrow down searches to specific sources, such as file:, clip:, or agents:.

Why it matters for builders

For developers and power users, ProtonSearch offers a significant reduction in context switching. By consolidating access to files, code, Git history, AI interactions, and system commands into one interface, it allows for more efficient task management. The Rust/Win32 native implementation suggests a focus on performance, which is critical for tools that are frequently used. The local-first design also aligns with growing privacy concerns among users.

Practical impact

Windows users can install ProtonSearch using the provided PowerShell script or by downloading the latest release from the GitHub repository. After installation, users should launch the application, press the configured hotkey (default Alt + Space), and allow the initial indexing to complete. They can then begin searching across their applications, files, clipboard history, and integrated AI agents. Power users can configure custom quicklinks and snippets for personalized workflows, and explore the Hermes agent integration for automated tasks.

Caveats and source limits

The source material indicates that ProtonSearch is a fresh release with 56 stars and 6 forks on GitHub as of the provided excerpt. While it mentions AI signals and developer signals, specific benchmark data or independent performance reviews are not included. The integration with AI agents is described, but the full range of supported AI models or specific agent capabilities beyond Hermes is not detailed. Pricing information is not applicable as it is an open-source project. The release date is noted as "fresh release" with a specific version number (v1.1.0) and a future-dated published_at timestamp, which may indicate a planned release or an artifact of data collection. The source does not provide details on specific context window sizes for AI interactions. The published_at date in the metadata (2026-07-04T21:00:40.000Z) appears to be in the future, suggesting it might be a placeholder or an error in the source data.