Everywhere: A Screen-Aware AI Assistant for Windows and macOS
On-screen aware AI assistant for your desktop. Uses current app context, multiple LLMs, and MCP tools to help you act across apps.
At a glance
- What is it?
- Everywhere is a desktop AI assistant for Windows and macOS that reads the active application's context through accessibility APIs, letting users query an LLM about what is currently on screen without copying text or switching windows. It supports multiple LLM providers, MCP tools, and sub-agent dispatch, activated by a global hotkey.
- Who is it for?
- Everywhere suits developers and knowledge workers on Windows or macOS who switch frequently between applications and want to query an LLM about whatever is currently on screen without interrupting their flow. Linux support is listed as in progress.
- Can I use it commercially?
- Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
- Is it still maintained?
- Yes. The repository last received commits 1 day ago.
- What is it written in?
- Mainly C#, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 29, 2026, and from our analysis. They are not legal advice.
Editorial analysis
What Everywhere Does: AI Assistance Without Switching Windows
The typical flow for using an AI assistant on a desktop involves selecting text, switching to a browser or separate app, pasting the content, and typing a question. Everywhere eliminates that sequence. The user presses a global hotkey and types a question; the assistant reads whatever is currently visible in the active application and uses that as context.
The README describes use cases that show the range: asking what an error message means while looking at it, requesting a summary of a webpage without leaving it, translating a word without opening a translation tool, requesting a tone adjustment for a draft email, or fact-checking a claim in a document. None of these require copying text to a clipboard or navigating to a new window.
The most recent release is v0.8.2 from 2026-09-16, and the last push to the repository was on 2026-09-27.
Context Awareness: Accessibility APIs and UI Automation
What separates Everywhere from a standard floating LLM chat interface is how it reads the screen. The README describes two mechanisms: multi-modal screen capture (screenshots) and direct integration with underlying accessibility APIs and UI automation technologies.
The accessibility API integration is described as allowing the application to accurately extract active structured environment data across a huge variety of software with low intrusion. This means that for supported applications, Everywhere reads the actual text and UI structure rather than just a rendered image, which produces more reliable context extraction than screenshot-only approaches.
On Windows, the README mentions integration with Everything (fast file search) as one of the agent capabilities. On macOS, integrated system apps are listed as a supported agent tool. These platform-specific integrations are noted separately from the cross-platform features.
Installing Everywhere on Windows or macOS
The README describes two acquisition paths under the heading Acquisition and Installation. Detailed installation steps are in the official documentation at everywhere.sylinko.com.
The repository root shows platform-specific solution files: Everywhere.Windows.slnx and Everywhere.Mac.slnx, alongside the combined Everywhere.slnx. The project is built with .NET 10 and Avalonia (a cross-platform .NET UI framework), as shown by the badges in the README. The global.json file in the root pins the .NET SDK version, ensuring contributors and CI builds use the same toolchain.
Releases are published on GitHub: v0.8.2 was released on 2026-09-16, with canary builds available before it. Canary releases are published to the same GitHub Releases page with pre-release labels, allowing early access to features before a stable release. The scripts/ directory in the repository root contains build automation used by the project's CI pipeline.
LLM Providers and MCP Tools
Everywhere does not lock users into a single AI provider. The README lists the currently supported model sources: the Everywhere Cloud Service, OpenAI, Anthropic (Claude), Google (Gemini), DeepSeek, Moonshot (Kimi), MiniMax, local deployment via Ollama, and custom API endpoints. This means a developer can use a locally running model through Ollama with no external API calls, or switch between providers depending on the task.
On the tooling side, the agent system includes a web browser, sub-agent dispatch, a local file system tool, terminal script execution, and MCP (Model Context Protocol) tools. MCP support means Everywhere can call any MCP-compatible server, extending its reach to databases, external APIs, or custom tools without changes to Everywhere itself.
Voice interaction and a memory system are listed in the README as work in progress. Mouse shortcuts are also marked as not yet available.
The Strategy Engine and Hotkey Activation
The README describes a scenario invocation and strategy engine as an advanced feature. The intended behavior is that when Everywhere is invoked via the global shortcut, it detects the current application and scenario context, and can push relevant execution strategies without the user needing to explain the situation. The README marks this feature as in progress (using a construction icon) and notes that the original intention was to end the workflow that disrupts working rhythm.
In the current released state, users invoke Everywhere with a global hotkey and type their request. The UI is described in the README as using a frosted glass design. The assistant renders responses with rich Markdown and math formula support. Global system hotkeys work across all applications, meaning Everywhere can be invoked regardless of which window is active.
Text selection interaction is listed as a supported feature, meaning that selecting text in any application before invoking Everywhere passes that text as part of the context.
Limitations and Features Still in Progress
Linux support is listed as work in progress. The platform-specific solution files in the repository confirm that Windows and macOS are the current targets, with no Linux build available in the current releases.
Voice interaction, mouse shortcuts, and a memory system are each marked as work in progress in the README feature table. The memory system is listed under the Powerful Agent System category, meaning the agent currently does not retain context between sessions.
The license field in the repository metadata shows NOASSERTION, which means the repository does not carry a standard SPDX license identifier at the metadata level. The LICENSE file in the repository root and ThirdPartyNotices.txt (which explicitly exists in the repository) should be reviewed directly before commercial or enterprise deployment.
The README notes that the i18n translations are partially AI-assisted and invites corrections, which is a signal that the localization quality is uneven across the supported languages.
Everywhere vs. Raycast AI
Raycast is a macOS launcher that includes an AI assistant layer. It activates via a global hotkey, supports multiple AI providers, and can call external tools through extensions. Raycast is macOS-only and is a closed-source commercial product with a free tier and a paid Pro plan for AI features.
Everywhere targets both Windows and macOS, is open-source, and places its differentiation specifically in the screen-context awareness through accessibility APIs rather than clipboard-based or manual context entry. Raycast's AI mode requires the user to describe what they are looking at or paste content; Everywhere attempts to read the active application's context directly.
The trade-off is maturity: Raycast is a well-established product with a large extension ecosystem, while Everywhere is at v0.8.2 with several features still in progress. Teams committed to macOS and wanting a more mature ecosystem will find Raycast more polished; teams on Windows or those who specifically want accessibility-API context extraction will find Everywhere covers ground that Raycast does not.
Another comparison point is the extension model. Raycast's extensions are JavaScript packages published to a curated registry, covering hundreds of services. Everywhere's tooling layer uses MCP, which is a protocol rather than a proprietary registry. Any MCP-compatible server can be connected to Everywhere without going through a centralized approval process. This is an architectural difference that affects how organizations with internal tools or custom integrations can extend the assistant's reach.
Editorial conclusion
Everywhere suits developers and knowledge workers on Windows or macOS who switch frequently between applications and want to query an LLM about whatever is currently on screen without interrupting their flow. Linux support is listed as in progress. The license terms are listed as NOASSERTION in the repository metadata, so teams with compliance requirements should review the LICENSE file and the ThirdPartyNotices.txt in the repository root before adopting it. Before installing, confirm that the LLM provider you want to use is in the supported list and that your workflow fits the shortcut-based activation model, since voice interaction and mouse shortcuts are listed as work in progress.
Frequently asked questions
Does Everywhere work on Linux?
Not yet. The README feature table lists Linux as work in progress. The current releases cover Windows and macOS only.
Can Everywhere use a locally running AI model?
Yes. The README lists Ollama as a supported model source, which allows running a local LLM with no external API calls. Custom API endpoints are also supported for other local or self-hosted model servers.
What is the difference between Everywhere's screen reading and a screenshot-based approach?
The README describes two mechanisms: standard screenshot-based multi-modality and direct integration with accessibility APIs and UI automation. The accessibility API path extracts structured text and UI data from the active application rather than interpreting a rendered image, which the README says produces more accurate context extraction for a wide variety of software. The accessibility path also works when the text of interest is not visible in the current viewport.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/sylinko-everywhere)