Inspiration
Most AI writing tools interrupt users with chats, popups, or dedicated editors. Even powerful tools like GitHub Copilot are limited to code editors, while everyday writing still feels fragmented.
I wanted to explore a different interaction model: AI that stays invisible and works everywhere.
Universal Copilot was inspired by how natural GitHub Copilot feels—inline ghost text, instant suggestions, and a single key to accept. This project asks a simple question: why shouldn’t that experience exist system-wide, across all apps and all kinds of writing?
What it does
Universal Copilot is a Windows-native, system-wide AI writing copilot.
As the user types in any application (Notepad, browsers, text fields, etc.), AI-generated suggestions appear as inline ghost text at the caret. The user can:
Press TAB to accept the suggestion
Press ESC to dismiss it
Keep typing to instantly update the prediction
There are no chats, popups, or editors. The AI quietly assists only when needed.
How we built it
The project is built as a native Windows desktop application, not a website or browser extension.
Key components include:
Low-level caret tracking using Windows APIs
A transparent, click-through overlay window for rendering ghost text
A keyboard hook that intercepts only TAB and ESC
Context capture from the active text field with correct UTF-16 handling
A streaming inference client using an OpenAI-compatible API
Token-by-token rendering with cancellation and confidence-based truncation
All components are designed to fail silently to preserve a smooth user experience.
Challenges we ran into
System-wide text detection: Not all applications expose caret or text information consistently.
Keyboard safety: Intercepting TAB and ESC without breaking app behavior required careful low-level handling.
Latency: Streaming AI suggestions had to stay fast (<150ms perceived delay) to feel natural.
Invisible UX: Avoiding flicker, jitter, and intrusive behavior was harder than adding features.
Accomplishments that we're proud of
Built a Copilot-style inline AI experience outside of any editor
Achieved real-time, streaming ghost text with smooth cancellation
Designed a system that feels invisible and unobtrusive
Created a working prototype that runs across any Windows text field
What we learned
Great AI tools are defined more by UX decisions than model size
System-level software introduces real-world complexity beyond web apps
Responsiveness and trust matter more than feature count
The best productivity tools disappear into the workflow
What's next for Universal Copilot
Potential future directions include:
macOS support
Optional local on-device inference
Smarter confidence scoring
Lightweight installer and onboarding flow
For this hackathon, the goal was to prove that system-wide, inline AI writing assistance is both possible and practical.
Log in or sign up for Devpost to join the conversation.