Clicky

Clicky fills a niche that most AI assistants ignore: real-time, screen-aware tutoring. Instead of alt-tabbing to a chat window, you talk to Clicky and it responds with voice, pointing at UI elements on your screen. The implementation is clever—push-to-talk streams audio over WebSocket to AssemblyAI, the transcript plus screenshot go to Claude via SSE, and the response is spoken through ElevenLabs. Claude can even embed [POINT:x,y:label:screenN] tags to move your cursor to specific buttons or fields, which is surprisingly effective for learning. When to pick Clicky: if you're a Mac user learning a new tool (Figma, DaVinci, Xcode) and prefer spoken guidance over reading docs. It's also great for developers who want to hack on a SwiftUI + Cloudflare Workers stack—the codebase is clean and MIT-licensed. When to pass: if you're on Windows or Linux—it's macOS-only. Also avoid if you want a cloud-managed assistant that just works without deploying a Cloudflare Worker and managing your own API keys. The setup is not trivial. Compared to alternatives: ChatGPT's desktop app can also see your screen (on macOS) and talk to you, but it's a general-purpose chatbot, not a tutor that points. GitHub Copilot stays in the editor. Clicky's superpower is that it lives in the cursor's context and can physically direct your attention. But it lacks Copilot's deep IDE integration or ChatGPT's broad knowledge base. A caveat: the original version is stable, but the creator mentioned on April 27, 2026 that new features will be private. That means community contributions become more important. If you rely on this daily, you might hit a ceiling on feature updates. Still, the MIT license means anyone can fork and extend it.