Julie – Screen-Aware Desktop AI Assistant with Computer Use Agents
Julie is an open-source desktop AI overlay for macOS and Windows that reads your screen, responds via voice or text, and runs browser and computer automation agents using Groq-powered LLMs.
TL;DR
TL;DR: Julie is a transparent desktop AI overlay for macOS and Windows that understands your screen context, responds in real time, and runs autonomous browser and computer-use agents powered by Groq-hosted LLMs.
Source and Accuracy Notes
⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.
- Project page: tryjulie.vercel.app
- Source repository: github.com/Luthiraa/julie
- License: MIT (verified via README.md shields.io badge — LICENSE file not present in repo root)
- HN launch thread: not found in HN Algolia (repo created 2025-12-19, no indexed Show HN post)
- Source last checked: 2026-08-20 (commit
7b633ef5)
What Is Julie?
Julie is a lightweight, transparent desktop AI assistant built to reduce context switching. It lives on top of your workspace, understands what is on your screen, and responds via voice or text without forcing you to switch tabs or break focus.
The description on the project page reads:
“Julie is for people who want the power of an AI copilot without inviting a second workspace into their life. We’re building an overlay layer for knowledge work. Julie sits quietly on top of your apps, sees what you’re looking at, and helps in real time.”
Unlike a chat window you switch to, Julie stays visible as a floating overlay. Its ghost mode lets it sit on top of your screen without blocking clicks — press Cmd+Shift+I on macOS and it becomes click-through so you can keep working while Julie watches.
The AI backend is Groq (Llama 3 70B and Llama 4 Scout), chosen for fast, reasoning-heavy responses. You bring your own API key.
Setup Workflow
Prerequisites
- macOS (Apple Silicon or x64) or Windows (x64 or ARM64)
- A Groq API key (starts with
gsk_)
Step 1: Install Julie
Download the latest installer from the releases page:
| OS | Architecture | Installer |
|---|---|---|
| macOS | Apple Silicon (M1/M2/M3) | Julie-0.0.1-arm64.dmg |
| Windows | x64 | Julie-Setup-0.0.1-x64.exe |
| Windows | ARM64 (Surface/Snapdragon) | Julie-Setup-0.0.1-arm64.exe |
Step 2: Configure Your Groq API Key
- Launch Julie.
- Click the Grid Icon to open Settings.
- Enter your Groq API Key (from console.groq.com/keys).
- Click Save.
Step 3: Start Using Julie
- Type or speak to ask questions.
- Press
Cmd+Shift+I(macOS) orCtrl+Shift+I(Windows) to toggle Ghost Mode — Julie becomes click-through. - Press
Cmd+](macOS) orCtrl+](Windows) to toggle visibility. - Use Smart Mode for complex, multi-step queries.
Agentic Capabilities
Julie goes beyond single prompts. It runs autonomous agents that can:
Browser Automation (Puppeteer)
- Navigate to any URL
- Click elements via CSS selectors
- Type text into input fields
- Scroll pages up or down
- Read page content and interactable elements
- Execute custom JavaScript
Computer Use (macOS only, via JXA)
- Mouse Movement — move cursor to specific screen coordinates
- Click Actions — left click, right click, double click
- Drag & Drop — drag from one point to another
- Scroll — scroll at any position on screen
- Screen Info — get screen size and cursor position
Keyboard and Terminal
- Type Text — simulate keyboard input to any active window
- Run Commands — execute shell commands directly from the assistant
Practical Evaluation Checklist
- [ ] App launches without errors on macOS or Windows
- [ ] Groq API key accepted and saved
- [ ] Ghost mode (
Cmd+Shift+I) works — overlay becomes click-through - [ ] Screen-aware vision captures current screen on click
- [ ] Browser automation agent can open a URL and extract content
- [ ] Keyboard shortcuts respond correctly (visibility toggle, ghost mode)
- [ ] No excessive CPU/GPU usage when idle
Security Notes
- All AI inference is handled by Groq’s API — screen content is sent to Groq when vision mode is used.
- The Groq API key is stored locally in app settings.
- Browser automation (Puppeteer) runs locally and can interact with any website — use caution when granting internet access.
- Computer use agents (JXA on macOS) can control mouse and keyboard — only enable when you want active automation.
FAQ
Q: Does Julie work on Linux? A: Not currently. The README only lists macOS and Windows installers.
Q: What models does Groq power Julie with? A: The README specifies Llama 3 70B and Llama 4 Scout via Groq’s API.
Q: Is there a cost to using Julie? A: Julie itself is free and open source. You pay for your own Groq API key usage, which has a free tier available at console.groq.com.
Q: Can Julie run locally without an internet connection? A: No — AI inference is handled by Groq’s cloud API. The app must have internet access to generate responses.
Q: What is Ghost Mode?
A: Ghost Mode (Cmd+Shift+I on macOS) makes Julie’s window click-through, so you can keep it visible while working in other apps without blocking interactions.
Conclusion
Julie fills a specific niche: a persistent, screen-aware AI overlay that can autonomously run browser and desktop automation. Built in a weekend and open-sourced, it is early-stage but functional. If you want an always-on AI copilot that understands your screen context and can automate repetitive browser and desktop tasks, Julie is worth a short.
The Groq-powered inference stack gives fast reasoning responses, and the transparent overlay UI avoids the context-switching tax of switching to a chat window. Its computer-use agents (JXA on macOS, Puppeteer for browser) are the standout feature for power users who want to automate multi-step workflows.
The main limitation is that it is a single-developer project in early stages — the 0.0.1 release label and lack of a formal HN launch thread suggest it has not yet seen broad adoption. If you try it, file issues on the repo.
- Project page: tryjulie.vercel.app
- Repo: github.com/Luthiraa/julie
Related Posts
ai-setup
Recall – Persistent Memory for Claude Code via MCP Hooks
Recall gives Claude Code a permanent memory store that survives session restarts and context compaction. Four hooks capture and restore context automatically — with cloud SaaS or self-hosted options.
2/28/2026
ai-setup
Sentrial – Catch AI Agent Failures Before Your Users Do
YC W26-backed AI agent observability platform. Trace sessions, detect silent regressions, and A/B test prompts in production before failures reach users.
5/28/2026
ai-setup
IonRouter – Fast Low-Cost AI Inference API
IonRouter is a YC W26 inference API routing open-source and fine-tuned models via an OpenAI-compatible endpoint, built on a C++ runtime optimized for GH200.
5/28/2026