A cute little AI assistant that lives on your desktop.
Download for Windows · License (MIT)
$100–$200. All for just a handful of tasks.
Yep; that’s what a lot of desktop AI assistant products charge. A hefty subscription for something that automates a few things and then sits there.
Then there are amazing open-source projects… and you open the repo and hit a wall:
- “Use Claude Code.”
- “Use Codex.”
- “Figure it out yourself.”
No beginner-friendly guide. No proper setup. No explanation of what anything does.
So even when the code is open, most people never actually get to use it.
The biggest advantage those companies have isn’t that they’re building something impossible.
It’s distribution — and clarity.
That’s why Mirrly exists.
An open-source desktop AI assistant with one goal: make it accessible to everyone.
Whether you’re technical or not, you get a real installer, clear docs, a website that speaks plain language, and a codebase you can actually read.
You can:
- Download it and run it locally
- Clone it and learn how it works
- Fork it and contribute
- Customize it
- Build your own assistant on top of it
- Or polish it into something that’s yours
I wanted everyone to have their own small AI helper on the desktop; always there, actually useful, and enjoyable to talk to. Not locked behind a subscription wall. Not buried behind “install seventeen CLIs first.”
Mirrly floats over your other apps. You bring your own AI key (or start with Pollinations / local Ollama — no cloud key required).
It can:
- Look at your screen and help explain, write, or fix something
- Listen on a call (if you allow the mic) and help with what to say or meeting notes
- Open apps and websites for you
- Focus windows, read/write project files (with approvals + local snapshots)
- Cowork on a live markdown note you and Mirrly edit together
- Keep projects with their own instructions, files, and memory
- Draft an email for you (you still click Send)
- Look things up on the real web instead of guessing
- Use MCP tools from servers you configure
- Hand hard jobs to a local Hermes Agent when you enable the bridge
- Remember facts you ask it to keep, on your own computer
- Talk with you in voice mode (hands-free)
- Learn your preferences through simple skill files
- Turn into a little cat, dog, fox, or bunny if you want something cuter than a pill
Your keys and memory stay on your PC. Mirrly doesn’t run its own cloud account or server.
- Open the latest release: github.com/Sadat-Rakib/Mirrly/releases/latest
- Under Assets, grab one of these:
- Mirrly-Setup.exe — normal installer (recommended)
- Mirrly-Portable.exe — no install; run from anywhere
Direct link: Mirrly-Setup.exe
- Double-click the file you downloaded
- If Windows SmartScreen pops up (“Windows protected your PC” / unknown publisher):
- Click More info
- Click Run anyway (or Keep anyway, then Run)
- That warning is normal for an unsigned open-source app. Nothing is wrong with your download.
- Mirrly opens as a small panel on your desktop (always on top)
- Follow the short walkthrough (click the logo anytime to see it again)
- Open Settings with
Ctrl+,
In Settings → your AI provider:
| Option | What to expect |
|---|---|
| Pollinations (default) | No key needed. Fine for a quick try. Weak at reading your screen. |
| Ollama | Local models at http://127.0.0.1:11434/v1. No cloud key. |
| Gemini / Groq / OpenRouter | Free-ish keys work well. Pick a vision model if you want screen help. OpenRouter can load Hermes-family models. |
| OpenAI / Anthropic / custom | Paste your key. Any OpenAI-style endpoint works. |
Speech tip: Listening, dictation, and voice mode need a Groq or Gemini key for speech-to-text.
Leave Mirrly open while you work.
| Want to… | Do this |
|---|---|
| Get help with what’s on screen | Ctrl + Enter |
| Explain what’s on screen | Ctrl + E |
| Solve a problem on screen | Ctrl + H |
| Ask anything | Type and press Enter |
| Cowork on a note | Notes button in the top bar |
| Dictate instead of typing | Mic next to Send |
| Talk hands-free | Waveform button in the top bar |
| Listen to a call | Mic in the top bar |
| Open Settings | Ctrl + , |
| Switch to a mascot | Settings → Appearance |
| Quit | Ctrl + Shift + X |
Try saying things like:
- “open notepad”
- “open youtube”
- “email alex@example.com about tomorrow’s meeting”
- “write a TODO plan in the cowork note”
- Skills — short
.mdfiles that teach Mirrly how you like things done (Settings → Skills). - Projects — scoped instructions, files, and memory under
Documents/Mirrly/projects/. - MCP — paste a server JSON list in Settings → MCP, enable it, hit Refresh.
- Hermes — enable the bridge and point at your Hermes CLI or HTTP endpoint for hard multi-step handoffs.
- Screen answers look made up → Switch to a vision-capable model (Pollinations alone often can’t see well)
- Mic / voice doesn’t work → Add a Groq or Gemini key for speech
- Panel disappeared → Check the system tray, or run Mirrly again from the Start menu
- Want a clean reinstall → Uninstall, then delete
%APPDATA%\mirrly-data.jsononly if you also want to wipe keys/memory
Need Node.js 18+.
git clone https://github.com/Sadat-Rakib/Mirrly.git
cd Mirrly/my-product
npm install
npm startBuild the Windows installer:
npm run dist:winWebsite (optional):
cd website
npm install
npm run devMirrly is meant to be a project you can read, reverse-engineer, and extend — not a black box.
- Desktop app: Electron + HTML/CSS/JS (
my-product/) - Agent loop: LangGraph (
@langchain/langgraph) - Tools: built-in desktop agency + Model Context Protocol (MCP) client
- Local models: Ollama (OpenAI-compatible API)
- Optional handoff: Hermes Agent bridge (CLI or HTTP)
- Cloud models: OpenAI, Anthropic, Google Gemini, Groq, OpenRouter, and more
- Website: React + TypeScript + Vite + Tailwind (
website/)
The important app code lives in my-product/main.js, my-product/renderer/, and my-product/src/.
Most “portfolio projects” online are another to-do app or another ChatGPT clone. They teach syntax. They rarely teach how to ship something that feels like a product.
Mirrly is open so you can:
- Study a real desktop agent (screen, voice, tools, memory, projects)
- Fork it and make it yours
- Turn ideas from it into your own startup
If you don’t want to download random binaries; fair. The whole codebase is public. Read it. Build your own version if you prefer.
The only thing I ask in return:
⭐ Star the repository if you find it useful, and tell me what you build. I’d genuinely love to see it.
Issues and PRs are welcome. Keep the desktop app easy to read.
MIT — use it, modify it, build a business with it.
Copyright (c) 2026 Mirrly.