Mudrik · مدرك
Stop pasting screenshots into AI chats. Mudrik is an open-source Windows AI assistant that sees what you see — and answers, types, pastes, or clicks for you.
Watch the 10-second demo
Install in 30 seconds
Install the AI runtime
One global npm package. Open PowerShell and run:
$ npm i -g opencode-aiFirst launch triggers Windows SmartScreen — More info → Run anyway. Code signing on the roadmap.
Built for the cursor, not the chat window.
Cursor-anchored
The panel pops up on click alt+space — your target stays in clear view. No more switching to a separate AI chat app or browser tab.
Acts on any Windows app
Drives buttons, inputs, and menus through Windows UI Automation. Anywhere accessibility reaches, Mudrik reaches.
Reads your files when asked
Look up local docs and code without giving the AI free run of your disk. Reads stay scoped to the working directory.
Area capture
Press Ctrl+Space and drag a rectangle. Mudrik grabs the pixels and scans the UI elements inside — perfect for charts, dialogs, or anything UIA misses.
Bring your own model
Anthropic, OpenAI, Google, Groq, DeepSeek, Ollama, Z.AI, OpenRouter — paste your key in the settings panel. No terminal auth dance.
English + Arabic
Full RTL when Arabic is selected — UI strings, prompts, and panel layout all flip. More languages welcome via PR.
Four keys, that's it.
| Alt + Space | Panel slides in on the opposite side of your cursor. Mudrik scans the whole active window — every visible button, field, label, and value — with the cursor element as the focal anchor. All preloaded as context, no screenshot needed. |
| Ctrl + Space | Area select: drag a rectangle. The region is screenshotted and UIA elements inside are scanned. |
| Esc | Cancels whatever's happening: stops a response, exits area-select, or closes the panel. Your chat history is preserved. |
| Enter | Send — Shift+Enter for a newline |
Both global hotkeys are rebindable from the ⚙ menu.
Actions are text, not tool calls.
Mudrik asks the model to write its actions inline as plain HTML comments — action markers. The host parses them out, runs them through a sandbox, and shows the natural-language reply. No JSON-mode juggling. No vendor-locked tool API.
Works with any model — pick your provider:
Done. <!--ACTION:{"type":"paste_text",
"selector":"Body",
"text":"Hi Ahmed…"}-->Sandbox first.
Mudrik defaults closed. Dangerous capabilities are off, accessibility-style automation is allow-listed, and screen capture only happens when you explicitly attach a region.
| Shell / PowerShell exec | ✕ Off |
| Filesystem write | ✕ Off |
| Windows UI Automation | ✓ Allow-listed |
| Filesystem read | ✓ Working dir only |
| Screen pixels | ✋ Manual attach only |
Hi, I'm Abdullah Magdy.
A dev who got tired of pasting screenshots into AI Chats — so I built Mudrik on nights and weekends. It's open source; contributions and ideas are welcome 🤝