Multi-AI — One prompt in. Every AI out.

🧰 The tools

📢

Broadcast

Type once, hit Send to all (or Enter). Every panel also has its own ➤.

🧩

AI picker

19 built-in AIs + add your own by URL — with an Advanced editor for custom selectors.

📚

Prompt library

Save prompts you reuse and drop them back into the composer in one click.

⬇️

Export / compare

Grab every panel's latest answer into one Markdown file — prompt on top, one section per AI.

⌘

Command palette

⌘/Ctrl+K to run any action or jump to a panel, no mouse needed.

⚙️

Settings

Default view, theme, paste-all, scroll-sync, and a User-Agent override — one home.

🍪

Import logins

Pull auth cookies from Edge/Chrome (macOS/Windows/Linux) so you skip re-logging-in.

✨ Highlights

📢 Broadcast to every panel — one prompt, all models, at once.
🧩 19 built-in AIs + custom sites — Claude, ChatGPT, Gemini, Grok, Perplexity, Copilot, DeepSeek, Qwen, Le Chat (Mistral), Kimi, Meta AI, Poe, Pi, You.com, Doubao, Groq, HuggingChat, Character.AI, LMArena. Add any site by URL; teach it selectors via Advanced.
🔐 Dedicated login window — a real, full-size browser sharing the panels' session (clean Chrome UA so Google/Cloudflare don't block sign-in). Log in once, everywhere.
🍪 Import logins from Edge/Chrome — decrypts your existing cookies on macOS, Windows and Linux.
🖼️ Paste an image to all panels — ⌘/Ctrl+V an image and it uploads into every AI (verified before the prompt sends).
🎛️ Built for focus — 2×2 grid ⇄ mobile columns, light/dark, drag headers to reorder, double-click to maximize, optional synced scroll, ↑/↓ prompt history, ⌘/Ctrl+K command palette.

Why no API keys?

These sites block plain web embedding (Cloudflare bot checks + anti-iframe rules). Multi-AI sidesteps that entirely: every panel is a genuine Chromium browser view pointing at the real website. You use your own logged-in accounts — nothing is proxied, no keys, no per-token cost.

🛠 How it works

You — one prompt → broadcast() → inject into every panel

Claude webview Gemini webview DeepSeek webview Qwen webview …

Each site in the catalog (renderer/catalog.js) holds its URL, login URL, brand color, and optional CSS selectors for the input box and send button. broadcast() injects a tiny script into every enabled panel to fill the box and click send. Answers render in place on each real AI website, using your logged-in session.

Tech: Electron 33 (Chromium <webview> panels) · vanilla JS/HTML/CSS · no framework, no bundler, zero runtime dependencies. Auth persists in an Electron session partition; UI preferences in localStorage.

🚀 Quick start

# clone and run
git clone https://github.com/battastudio/MulAI.git
cd MulAI

npm install     # downloads Electron (~150 MB, one-time)
npm start       # opens the Multi-AI window

First run: click Log in (or Import) and sign in to the AIs you want. If a panel shows a Cloudflare check, complete it right inside the panel — it's a real browser. Then type a prompt and Send to all.

Build a standalone app bundle:

npm run package:mac      # → dist/Multi-AI-darwin-*
npm run package:win      # → dist/Multi-AI-win32-x64
npm run package:linux    # → dist/Multi-AI-linux-x64

No .env, no config, no API keys.

🤔 Honest limitations

Windows/Linux cookie import is code-complete but lightly tested — the decrypt math has self-checks, but real-profile import was verified on macOS only. On Windows, sqlite3 must be on PATH.
Selectors can break. Injection depends on each site's CSS selectors. When a site redesigns its input/send button, fix it in renderer/catalog.js (or the picker's Advanced fields).
Some logins are device-bound. Google/Gemini cookies won't import; use the Log in window for those.
Bot-detection sites. ChatGPT and Grok may challenge automation — complete any check inside the panel.
Per-site image/answer support varies. Image paste and answer-export are best-effort heuristics across differing chat DOMs.