BroPilot — AI Browser Pilot & Assistant

AI Browser Pilot: navigate, search, deep research, and control web tasks with in-browser WebGPU and local AI models.

BroPilot is an autonomous AI browser pilot and research assistant that connects directly to local AI runtimes (in-browser WebLLM WebGPU, Apple Silicon MLX, Ollama, LM Studio, and cloud providers). Control your browser, automate complex tasks, clip research, and summarize web content with complete privacy and zero data leakage. ⚡ WHAT'S NEW IN VERSION 0.3.0: • ModelFit Hardware Profiler: Automatically profiles your GPU VRAM, system RAM, and unified memory architecture. Recommends native Apple Silicon MLX models or calculates FreeToken CPU+GPU hybrid offloading for Windows & Linux. • Direct In-Browser WebGPU: Run verified local models (Llama 3.2, Qwen 2.5, Phi-3.5) 100% in your browser with zero local server requirements. • One-Click Web Clipper: Dedicated header web clipper icon to save clean, formatted Markdown notes straight to your local research vault. 🎯 AUTONOMOUS GOAL AGENT & BROWSER CONTROL: Provide a high-level goal in plain English (e.g. "Find alternatives to a mini AI workstation GPU on Amazon and compare prices" or "Find the download link for dataset weights and copy the magnet link"). • Compact Accessibility Perception: The agent scans interactive elements with compact references and accessibility labels instead of fragile CSS selectors. • Reinforcement Learning Early Stopping: Evaluates page information sufficiency after each step and triggers early completion the moment answers are identified. • Built-in Safety Guardrails: Sensitive inputs (passwords, payment fields, deletions) automatically pause and require your explicit one-click confirmation. • In-Page Floating HUD: Real-time progress badges and boundary illumination let you monitor automated actions with an instant emergency stop button. 💻 100% LOCAL & PRIVATE AI (NO SERVER REQUIRED): • In-Browser WebGPU (WebLLM): Run Llama 3.2 3B, Qwen 2.5 3B/7B, and Phi-3.5 Mini directly inside your browser cache. • Local MLX Acceleration: Seamless integration with Apple Silicon native MLX models for blazing-fast inference on M-series Macs. • Ollama & LM Studio Integration: Automatic detection of locally running models at http://127.0.0.1:11434 and http://127.0.0.1:1234. • FreeToken Hybrid Offload: For PCs with discrete GPUs, BroPilot automatically calculates optimal layer splitting between GPU VRAM and System RAM to run larger parameter models without Out-Of-Memory crashes. 📚 RESEARCH & PRODUCTIVITY TOOLS: • One-Click Web Clipper: Save articles, GitHub repos, and academic papers with frontmatter metadata and structured Markdown. • Document & PDF Chat (RAG): Attach PDFs, Markdown, JSON, or CSV files to ground conversations directly in your documents. • Voice Dictation & Read Aloud: Hands-free voice speech-to-text dictation and natural speech audio responses. • Auto Form Filler: Local secure profile management to detect and autofill multi-step forms. 🔒 PRIVACY FIRST: BroPilot operates on-device. We do not track your browsing history, sell personal data, or transmit your page contents to third-party ad networks. Keyboard Shortcuts: • Command+Shift+A (Ctrl+Shift+A): Toggle BroPilot Sidepanel • Command+Shift+S (Ctrl+Shift+S): Clip current page