MBS Workbench is a desktop IDE where everything runs on your own GPU. Load any GGUF model (Llama, Qwen, Mistral) with CUDA acceleration. A 35-tool autonomous agent reads/writes files, runs terminal commands, and iterates on tasks. Full Monaco editor with LSP, DAP debugger, native Git, and integrated terminal. Generate images offline with Stable Diffusion. Fine-tune LoRA/QLoRA adapters. Deploy to Docker, Kubernetes, or 10 cloud providers. Private beta — free access for 30 days.

AppSignal — Real-time monitoring that helps you ship with confidence

Real-time monitoring that helps you ship with confidence

Hey Product Hunt! 👋 I'm Balasundaram, founder of MBS Workbench. I've been building this for months — 250,000+ lines of Rust and TypeScript. Every AI coding tool today sends your code to the cloud, require internet, all have monthly fees, and all route your proprietary code through someone else's servers. For defence, healthcare, and finance teams, that's a dealbreaker. ITAR and HIPAA literally prohibit it. So I built MBS Workbench: a complete IDE where everything runs on your own GPU. No cloud. No subscriptions. Your code never leaves your machine. • Local LLM inference — any GGUF model, CUDA-accelerated (10-12 t/s on 30B 3Ab with 4GB VRAM) • 35-tool autonomous agent — reads/writes files, runs terminal commands, iterates on tasks • Full code editor — Monaco, LSP for 9 languages, DAP debugger, native Git • Image generation — Stable Diffusion offline • Model training — LoRA/QLoRA fine-tuning • One-click deploy — Docker, K8s, 10 cloud providers Current status: 81 phases shipped. 4,600+ community members. 0 security advisories. Private beta open. Try the beta. Break it. Tell me what sucks — your feedback shapes the product. Get access: mbsworkbench.com 🙏

AppSignal — Real-time monitoring that helps you ship with confidence

Real-time monitoring that helps you ship with confidence

MBS Workbench

Local AI dev platform — code, LLMs, agents, all offline

Local AI dev platform — code, LLMs, agents, all offline