Firecracker microVM Preview
VULK's live preview system — every generated project runs in an isolated Firecracker microVM with sub-second cold-start, full Node + Vite, and a public URL. Replaces fragile WebContainer / iframe sandboxes.
Firecracker microVM Preview
Firecracker is the open-source virtual machine monitor that powers AWS Lambda and Fargate. It boots a Linux guest VM in tens of milliseconds with strong hardware-level isolation, rather than relying on browser sandboxes (WebContainer, StackBlitz) or shared containers (Docker). For an AI app builder this matters because previews need to run real npm install, real node, real pnpm dev, real native modules — not a simulated subset.
VULK runs every preview in its own Firecracker microVM on the dedicated preview server (webapp.vulk.dev). When a generation completes, the rootfs (Node 22 + Vite + Tailwind v4 + shadcn pre-warmed) is cloned, the project files are mounted, pnpm install and pnpm dev start, and a public URL with HMR is returned within ~12 seconds. Crashes are isolated to the VM; the host stays clean.
MCP Server (vulk-mcp-server)
VULK's official Model Context Protocol server, published on npm as vulk-mcp-server. Lets Claude Desktop, Cursor, and any MCP client trigger VULK generations, list projects, and stream previews.
BYOM (Bring Your Own Model)
A pricing pattern where the user supplies their own API key for the underlying LLM provider (OpenAI, Anthropic, Google, OpenRouter) and the platform passes the model cost through at zero markup. VULK supports BYOM on Pro and above.