I built pneuma because I kept running into the same friction: needing a quick tool, then spending 20 minutes finding an app, installing it, configuring it, only to use it once. I thought: if AI can generate working code reliably, why not just describe what I need and have it appear? The technical approach: an LLM generates a self-contained Rust module, rustc compiles it to WebAssembly in under a second, Wasmtime JIT-compiles it, and wgpu renders everything on the GPU. If compilation fails, the error is automatically fed back for correction. About 90% first-attempt success rate. Agents are sandboxed (one crash can't affect anything else), they persist so you can reuse them, and they can talk to each other. There's also a community store for sharing agents. Would love feedback on the concept and whether this is something you'd actually use day to day.

pneuma

An OS where software materializes from a prompt

An OS where software materializes from a prompt