One interface.
Multiple runtimes.
Lwati separates the application control plane from the browser intelligence and compute plane. That keeps local execution first-class without pretending the browser can replace every native development environment.
Laravel control planeAuth · projects · sync · policies · deployments
↓React / StudioBuild · Science · Learn
↓Local AIWebLLM · WebGPU
SciencePyodide · DuckDB · webR
WorkspaceBrowser storage · source · exports
CAPABILITY MODEL
Do not confuse “WebGPU exists” with “every model will run”.
Local AI needs real adapter, feature, limit and model-load checks. The rest of the workspace should remain usable even when a selected model cannot initialize.
Build editorWorks without GPU
Local generative AIWebGPU capability dependent
PythonWebAssembly / Pyodide
Analytical SQLDuckDB-Wasm
RwebR / WebAssembly
Heavy native workloadsBridge or remote compute path
OUTPUT
Ordinary software remains the deployment target.
The build environment may be AI-native. The production application does not have to be.
intent
→ agent
→ deterministic tools
→ source code
→ deployment