AlPa: Compile Once, Run Everywhere — Edge AI Inference on WebAssembly + WebGPU

Compile once, run everywhere

AlPa is a secure, portable, vendor-neutral AI inference platform built on WebAssembly + WebGPU, with Dawn as the native backend. The idea is simple and radical at the same time: compile once, and run the same inference stack everywhere — browsers, edge devices, servers, and embedded systems.

Why an inference platform instead of another runtime?

Most inference tooling stops at the model. AlPa treats the whole execution context as the product: who is allowed to run inference, on which hardware, against which model, under what license terms. That framing shapes every layer of the platform.

• WebGPU backend — GPU-accelerated inference via Dawn or browser WebGPU, delivering native-class performance on consumer hardware.
• GGUF model loading — progressive loading of llama.cpp-compatible models, so inference can start before the full model has arrived.
• OpenAI-compatible API — a standard chat completions interface means existing tooling and clients work with AlPa out of the box.
• Browser + native — one codebase for both browser (WASM) and native builds, with no forked runtimes to maintain.
• Multi-tenant — white-label support with per-tenant branding and SSO for platform-style deployments.
• Offline support — Ed25519-signed offline licenses with a grace period keep deployments running without a constant connection.

Portability is the product

WebAssembly and WebGPU together define a vendor-neutral hardware abstraction that now spans nearly every modern browser and a growing set of native environments. AlPa treats that layer as its runtime target: the same C/C++ core ships as a WASM module in the browser and as a native binary at the edge, backed by WebGPU acceleration through Dawn. Inference travels to where the data lives instead of forcing data toward the cloud.

License-gated inference, by design

AlPa's security model is enforced at the runtime layer, not bolted on. The C/C++ runtime validates a license token before every model load and every inference call. Heartbeat re-validation runs in the background, so revoked or expired licenses are enforced mid-session. Offline use is supported through Ed25519-signed licenses with a grace period — enabling air-gapped and edge deployments without weakening the licensing chain.

Getting started in five minutes

The quick start is deliberately short. Get a license token from the AlPa platform, pick your interface — the browser demo in Chrome 113+ or Edge 113+ for a guided start, or the CLI binary for command-line use — and point it at any llama.cpp-compatible GGUF file:

alpa_cli_demo --license "YOUR_TOKEN_HERE" --model /path/to/model.gguf

Type your message, press Enter, and the model responds through the OpenAI-compatible chat completions interface. That's the entire loop: licensed user, local hardware, portable runtime.

Where this fits in our portfolio

AlPa is currently shipping as version 1.0.0-beta and sits in LOTUS CHAIN Hub's portfolio direction as edge AI infrastructure — the vendor-neutral counterpart to DiagFlow's governed medical vision workflows. The same principle drives both: bring computation to the user's own hardware, with governance and security built into the runtime rather than added afterward.

AlPa is proprietary software. To learn more about the platform, the security model, or licensing for your deployment, get in touch with the team.