Web AI_Parallaxis R&DBrowser SLM prototype

Running a website assistant entirely in the visitor’s browser_

Parallaxis tested a WebGPU-first architecture with small quantized models, cached assets, retrieval over approved website content, capability detection, and a non-AI fallback for unsupported devices. The prototype define

WebGPU browser AI
WebGPU capability detection
Quantized in-browser inference

Challenge_

A website assistant that calls a remote LLM adds recurring cost, data transfer, latency, and availability dependencies. A browser model avoids those costs but must download quickly and remain grounded in a narrow knowledge set.

Approach_

Parallaxis tested a WebGPU-first architecture with small quantized models, cached assets, retrieval over approved website content, capability detection, and a non-AI fallback for unsupported devices.

Outcome_

The prototype defines where private, zero-API website assistance is viable and where device memory, first-load size, or answer quality still justify a server model.

Overview

Parallaxis tested a WebGPU-first architecture with small quantized models, cached assets, retrieval over approved website content, capability detection, and a non-AI fallback for unsupported devices.

The engineering decision

The product promise is deliberately narrow: answer from approved site material, disclose when the device cannot run the model, and never invent company facts outside the retrieval context.

How the system works

The implementation separates intake, validation, state changes, side effects, and reporting. That separation makes failures visible and allows one layer to change without rewriting the entire workflow.

Operational users see explicit statuses and exceptions; technical teams retain identifiers, timestamps, versions, and logs needed to reproduce a result.

Outcome