Reflex

All projects

Apps & Tools

A WebGPU-powered local LLM tool for structured decisions and probability estimation.

byKshetrajna Raghavan

PlayableOpen source

Reflex media is blocked

Allow external media to connect to the provider and play this content.

Description

Reflex is an experiment in running small multimodal LLMs locally in the browser via WebGPU. It allows users to provide a 'state' (as text or an image) and ask questions with fixed answers, returning precise probability distributions for each option without uploading data to any server. It is presented as an open re-creation of TypeSafe's Jev.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: An impressive demonstration of local, browser-based multimodal inference for structured decision-making using WebGPU.
Attributed to: @kshetrajna
Built with: Jev GPT-5.5 Pro LTX 2.3 Pro 3.1 Pro Vidu Q3 Pro Seedream 5.0 Pro Seedance 1.5 Pro Flux 2 Pro Nano Banana Pro Gemini 3 Pro Gemini 2.5 Pro GPT Image 1.5
Links: kshetrajna12.github.io
View tweet
View GitHub repo
Tags: #webgpu #local-llm #structured-output #multimodal

Discovered: X
First posted: X · Sep 18, 2026
Listed: Added 2h ago
Build: Strong
Models: Moderate

Reflex

Description

Timeline

Reflex

Description

Timeline