TensorFire

TensorFire is a framework for running neural networks in the browser using WebGL, with no setup or installation required. It lets applications perform deep learning directly on the client side, avoiding the need for native apps, compute farms, or server round-trips, and keeping data local to the user.

It has two parts: a low-level language based on GLSL for writing parallel WebGL shaders that operate on 4D tensors, and a high-level library for importing models trained with Keras or TensorFlow. It runs on any GPU regardless of CUDA support, and supports low-precision quantized tensors for browsers without full OES_texture_float support. The low-level API can also be used for general-purpose parallel computation such as matrix multiplication, linear systems, PageRank, cellular automata, and image processing.

TensorFire is aimed at developers building browser-based deep learning applications, including image classification, style transfer, text generation, and real-time webcam inference. The pages describe it as open source in progress, with API documentation still being written; no pricing, plans, or editions are stated.