PrismML

Meet PrismML

PrismML develops compact AI models for applications that need to run on laptops, phones, and other devices with limited memory. Its Bonsai family uses low-bit model weights to reduce storage and inference requirements, giving developers a way to build local assistants without relying on a remote model for every request.

The current Bonsai 2 models accept text and images and support reasoning and tool calls. Developers can choose among several model sizes, download the weights, and use the documented runtimes and integration examples to build an application. PrismML also offers Bonsai Image for local image generation. The models are useful for projects where device capacity, response time, and keeping inference on local hardware shape deployment choices.