Slotstream

Slotstream is an open source inference engine for running a very large language model on a Mac with limited memory. According to the vendor, it runs a 105 GB model on Macs with 16 to 64 GB of memory by loading parts of the model from the SSD as needed. It is written in native Swift on Apple's MLX and Metal and is tuned as one system for that model. The engine is the foundation of Sevra, the vendor's private local AI app, which is still in development and open to a waitlist. Sevra is described as an app that will choose and update the model for a given computer, work with files and tools the user grants access to, keep searchable conversation chapters and a journal, and let users create small apps or save repeatable tasks as skills. After the model download it is meant to work offline, with conversations and personal files staying on the machine and no account needed. A Mac version comes first, followed by Sevra OS for PCs. The audience is people who want capable AI models running locally on Apple hardware, and developers interested in the engine itself. The site links to the code and published measurements. No prices are published. Slotstream is open source, and Sevra is available only through the waitlist.