The October 23 announcement describes the model as a 4B decision model and says it can answer a question about text or an image with a probability.
The application exposes a local Decisions API at 127.0.0.1, allowing software on the machine to make requests without a Perplexity API key. Perplexity says the local request format matches its hosted Decisions API, with the hosted endpoint requiring a URL change and authentication.
According to the linked documentation, first launch downloads a 5.8 GB q8_0 GGUF build. The stated requirements are an Apple-silicon Mac, macOS 13 or later, 16 GB of RAM, and about 6 GB of disk space. Examples include labeling private files, a pre-commit secret-scanning hook, and screenshot sorting.
The announcement does not establish the model's licensing terms, measured inference speed, accuracy beyond Perplexity's description, or full feature parity with the hosted service.