Start with the morning list
Opening an inbox and working out who needs what is hardly a spectacular AI demonstration. Yet that is one of the jobs proposed for Perplexity Computer's fully local Mac mode: find requests and deadlines in Messages and Mail, then make a priority list in Notes. The appealing part, to my mind, is gathering the pieces of one job from several applications. The person's question remains wonderfully ordinary: what needs doing today?[1]
Before the task list comes a setup list. There is a supported Apple Silicon Mac, a paid Pro, Max or Enterprise subscription, the current application and a downloaded local model. The user then selects fully local mode. Perplexity recommends 36 GB or more of unified memory. That recommendation is not a hard admission rule, but having the Mac application installed is no guarantee of a comfortable experience either. The model choices are Qwen 3.6 and Perplexity's adapted PPLX Pro.[1]
What the list-maker can reach
Permissions are the next stop in everyday use. Computer reaches those applications and local files only when allowed. That access supplies the raw material for the morning list: it cannot discover a deadline in a message it cannot read. Making the list and carrying out everything on it are also different requests. Perplexity's examples include preparing follow-up replies as drafts for review. That review point is a more useful detail than a demonstration that compresses an entire working day into one instruction.[1]
A local model does not turn networked applications into offline ones. Sending an email or looking for current information still requires a connection. Computer lets users separately authorize cloud help for browser use and connected services. The same morning workflow can therefore contain different boundaries: summarizing existing correspondence can finish on the device, while sending a new message outward is another step. An intelligible experience needs that transition to remain understandable instead of becoming a surprise menu halfway through the job.[1]
The job that repays the setup
It would be unfair to dismiss the setup effort outright. For someone already using a Mac with suitable memory and working with the same files every day, downloading a model once can be a reasonable exchange. Work completed locally does not consume Computer credits, although the subscription requirement remains. For an occasional user, learning the application permissions and two execution modes represents a larger burden. Different valuations of the feature do not mean that one of those users has misunderstood technology.[1]
For me, the unresolved question is the review burden rather than how polished the morning list looks. If finding a missed deadline requires rereading every conversation, automatically gathering the files addresses only part of the job. That is a criterion for the advertised task, not a claim that I encountered such a failure. A repeatable independent test using the same messages can make the correction work after setup visible. The current examples describe the jobs the product proposes; they do not measure time saved on an ordinary morning.[1]