Mechanism: The product runs planning, reasoning, and file processing locally, while allowing a user-authorized transition to cloud services for selected capabilities. Perplexity says Computer now runs its model, agent harness, and task execution fully on Apple silicon Macs. The company identifies Qwen 3.6 or PPLX Pro as the local models, and says planning, reasoning, and file processing can remain on device. It also says users may authorize cloud escalation for current information, browser use, connected apps, or advanced reasoning through one of 15+ frontier models. Observed facts: the announcement names Notes, Messages, Mail, and Numbers among supported Mac apps; it also says code and tool execution run in isolated sandbox environments with controlled access to files and connected apps.
Why now: Perplexity frames fully local execution as a further step beyond its earlier local-file and hybrid-compute options. The immediate first implication: Deployment architecture becomes a user-facing control over where planning and sensitive inputs are processed, not merely a behind-the-scenes model choice. Teams evaluating the mode should map which subtasks truly need network access before authorizing escalation.
Evidence boundary: The evidence is Perplexity’s product announcement; it does not independently validate data flows, security isolation, performance, or quality. The source describes capabilities and examples, not comparative benchmarks or an audit of whether every workflow stays local. First watch condition: Look for independent testing or product documentation that makes cloud escalation, permissions, and network behavior observable per task.
The design’s significance will depend on whether that boundary is operationally legible: a local model alone does not guarantee a local workflow if tools or an authorized escalation move inputs elsewhere. The useful question is whether users can inspect and govern those transitions without breaking the task. The source establishes the announced option; it does not establish how reliably users can enforce it in routine work.