Perplexity unveils a Computer feature that splits tasks across local models and cloud-based models, to keep private data on-device and maximize token efficiency
Context & Ripple Effects
Perplexity has been expanding Computer from a cloud-based general-purpose worker that routes work across multiple AI models into both a macOS-oriented Personal Computer and an enterprise offering. This feature adds a routing layer based not only on model choice but also on where work is performed.
The move also sits alongside Perplexity’s reported commitment to deploy models through Microsoft Foundry, making the boundary between local execution and hosted inference strategically important rather than incidental.
First-order effects
- Perplexity Computer can keep private inputs on-device while sending other portions of a task to cloud models, changing how its users’ work is allocated across its local and hosted capabilities.
- Perplexity gains another way to manage token use in Computer, alongside its existing multi-model routing approach.
Second-order effects
- The enterprise version of Computer becomes more relevant for workloads where data handling constrains use of fully cloud-based agents, while hosted models remain available for tasks that need them.
- Perplexity’s cloud-infrastructure partners and model providers may see demand become more selectively routed: cloud capacity is used for the portions of work not handled locally.
Third-order effects
- If hybrid routing proves usable in agent products, competition will increasingly center on orchestrating local and cloud models around privacy, cost, and task fit—not simply offering access to the largest hosted model.
- The pattern could make deployment architecture a key dividing line between consumer and enterprise AI agents, as vendors try to reconcile on-device control with cloud-model breadth.
The trend: AI-agent platforms are evolving toward hybrid local-cloud orchestration, using task routing to balance private-data handling, model access, and inference efficiency.