Perplexity Launches 'Portable Computer' to Shift AI Processing On-Device
The new local version of Perplexity's agentic platform eliminates cloud dependency to slash token costs and enhance data privacy.
Perplexity has launched "Portable Computer," a local iteration of its agentic platform that runs entirely on a user's own hardware. The move marks a strategic shift toward on-device processing to eliminate cloud dependency and give users total control over their data.
According to the company, the offering is designed to run on local hardware, specifically targeting Nvidia DGX Spark and RTX Linux PCs. By moving the runtime—including the orchestrator LLM, subagent LLM, and agent harness—directly onto the user's machine, Perplexity removes the need for constant cloud communication. This architecture allows for zero token costs for all local work, removing the recurring operational expenses typically associated with cloud-based LLM inference.
The Shift to Edge AI
This launch aligns with a broader industry trend toward "Edge AI." For years, the AI sector has relied heavily on massive cloud clusters to handle the computational load of large language models. However, this model creates two primary friction points: the exorbitant cost of inference tokens and growing concerns over how sensitive data is transmitted and stored in the cloud. By shifting the processing load to the edge, companies can bypass the bottleneck of centralized servers.
Implications for Privacy and Cost
For Perplexity, the transition to on-device capabilities serves a dual purpose. First, it significantly reduces the company's overhead by offloading the compute costs to the end user's hardware. Second, it creates a powerful value proposition for privacy-conscious users. Because the data remains on the local device rather than being sent to a remote server, the risk of data leakage is minimized, and the user maintains absolute sovereignty over their information.
What to Watch
As Portable Computer rolls out, the industry will be watching to see how effectively local hardware can maintain the performance levels of cloud-based agents. While the current focus is on high-end Nvidia hardware, the long-term success of the initiative may depend on whether these agentic capabilities can be optimized for a wider range of consumer-grade devices. It remains to be seen if this will prompt other AI search and agent platforms to accelerate their own local-first roadmaps.