Perplexity Portable Computer routes work between local, cloud

The strongest local-AI pitch is not replacing frontier models—it makes them economically viable by reserving expensive inference for genuinely hard work.

• –Local orchestration, planning, tool routing, search indexing, and task queues reduce recurring cloud usage.
• –Selective escalation preserves access to frontier reasoning, web research, and connected apps when local models hit their limits.
• –The privacy model is compelling because cloud handoffs require permission, but users still need clear visibility into what data crosses the boundary.
• –DGX Spark’s cost and Linux-first availability make this an early adopter platform, not yet a mainstream desktop product.
• –The broader pattern points toward hybrid agents that optimize every task for privacy, latency, capability, and cost.