Field note
Perplexity Portable Computer treats local execution and cloud escalation as one explicit agent-runtime boundary rather than two unrelated products.
The release is timely because stronger local models and desktop AI hardware make privacy, latency, and per-task cloud cost configurable at the workflow level. Perplexity says Portable Computer runs its orchestrator, planner, tool router, scheduler, durable task queue, and local search index on an NVIDIA DGX Spark. Private data stays local by default, while cloud research, browser use, connected apps, or frontier models require user-authorized escalation when device content must leave the machine.
A local orchestrator attempts the task on device, keeps files and persistent state local, and exposes a permission checkpoint before sending selected context to a remote capability. Agent builders should label every boundary crossing, minimize transferred context, retain a local audit record, and make cloud fallback optional rather than invisible.
Perplexity's launch claims describe its own hardware and model stack; independent tests must still measure task coverage, leakage resistance, power, latency, and failure recovery. Revise the conclusion when independent workloads show how often cloud escalation is needed and whether permission prompts remain understandable across long-running tasks.