AI search provider Perplexity introduced a local-first version of its agent platform, named Portable Computer, on August 25, 2026. Developed in partnership with Nvidia, the system is designed to execute knowledge-intensive agent workflows without mandatory cloud dependencies. At launch, the entire software stack operates on Nvidia DGX Spark desktop systems, with support for standard consumer RTX PCs scheduled to follow.
The technical architecture of Portable Computer integrates a local orchestrator, a structured task queue, and an on-device search index. The system relies on local foundation models, specifically Qwen 3.8 27B and the fine-tuned PPLX 27B. Because all processing occurs directly on the local hardware, routine operations generate zero recurring cloud inference costs and prevent sensitive operational data from leaving the local environment.
To handle edge cases and broader research queries, the platform incorporates a hybrid escalation mechanism. When a specific task demands advanced reasoning or external web searches, Portable Computer can consult high-capacity cloud advisor models, such as Claude Opus 5. This transition to external compute does not happen automatically, requiring explicit approval from the user before any data is sent off-device.
This release highlights a broader industry shift from purely cloud-hosted AI agents toward local edge execution. By leveraging capable 27-billion-parameter models on dedicated desktop hardware, automated processing of internal documents becomes viable for privacy-sensitive enterprise environments. The deployment model shows how hybrid setups can combine strict local data sovereignty with on-demand access to massive frontier models.
As support expands to standard Nvidia RTX hardware, Portable Computer is positioned to reach developers and independent teams without specialized enterprise rigs. The architecture illustrates that on-device inference has matured beyond basic chat interfaces into comprehensive task pipelines capable of orchestrating multi-step workflows locally.