Perplexity has launched Portable Computer, a local version of its agentic “Computer” platform, according to VentureBeat. The report says the product runs entirely on hardware users already own, rather than relying on remote token-metered inference. The initial hardware scope is Nvidia-heavy. VentureBeat says Portable Computer starts with Nvidia’s DGX Spark desktop supercomputer and Linux machines equipped with Nvidia RTX graphics processing units. Techmeme also summarized the launch as a fully on-device AI agent platform starting with Nvidia DGX Spark and RTX Linux PCs. The core claim is economic as much as technical: the product is described as having zero token costs. Based on the provided summaries, that appears to mean the agent runs locally on supported machines instead of charging per remote model call. The reports do not provide enough detail to verify pricing terms, performance, supported models, or what tasks the agent can complete. This is a developing item because the cluster contains one original report, from VentureBeat, and one aggregator item pointing back to that same report. The launch itself is consistent across the items, but the available summaries do not establish broader availability, minimum hardware requirements, enterprise terms, or whether additional platforms are planned. Who benefits: Users and teams with supported Nvidia hardware are the clearest beneficiaries named in the reports. Perplexity also gets a path to distribute its agentic platform outside a purely cloud-token model. Who's exposed: Teams without Nvidia DGX Spark or RTX Linux PCs are outside the initial hardware scope described in the summaries. Cloud-only AI agent workflows may face comparison pressure where local execution is practical, but the reports do not quantify that impact.