Perplexity partners with Nvidia to launch Portable Computer, a fully local AI agent with zero token costs

Summary

Perplexity has launched the Portable Computer, developed in collaboration with Nvidia. This platform is a fully local AI agent infrastructure that runs on RTX 3090 or higher GPUs or DGX Spark hardware. It offers zero token costs for knowledge-work tasks, and cloud transitions occur only with explicit user consent.

Why it matters

This is one of the most concrete initiatives to decouple serious AI agent workloads from cloud billing. The move directly challenges the token-based revenue model that underpins most AI services. For companies handling sensitive financial or personal data, keeping models, files, and inference entirely on-device removes a significant compliance and trust barrier. It also signals that Nvidia is shifting its strategy for local AI hardware from a hobbyist curiosity to a practical tool.

Highlights

• Token costs are zero for local operations. Cloud transitions require step-by-step user consent rather than a silent default.
• The co-designed model + minimal skeleton outperforms general-purpose frameworks: 82.6% vs. 74–77.6% in Perplexity’s internal benchmarks, and 66.7% vs. 43.9–50.2% on BrowseComp.
• OS-level control is always active. If this feature is unavailable, the agent disables itself instead of executing tools with full user privileges.

Critical take

All headline comparisons come from Perplexity’s own internal evaluations, with the company merely stating it 'plans to open-source' the data. The 24 GB VRAM minimum requirement, necessitating a DGX Spark or high-end RTX card, effectively redefines 'zero token cost' as 'zero marginal cost on hardware you paid thousands of dollars for.' The marketing language deliberately blurs this distinction.