Perplexity launches Portable Computer with Nvidia for fully local AI agents

Aug 25, 2026

Perplexity has partnered with Nvidia to launch Portable Computer, a local-first AI agent platform that runs entirely on-device to eliminate cloud token costs and enhance data privacy. Debuting on Nvidia's DGX Spark and Linux systems, the software packages local models like Qwen 3.8 27B and PPLX 27B with an OS-enforced sandbox. It requires high-end hardware, including Nvidia RTX GPUs with at least 24GB of VRAM. The system operates locally by default, utilizing a hybrid workflow that requests explicit user permission and performs PII screening before escalating complex tasks to cloud models.

Portable Computer local AI launch

▪Portable Computer packages the local model, inference engine, agent harness, tool sandbox, and app connectors into a single system to simplify local AI deployment
▪Perplexity launched Portable Computer on August 25, 2026, a local-first version of its agentic Computer platform that runs AI models entirely on-device
▪At launch, Portable Computer users can run the Qwen 3.8 27B model or Perplexity's post-trained PPLX 27B model, with Nvidia's Nemotron 3.5 Lightning coming soon

Nvidia DGX Spark partnership

▪Nvidia's partnership with Perplexity on Portable Computer signals a strategic shift to validate local AI hardware for practical, non-hobbyist enterprise workloads
▪Perplexity co-designed Portable Computer in partnership with Nvidia, launching first on Nvidia's DGX Spark desktop supercomputer

Hardware requirements RTX GPUs

▪To run Portable Computer, Windows and Linux PCs require an Nvidia RTX GPU with at least 24GB of VRAM, such as a GeForce RTX 3090 or newer
▪Portable Computer is available on August 25, 2026, for paid Pro, Max, Enterprise Pro, and Enterprise Max subscribers on Linux, with Windows support coming in September 2026
▪Running Portable Computer on a DGX Spark requires a GB10-class superchip, 128GB of memory, and at least 1TB of storage

Local-first privacy architecture

▪Work completed locally on Portable Computer consumes zero billing credits or per-token fees, keeping sensitive data within the local device boundary
▪Portable Computer executes code and tool calls inside an OS-enforced sandbox that disables tool execution entirely if the sandbox is unavailable
▪Perplexity designed a minimal agent harness for Portable Computer to prevent small local models from degrading when context exceeds 100,000 tokens

Cloud escalation permission workflow

▪On Terminal Bench 2.1, escalating local Qwen execution to a cloud-based Claude Opus 5 advisor achieved a 73.0% score at an estimated $0.415 per task
▪Portable Computer starts all tasks on-device by default and prompts the user for explicit permission before escalating any individual step to cloud models
▪Before escalating a task to the cloud, the Portable Computer harness runs a PII classifier over the context to show users exactly what data will leave the device

Enterprise use cases

▪Portable Computer targets industries with strict data residency or confidentiality constraints, including finance, legal, healthcare, government, and defense
▪Portable Computer supports local integrations with Google Drive, Gmail, Slack, and GitHub, allowing local data analysis to be pushed directly to external channels