Perplexity launches a local AI agent with zero token costs

AI company Perplexity is expanding its capabilities into local AI services, unveiling a new local agent called Portable Computer. This local agent is designed to control AI costs, addressing user concerns about high cloud bills. Portable Computer is built upon a hybrid agent inference orchestrator that connects cloud and local inference, outperforming cloud-based agents in accuracy while utilizing Nvidia DGX Spark workstations to run local AI models.

These models, including Nvidia Nemotron 3.5 Lightning, Qwen 3.6, and Qwen 3.8, are small and efficient, capable of handling complex agentic workflows. By running locally, Perplexity avoids per-token API fees and ensures privacy and intellectual-property protection by keeping tokens within the local device. Although Perplexity's current privacy policy does not directly address Portable Computer, the company has shown its commitment to user privacy through other services.

In benchmark tests, Portable Computer demonstrated competitive performance against other agents like Pi and Hermes, matching or exceeding them in accuracy while using fewer tokens. Perplexity's success hinges on the availability of open-weight models and powerful local hardware, with the possibility of Nvidia investing $30 billion in the company potentially paving the way for future revenue streams.