Running intelligent agents locally on devices represents a qualitative shift that grants users full productivity power while ensuring the confidentiality and privacy of sensitive data.
Article Index:
- Launching the portable computer in collaboration with Nvidia
- Working mechanism and local model processing
- Supported hardware and platform availability for subscribers
- Ecosystem evolution and competition in local intelligence
- Frequently asked questions
Launching the portable computer in collaboration with Nvidia
Perplexity launched the “Portable Computer” tool on Tuesday, a fully local and independent version of its AI agent platform that runs entirely on user hardware and keeps sensitive data and work logs completely away from cloud servers. Developed in close collaboration with chip giant Nvidia, this product represents a strategic shift in how smart agent privacy is handled by ensuring that the orchestrator, sub-agent models, and execution environment run directly within the user’s local device.
This move addresses the growing demand from professionals and enterprises who want to leverage complex task automation and deep search capabilities, but are prevented by strict regulatory policies or intellectual property leakage concerns from uploading their sensitive data to the public cloud.
Working mechanism and local model processing
The Portable Computer system relies on running a specially trained language model comprising 27 billion parameters locally on hardware, alongside making the 27-billion-parameter “Qwen 3.8” model available, with upcoming support for Nvidia’s “Nemotron 3.5 Lightning” model. The system operates without any cloud dependency for most daily tasks and commands, allowing users to execute workflows without consuming paid token limits.
When a specific task requires advanced reasoning capabilities that exceed the capacity of local models and necessitates cloud frontier models, the orchestrator requests explicit user consent before routing the request to the cloud—a process Perplexity describes as “governed by user gateways, marked with personal data protection, and restricted to text routing only.” The company emphasizes that the local 27-billion-parameter model with its built-in environment achieves an 82.6% success rate in benchmark tests based on real-world scenarios.
Supported hardware and platform availability for subscribers
In its initial phase, the new product is available on Nvidia’s DGX Spark super desktop computer, equipped with a massive 128GB unified memory and up to 1 petaflop of AI processing power, as well as Linux systems equipped with Nvidia RTX graphics cards featuring at least 24GB of video memory. Support for Windows systems is expected to launch next September.
Perplexity announced the availability of the Portable Computer to all Pro and Max tier subscribers at no additional cost beyond their current subscriptions, and the service is also automatically available to Enterprise Pro and Enterprise Max subscribers to safely boost team productivity.
Ecosystem evolution and competition in local intelligence
This launch is an extension of Perplexity’s “Computer” platform, which first appeared in February 2026 as a multi-model cloud agent for cognitive tasks, before expanding to Windows desktops in July via the “Personal Computer” program. The new Portable Computer adopts the opposite approach by prioritizing local execution and data security at the expense of cloud convenience, putting it in direct competition with a growing sector of local agents.
Perplexity stated in its launch release: “Continuous developments in models, chips, and electronic devices will continue to constantly elevate local capabilities first,” reflecting its firm conviction that the future of intelligent computing will strongly trend toward decentralization and edge data processing.
Frequently asked questions
Question: What is Perplexity’s Portable Computer platform?
Answer: It is a smart agent platform that runs entirely locally on user devices in collaboration with Nvidia to protect the confidentiality of sensitive data and files.
Question: What hardware is required to run the platform locally?
Answer: It requires an Nvidia DGX Spark computer or Linux systems and Nvidia RTX graphics cards with at least 24GB of video memory.
Question: How does the system handle complex tasks that require cloud models?
Answer: The system requests explicit and direct permission from the user before sending abstract texts protected from personal data to be processed via cloud models.
