The goal of the Perplexity Portable Computer is a more local setup for its AI computer plans. The company shared the news on August 25 and called Portable Computer a fully local take on Perplexity Computer. Perplexity says many AI assistants rely on outside servers for most of the work. Portable Computer is built to run the main Perplexity Computer runtime on the hardware near the user instead.
In Perplexity’s description, the orchestrator LLM, the subagent LLM, and the agent harness all run on the user’s own machine. It also claims this design removes the need for a cloud link just to run the runtime. Perplexity adds that this may appeal to people who want more say over where AI tasks and data are handled.
Perplexity Portable Computer: What It Is
Perplexity Portable Computer is based on the idea of an AI setup that can manage multiple parts at once. The goal is to handle harder jobs, not just answer a single question. With Portable Computer, Perplexity aims to run that same style on devices at hand. The hardware is powered by NVIDIA DGX Spark.
The company states that the full run happens on-site. That includes the parts that set up the work, handle smaller agents, and use the agent harness. It also differs from tools that look local. Some apps may show a local screen, but still send the main work to the cloud.
Perplexity frames Portable Computer as a fully local version.
A Runtime That Does Not Rely On The Cloud
One of the key points is the claim about no cloud need for the runtime. In practice, the core system pieces are meant to run on the user’s own hardware. They are not said to need steady contact with Perplexity servers for local AI work.
This setup may fit users who care about privacy.
When the runtime stays on local hardware, people may have more control over private items. That can matter when AI agents receive documents, files, or other data to finish a task.
Still, local use does not mean every AI feature is locked to the device forever. Perplexity also described a way to pull frontier models from the cloud when a job needs stronger reasoning.
Frontier-style Work Needs A Go-ahead
Perplexity shared details in a later post about tasks that need stronger reasoning. If the local part of the system thinks a frontier model is needed, it can pause. Then the system asks the person using it to approve the move to a cloud model. This keeps local work separate from optional cloud reasoning. Instead of sending a task out without telling you, it makes the choice visible.
Perplexity calls this process user-gated.
That step may matter a lot when a Portable Computer is used with private or sensitive data. Sensitive files do not leave the device. Perplexity also says sensitive documents stay on the user’s machine. The company adds that cloud requests get flagged for PII. It also says those requests are limited to text guidance.In contrast, the sensitive documents remain local. So the setup points to local AI handling the user’s data. Then, when needed, cloud frontier models can add extra reasoning.
For people with confidential documents, this split may be a key reason to use a Portable Computer.
It also mirrors a larger trend in AI design.
More systems now mix on-device processing with limited cloud help, instead of treating them as opposites.
Why The NVIDIA DGX Spark Piece Is Important
NVIDIA DGX Spark is the hardware base for this local AI setup. Perplexity choosing DGX Spark for Portable Computer shows how much people want stronger local machines. This matters as models and agent-style software ask for more computation.
When you run an orchestrator, subagents, and an agent all on your own system, the load is much higher than a normal desktop app.
So DGX Spark acts as a starting point for anyone who wants to test a more capable AI computer.
They can do that without sending the full job to the cloud.
The note provided does not share extra hardware details or speed results for the Portable Computer.
Because of that, the actual performance gap versus Perplexity Computer in the cloud is not yet clear.
A New Way To Build AI Agents
This launch also points to a shift in how AI assistants are made. Rather than using one model like a chat screen that waits, agent systems can coordinate multiple models and tools. They can then work through more complex jobs step by step.
Portable Computer, from Perplexity, pushes a large part of that setup to local hardware.
What you get is a mix of local AI work, agent coordination, and optional cloud reasoning. If privacy is a top concern, this split may matter more than people expect.
Conclusion
Perplexity’s Portable Computer on NVIDIA DGX Spark feels like a real move toward running agentic AI on the hardware people already have.
They say the orchestrator model, the subagent model, and the agent harness can all run on the user’s machine. They also say the main runtime does not need a cloud service.
At the same time, the user can choose to use cloud-based frontier reasoning when the job needs more than the local setup can handle.
Privacy is a key part of the pitch. They focus on keeping sensitive documents on the device and routing work to the cloud only with user approval.
The open issue is simple: how does this local setup hold up in day-to-day tasks? If it can run useful agentic AI while still keeping sensitive data on-device, it may be a strong option versus AI tools that rely mostly on remote systems.