Perplexity launches AI agent that runs entirely on your computer
What's the story
Perplexity has launched a new version of its Personal Computer agent, called Portable Computer. The innovative tool leverages local AI models to fulfill user requests and can switch to cloud-based processing when needed and with permission. The launch comes as a response to some of the limitations associated with cloud-based AI solutions.
Features
The agent offers all standard commands and features
The Portable Computer provides access to all the standard commands, features, and tools of Perplexity's agent. This allows it to perform tasks with minimal user intervention.
The main advantage is that models, files, and work remain on your machine. You have control over when the AI should go online for its tasks.
It also supports integration with third-party services like Google Drive, Gmail, Slack, and GitHub for accessing cloud-based files and information.
Benefits
Local processing enhances speed, security, and cost-effectiveness
Local AI models are faster and more responsive than their cloud-based counterparts.
They also provide enhanced security as tasks are performed in an isolated sandbox on your machine, reducing the risk of data breaches.
Plus, running AI locally doesn't consume any billing credits, unlike cloud-based solutions, which can be expensive due to usage of credits and tokens.
Cost-effectiveness
Currently available for paid subscribers only
Perplexity has described the Portable Computer as a "local-first agent," emphasizing that sensitive data never leaves the device without permission.
The company also noted that local models carry no inference fee, making this system both private and cost-effective by design.
However, it's worth noting that access to this new local AI is currently limited to paid accounts only, Pro, Max, Enterprise Pro, and Enterprise Max subscribers.
Compatibility
Currently limited to Linux systems
Currently, the Portable Computer is only available for Linux systems. However, Perplexity has promised that it will be available for Windows users in September.
On the downside, local AI solutions like this one often require high-end hardware to run effectively.
For Linux users, an NVIDIA DGX Spark or another Linux machine with an NVIDIA RTX GPU is needed, while Windows users need a PC with an NVIDIA RTX GPU with at least 24GB of VRAM.
AI models
You can choose between 2 local models
The Portable Computer offers a couple of options for local models.
You can use Qwen 3.8 27B, a popular open-source model known for its speed and performance in coding, research, professional work, and more complex agentic tasks.
Alternatively, you can use a post-trained version of the model called PPLX 27B that has been refined beyond its original instructions to improve accuracy and efficiency.