Perplexity brings always-on AI systems for individual and enterprise use

Perplexity has launched Personal Computer and Computer for Enterprise, which are designed to act as digital proxies that can perform complex tasks on behalf of a person or a company.

While traditional AI has functioned mostly as a chatbot that answers questions, these new releases aim to be autonomous digital workers that can manage projects from start to finish.

Perplexity Computer functions as a virtual analyst or a chief of staff. The core idea behind this is 'AI is the computer' This means that AI does not just give you a list of links or a summary of text. Instead, it uses an orchestration harness of 20 frontier models to understand a goal and move across different tools to finish a task.

In this context, orchestration means AI’s ability to coordinate multiple AI programmes to handle various parts of a single job. It can keep working even after the user has stepped away from their desk.

While Personal Computer focuses on an individual's work, Computer for Enterprise is designed for the scale of an entire company. The logic is: what works for one person can also work for a team, reasoning across tools and keeping work going.

Always-on digital proxy

The Personal Computer runs on a dedicated Mac mini, which is a small desktop computer, which stays connected to Perplexity’s servers and user's local files 24/7. As it is always on, it can function as a persistent digital proxy for the user.

In short, Personal Computer turns your Mac into an AI agent.

Perplexity's co-founder Aravind Srinivas points out that the biggest disadvantage for a single person trying to build a massive company is: “They have to sleep at night."

Personal Computer can solve this challenge as it is designed to never sleep. It allows a user to orchestrate all of their tools, tasks, and files from any device at any time. Effectively, users can control the Personal Computer from their phone, while the actual machine sits on the desktop at home, working through tasks and merging local files with the power of the internet.

Since this system has access to personal files and applications, security is a critical factor.

Perplexity has included safeguards such as a full audit trail, which is a detailed log of every action the AI takes. Sensitive actions require the user’s explicit approval. A kill switch allows the user to stop the system immediately, if necessary.

Efficiency for modern business

The Enterprise version connects directly to the tools a company uses, such as Salesforce, a popular sales management platform, or Snowflake, a data storage service. This allows the AI to act as a bridge between different departments.

For example, a finance analyst can ask for revenue data, while a sales team pulls customer information at the same time. The Computer writes the necessary queries, runs them, and returns the results in an organised format.

Furthermore, teams can teach the Computer specific skills or workflows. It also integrates with Slack, the workplace messaging app, where it can help with coding, creating financial models, or building data dashboards without needing to wait for a specialist data scientist.

Perplexity says that, in a study of over 16,000 queries, its internal teams saved $1.6 million in labour costs by using the Computer for Enterprise system.

The company claims the AI performed the equivalent of over three years of work in just four weeks. The numbers are based on internal benchmarks from institutions such as McKinsey and MIT.

Interacting with AI in different environments

Apart from the two Computer platforms, Perplexity has also introduced Comet Enterprise. This is an AI-native browser designed for a managed office environment.

Much of a person's workday happens inside a web browser, and Comet is built to understand the context of various open tabs and automate repetitive tasks.

Admins in a company can control how Comet operates, deciding where it can take actions and where it should only answer questions. This is part of a responsible use framework that includes protections from partners like CrowdStrike to prevent sensitive information from being entered into the AI.

The broader goal of these releases is to change how we think about operating systems and how we interact with AI in different environments.

“A traditional operating system takes instructions. An AI operating system takes objectives,” Srinivas explains.

Instead of telling a computer exactly which buttons to click, a user tells the AI what they want to achieve, and the AI reasons out the best way to do it.