Perplexity Computer

Perplexity's October 5, 2026 changelog adds Automations to Computer on the web. A task can run on a schedule or when something happens in a connected app, use Memory, Skills, and connectors, build on earlier runs, and pause for review before a consequential action. The same notes add inline charts, including TradingView charts for candlesticks, volume, and moving averages, and GPT-6.1 Sol for eligible paid subscribers in Perplexity and Computer, where it also powers Light effort.

On October 1, 2026, Perplexity announced 10 ready-made skills for Perplexity Computer, made for US American Express business card members who also have a Perplexity Enterprise subscription. The skills cover jobs such as cash-flow forecasts, tax-season prep, marketing drafts, and hiring. Members open Personal CFO from the connectors page, link a card through Plaid, and running a skill uses Computer credits.

On September 30, 2026, Perplexity CEO Aravind Srinivas said anyone can forward or CC [email protected] to hand the agent a task, with no Perplexity account. The agent finishes the work in the background and keeps the context from the email. Perplexity said those emailed tasks are free for a limited time.

On September 24, 2026, Perplexity brought Portable Computer, the version of its Computer agent that runs entirely on your own hardware instead of the cloud, to Windows PCs built on AMD's Ryzen AI Max series chips, including AMD's Ryzen AI Halo developer platform. It had launched in August 2026 for Nvidia's DGX Spark and RTX Linux machines, then expanded to Nvidia-powered Windows PCs in September. The AMD version needs at least 24GB of GPU-accessible memory and runs one of two local models, Qwen 3.8 27B or Perplexity's own PPLX 27B, along with the agent's scheduler and sandbox, entirely on the device. It's available now to Pro and Max subscribers through the Perplexity app for Windows.

Perplexity Research described a new way the company trains the model behind Perplexity Computer. It learns from real user sessions, copying steps that worked and correcting specific mistakes, like a wrong tool call, even when the session succeeded overall. In a live test with about 100,000 users in each group, failed tool calls fell from 2.24% to 1.77% versus an earlier trained version, a drop of about 21%.

Perplexity Computer now has an effort slider with four settings: Light, Standard, High, and Ultra. Each setting picks a different orchestrator model and how deeply it reasons before acting and delegating work to other models. Lighter settings use cheaper models and cost less, so spending tracks the setting picked and the work actually done. People who want more control can still pick a specific model and reasoning level by hand. It's live on the web now, with Android and iOS support planned but not dated yet.

On September 15, 2026, Perplexity announced that HP will pre-install its Windows app, including Perplexity Computer, on select HP PCs. The rollout starts with the HP ZBook Ultra G3a mobile workstation, with the app placed directly in the Windows taskbar. Perplexity says Computer now connects to more than 400 other apps and data sources, including Snowflake, Salesforce, HubSpot, and Autodesk tools. More HP Windows devices are expected to get the app in the coming months.

On September 14, 2026, Perplexity brought Portable Computer, the version of its Computer agent that runs entirely on your own hardware instead of the cloud, to Windows. It had launched on August 25, 2026 for Nvidia's DGX Spark and Linux machines with Nvidia RTX GPUs; now it's available through the existing Perplexity app for Windows (downloadable from the Microsoft Store) for Pro and Max subscribers. It needs an Nvidia GeForce RTX or RTX PRO GPU with at least 24GB of VRAM. The app runs the model, agent harness, orchestrator, and scheduler entirely on your PC, so sensitive data stays on the device and work done locally doesn't use Perplexity Computer credits.

On September 4, 2026, Perplexity added OpenAI's newly released GPT-6 Astra model to Perplexity Computer, available to Pro and Max subscribers.

On September 2, 2026, Perplexity open-sourced Lily, the local inference engine that runs the on-device half of Perplexity Computer's hybrid compute feature on Mac. Lily is built specifically to run the Qwen3.6-35B-A3B model on Apple silicon, using a Rust and Metal setup instead of general-purpose tools like Apple's MLX. In Perplexity's own tests on an M5 Max Mac, it processed prompts about 23% faster and generated text about 35% faster than MLX-LM, and needs at least 32GB of memory to run.

On September 1, 2026, Perplexity added hybrid compute to its Mac app. A single task can now split between cloud models, which handle reasoning and web search, and a local model on your Mac, which handles private files and sensitive information using an on-device privacy filter Perplexity open-sourced. It needs Apple silicon, macOS 15 or later, and at least 24GB of memory, and is only available on Pro, Max, and Enterprise plans.

On August 25, 2026, Perplexity launched Portable Computer, a version of its Computer agent that runs entirely on hardware you already own instead of the cloud. It works on Nvidia's DGX Spark desktop and on Linux machines with Nvidia RTX GPUs, built in partnership with Nvidia. Because it runs locally, tasks don't cost any tokens.