Perplexity's AI agent now runs locally on high-end RTX PCs
Po / Money and Products desk
Perplexity’s AI agent no longer has to live entirely in the cloud. On the right Windows hardware, it now runs on the machine itself.
What launched
Perplexity brought Portable Computer, the local version of its Perplexity Computer agent, to its Windows app. The company announced it on September 14, 2026.
The app needs an Nvidia GeForce RTX or RTX PRO GPU with 24GB or more of video memory, and it is available to Perplexity Pro and Max subscribers.
How it handles data
The agent works on the user’s files locally. According to Nvidia, it asks permission before sending anything to cloud models.
Why it matters
Perplexity’s earlier Windows agent ran in the cloud; this version runs the model on the PC. For people and companies who do not want documents leaving their machines, an agent that has to ask before using the cloud is a meaningful change in the default.
The trade-off is hardware. Work that runs on the device does not use Perplexity Computer credits, but it still requires a Pro or Max subscription and a graphics card with at least 24GB of video memory.
Sources
Related
Nvidia, Palantir and Booz Allen restrict Claude over data fears
The Information reported on September 14, 2026, according to Reuters, that Nvidia, Palantir and Booz Allen are restricting their use of Anthropic's models over concerns about their proprietary data. Nvidia limits Claude to less sensitive work, Palantir wants a zero-data-retention guarantee before offering Anthropic's models through its software, and Booz Allen barred staff from running Anthropic's commercial model on cybersecurity projects that touch proprietary data. On September 1, Anthropic announced Enterprise Frontier Safeguards, which let enterprise customers keep monitoring data in their own cloud accounts under their own encryption keys.
Perplexity built its search database with two engineers and AI agents
Perplexity published research on September 15, 2026 describing CobbleDB, a key-value store purpose-built for the repeated batch reads that serve prepared web page records to its search product. It replaced DynamoDB in production. Median batch-read latency fell from 31.4 to 5.60 milliseconds and the p99 from 123 to 24.2 milliseconds, and the company's internal cost model estimates at least 20 percent savings against DynamoDB. Perplexity says two human engineers and hundreds of proactive, always-on AI agents built the core infrastructure in two months, and that it plans to open-source CobbleDB.
A modded DLSS 5 redraws GTA V's paused frame into smears
A widely shared clip shows an unofficial modded build of NVIDIA's DLSS 5 running while GTA V sits on its pause menu. With the scene frozen there are no fresh motion vectors or depth information, so the model keeps reprocessing the same still, blurred frame and takes its previous output as its next input. Faces and textures drift further from the original with every pass until the image collapses into dripping smears of colour. It is a mod rather than shipping NVIDIA behaviour.
ANOTHER News is published by ANOTHER, an AI-native content agency. Daily coverage also runs on Instagram.