A user report has said local AI can handle some personal and work tasks without uploading sensitive data to the cloud.

A Verge reporter on Oct. 11 (local time) described using a local AI agent running directly on a PC for real work tasks.

The reporter installed the open-source Hermes Agent on an M5 Ultra Mac Studio with 256GB of unified memory. Hermes is a self-hosted AI agent app that runs on macOS, Windows and Linux. It is free if used only with local models. The reporter chose Qwen 3.8 Flash Next, with 125 billion parameters and a size of about 105GB. The reporter also connected a Telegram bot so it could be controlled on a smartphone.

The installation finished quickly. The problem was deciding what to ask it to do. The reporter first built a morning briefing that checks email and schedules and provides weather information. At first, the briefing kept failing because the Mac was asleep at the 7:30 a.m. run time. The reporter said it was not a task that required a $12,000 computer.

A more useful task was organising more than 400 Steam games. Steam does not automatically categorise games. After the reporter issued and provided a Steam Web API key, Hermes sorted the games by genre in a few minutes. It left intact categories the reporter had created, such as favourites and co-op games. The API key was discarded immediately after the task was completed.

The reporter also assigned it to analyse financial records and to draft a comparison table of specifications for a new laptop under embargo. The reporter said the deciding factor was whether sensitive data they did not want to upload to the cloud could be processed on their own computer. The reporter is currently building Python scripts with Hermes to automate laptop benchmark tests.

The reporter said even powerful local models are not magic assistants that can do anything. The morning briefing has already run into problems many times. The reporter said they would treat local AI like a computer program rather than a digital friend.