Leaker Aaron Perris uncovered evidence of an unreleased Apple Mac app called Harbor, designed to turn a Mac into a private AI inference server for running open-weight models locally. The app reportedly lets users connect multiple Macs over Thunderbolt 5 to pool unified memory, supporting larger models and higher concurrency than a single machine could handle, with automatic device detection and no manual setup required.
9to5mac.com
· 2026-11-10
New testing data compares how different hardware handles local large language models, measuring output in tokens per second. Nvidia's RTX 5090 generated 205 tokens per second on gpt-oss-20b, Apple's M4 chip produced about 24 tokens per second on a 7B model, and Nvidia's DGX Spark ranged from 38 tokens per second on Llama 3.1 8B down to 4.42 on its 70B version. Separate hands-on testing on a 2024 Asus Zenbook Duo showed Intel's OpenVINO toolkit offloading work to an NPU kept the laptop responsive, versus heavy CPU and memory strain when running the same model through Ollama.
engadget.com
· 2026-11-09
A new tool called Batch applies local AI processing to help users quickly identify their best images, addressing a gap left by Google Photos. The app works alongside existing photo libraries rather than replacing them, running its analysis on-device instead of in the cloud.
fastcompany.com
· 2026-11-07
OpenAI CEO Sam Altman said on the Cheeky Pint podcast that people running AI models locally on personal computers is an inefficient use of the world's limited DRAM supply. He argued that cloud-based inference through hyperscalers achieves better utilization than individual users hoarding memory chips for home use. His remarks came as Intel shares fell nearly 30 percent Friday amid a PC market slowdown that analysts linked partly to AI-driven component shortages.
futurism.com
· 2026-10-30
Opera announced it will ship open-weight AI models directly inside its browser, powered by the newly proposed WebAI web standard backed by Google and Microsoft. Users will be able to run AI queries locally without installing separate tools like Ollama, and the feature is opt-in, with nothing downloaded unless the user chooses to.
zdnet.com
· 2026-10-28
Georgia Tech researchers built ORISA, a tool that scans a device's working memory for remnants of deleted AI chatbot conversations. In tests using the ChatterUI Android app loaded with 100 fictional facts that were later deleted, ORISA recovered 26 of them from memory, more than the 14 the AI assistant itself could still access.
zdnet.com
· 2026-10-27
A new Entrepreneur analysis compares three local AI hardware options for developers: NVIDIA's DGX Spark, built on 128GB unified memory and the Grace Blackwell GB10 chip, offers the most mature CUDA-based software compatibility but has slower memory bandwidth at 273 GB/s. The piece frames local AI machines as tools for data privacy, cost control and predictable performance rather than replacements for cloud-based large-scale training.
entrepreneur.com
· 2026-10-23
Perplexity has introduced a local-first option for its Computer agent in the Mac app, allowing both the AI model and the task-directing agent to run entirely on-device, including offline. The feature lets Computer plan, reason and process files locally, working with apps like Notes, Messages, Mail and Numbers, and only sends data to the cloud with user approval.
9to5mac.com
· 2026-10-20
A Verge laptop reviewer describes starting to experiment with running AI models locally rather than relying on cloud services, citing privacy concerns about sharing personal data with companies like OpenAI, Google, Microsoft, and Anthropic. The testing includes Apple's new M5 Ultra Mac Studio and upcoming Windows RTX Spark machines with up to 128GB of RAM, starting with the open-source self-hosted tool Hermes Agent, which runs free when used with local large language models.
theverge.com
· 2026-10-11
Apple has updated its Mac desktop lineup, giving the Mac Mini new M6 and M5 Pro chip options and refreshing the Mac Studio, with both machines now emphasizing local AI processing. The base Mac Mini price has risen by about $300 compared to earlier models, though specs have also increased. A tech reviewer who has tested Apple desktops for over a decade recommends the M6 Mac Mini as the best choice for most buyers.
wired.com
· 2026-10-09
A Show HN post introduces Screen Memories (SCM), a macOS application that lets users search photos and every frame of video using natural-language queries, OCR, and spoken-dialogue search via Whisper. All processing runs locally on-device through a local vision model, with no cloud uploads or accounts required, and the app auto-imports watched folders and re-embeds content when models change.
github.com
· 2026-10-04
Nvidia and partners including Acer, Asus, Dell, Gigabyte, HP and MSI will release a 64GB version of the GB10-based DGX Spark system on October 23, priced from $4,999. The new tier sits alongside the existing 128GB configuration, which currently sells for roughly $7,000 to $9,000 due to memory shortages.
tomshardware.com
· 2026-10-02