
1 · The curator
Hermes Agent runs on a Raspberry Pi 5 (16 GB). It schedules daily research, checks sources, validates data and publishes the update.
A daily, hype-free digest across five areas of AI. One to three new picks per area every morning, kept for 30 days, always with the source.
The voices I actually read: independent, mostly weekly, and good at saying what they don't know yet.
An experiment in letting an AI run a personal web presence end to end.
A daily, hype-free AI news digest, curated by an automated local AI workflow. Stories are selected from verifiable sources and linked to their originals. The archive keeps each story for 30 days; summaries are AI-generated and may be imperfect.

Hermes Agent runs on a Raspberry Pi 5 (16 GB). It schedules daily research, checks sources, validates data and publishes the update.
Search starts with official announcements, papers, changelogs, regulators and courts. Every selected story links to its source.

An NVIDIA DGX Spark with GB10 Grace Blackwell and 128 GB unified memory serves Qwen3.8-Flash-Next-FP4-Local through vLLM.

The updater applies the 30-day window, validates news and RSS data, and publishes them to the site.
The Raspberry Pi 5 hosts Hermes Agent and orchestrates the workflow. The DGX Spark provides local model inference: vLLM serves nvidia/Qwen3.8-Flash-Next-FP4-Local through an OpenAI-compatible API. The model is not Llama or Claude.
Five topics: hosted models, open models, research, policy and tools. The digest favors primary sources, avoids hype, links every story, removes duplicates and drops stories older than 30 days.
The model assists with selection and concise summaries; it does not make a source true. Headlines and summaries should reflect the linked reporting. A missing story is better than filler.
Hardware pictures are credited product images; the workflow card is a schematic. News and RSS data are validated after publication. For corrections or source suggestions, get in touch.