LAB — Local AI Brief

BIOS v2026.04 — NVIDIA DGX Spark [GB10 · 128GB]

[ OK ] Ollama daemon started — gpt-oss:120b loaded (65GB MXFP4)

[ OK ] blogwatcher: 21 RSS feeds registered

[ OK ] fetch-scrape-sources.sh: 8 scraped sources ready

[ OK ] GitHub APIs: TensorRT-LLM · NeMo · Merlin

[ -- ] Awaiting cron trigger — next run 06:00 CST

[ OK ] ElevenLabs TTS · Drew voice · eleven_flash_v2_5

[ OK ] Cloudflare R2 · D1 subscribers · Resend MTA

[ OK ] Healthchecks.io watchdog armed

SYSTEM READY — Local AI Brief v1.0 ████████████ 100%

██╗      █████╗ ██████╗
██║     ██╔══██╗██╔══██╗
██║     ███████║██████╔╝
██║     ██╔══██║██╔══██╗
███████╗██║  ██║██████╔╝
╚══════╝╚═╝  ╚═╝╚═════╝
LOCAL AI BRIEF

AI news. Runs local. Lands at 6 AM.

3–4 minutes. 31 sources. Written by a 120B model on personal hardware. In your inbox before your first coffee.

name$

mail$

free during beta · no spam · unsubscribe anytime · MIT licensed

cat features.txt

Why LAB is different

01 // HARDWARE

Runs on personal hardware

Written by a 120B model on an NVIDIA DGX Spark in Eric's office. No ChatGPT. No cloud dependency. Nobody else controls it.

02 // FORMAT

Listen, don't read

3–4 minutes of natural-sounding audio. Your commute, your workout, your coffee. Not another podcast you'll never finish.

03 // ACCURACY

No hallucinations

The model is instructed never to invent facts. If it can't confirm something, it says so. Shorter some days — but always grounded.

cat pipeline.sh | head -4

How the pipeline works

→

01_SCAN

Gather the news

31 sources checked at 6 AM — 21 RSS feeds, 8 scraped sites, 3 GitHub APIs. New stories deduped against the briefed-articles log.

→

02_ANALYZE

AI writes the script

Headlines sent to a local 120B model. Structured audio script written. No URL fetching, no hallucination.

→

03_VOICE

Text → audio

Script cleaned by clean-for-tts.py then synthesized by ElevenLabs. Drew voice, MP3 output.

04_DELIVER

Delivered to you

R2 upload → GitHub Pages → Resend email → Taildrop to iPhone → Healthchecks.io ping.

mpv LAB-2026-04-05.mp3

Hear a briefing

LAB — audio player — Drew @ ElevenLabs

LAB_BRIEFING · APR 05, 2026

LAB // Saturday Briefing

31 sources monitored · ~4 min · gpt-oss:120b on DGX Spark

0:00 0:00

choose_your_path.sh

Choose your path

💊

blue-pill.txt

free forever · no credit card

Daily briefing every weekday
Full audio + text formats
Delivered before 6 AM
31 sources monitored

Stay comfortable. Just subscribe.

💊

red-pill.txt

☕

pay what you want · your call

Everything in free
Keep the hardware running
Fund new sources & features
Early access to experiments

See how deep the rabbit hole goes.

grep -i "faq" README.md

Questions

What makes this different from other AI newsletters?+

Most AI newsletters use ChatGPT. LAB uses a 120B model running on personal hardware — an NVIDIA DGX Spark. No tech company controls the content, the cost, or whether it keeps running. The full pipeline is MIT licensed and public on GitHub.

What sources does the briefing cover?+

31 sources: 21 RSS feeds (NVIDIA, Google AI, OpenAI, AWS, Hugging Face, VentureBeat, MIT, arXiv, Cisco, Dell, LangChain, and more), 8 scraped sites (Cohere, Mistral, Intel, AMD, HPE, Modal), and 3 GitHub release APIs (NVIDIA TensorRT-LLM, NeMo, Merlin).

How do I know the information is accurate?+

The model is instructed never to invent details. If it can't confirm something from the source, it says so. Some briefings are shorter — that's intentional. Everything you hear is real.

Can I share this with my team?+

Yes — have each person sign up. It's free. If your team finds it valuable, the Pay What You Want option helps cover hardware and running costs.

What if there's no major news on a given day?+

The pipeline requires at least 3 fresh articles to run. No news worth covering means no briefing that day. You'll never get padding.

Who built this?+

Eric Davis — Principal Consulting Engineer on the AI Factory team at CDW. Personal project, personal hardware. Full source at github.com/GitWizrd/lab-local-ai-brief. Only non-local services: ElevenLabs (voice) and Groq (chat widget).

./subscribe.sh --interactive

Start receiving transmissions

$ subscribe --free --briefing=daily --format=audio+text

Entering name and email to join the list...

name$

mail$

# Your first briefing arrives before end of week · unsubscribe anytime