1. Show HN: Halo – open-source, tamper-evident runtime evidence for AI agents (github.com)
Tamper-evident runtime records for AI agents. Hash-chained, dependency-free, verifiable by anyone. - bkuan001/halo-record
2. GitHub Freno: cooperative, highly available throttler service (github.com)
freno: cooperative, highly available throttler service - github/freno
3. But Nothing Has Changed on Our Side (cacm.acm.org)
4. Not Dark Yet (agoodhardstare.substack.com)
5. Show HN: InstantVideos.org – short documentaries in ~30 seconds (instantvideos.org)
Type any topic. Get a finished documentary video in under a minute. Powered by AI.
6. Reducing Doom Loops with Final Token Preference Optimization (www.liquid.ai)
Antidoom uses Final Token Preference Optimization to cut doom-loop rates in reasoning models by up to 90%, with minimal disruption to model behavior.
8. Sodium-ion "salt" batteries will revolutionize electric-vehicle and grid storage (www.newscientist.com)
Today, most rechargeable batteries are made from lithium ions, but sodium-ion alternatives could make battery tech much cheaper and offer other advantages
9. Show HN: LLM Thought Visualization (github.com)
Contribute to ninjahawk/Subtext development by creating an account on GitHub.
10. The Music of Destruction (thebaffler.com)
Grossman was split into two figures by the Soviet authorities, dissident Jew and voice of the people.
11. Show HN: HTMLDrive – serve HTML files from your Google Drive (html-drive.com)
12. Software Bonkers (craigmod.com)
13. Tiny-C Reference Manual Excerpt (permacomputer.solarpunk.au)
The project found this little manual instructive. You could say it is a kind of C flavour that is inspired by BASIC!
14. Show HN: L9gpu – GPU telemetry that ties each GPU to the K8s pod or Slurm job (github.com)
GPU Observability with workload attribution. One OTLP agent per node ties hardware metrics (NVIDIA, AMD, Intel Gaudi) to the K8s pod or Slurm job burning the GPU. - last9/gpu-telemetry
15. Show HN: Baerly-storage, a document DB that runs per request, no DB server (github.com)
Document database with no database server. Tiny, LLM-friendly TypeScript API; DB engine runs to completion in each JS request handler; state lives in S3/R2. - Gusto/baerly-storage
16. Show HN: Shellular – run Claude Code, Codex, Pi from your phone (shellular.dev)
Run Claude Code, Codex, Cursor, and OpenCode on mobile and write code from your phone. Shellular gives you secure remote access to your dev machine: a real agent UI, terminal, files, Git, ports, and browser DevTools, all from your phone.
18. Craig Mod built his own Good Reads (craigmod.com)
Go Knicks, A Better Goodreads, 'Cheap' Killer Films — Roden Newsletter Archive
20. Sites that block AI training crawlers mostly ignore the answer time bots (sitedex.dev)
We read the robots.txt of the top 10,000 sites. 38% of the GPTBot rules we could date were written in a single quarter of 2023 — right after GPTBot launched and the Times sued. Then every AI company quietly added a second bot, the one that answers questions in real time, and almost nobody wrote a rule for it.
21. The Making of Claude Code (www.anthropic.com)
The inside story of how Claude Code went from an internal CLI to Anthropic's coding agent, told by researchers, engineers and early users who built it.
23. Dropping in on Gottfried Leibniz (2013) (writings.stephenwolfram.com)
Visit to archive provides a look at Leibniz’s papers, notation, calculator. His near misses on universal computation, quest for the systemization & formalization of knowledge.
24. Weighing smoke: why AI visibility dashboards are mostly useless (betterthangood.xyz)
GEO dashboards launder daily noise into a decimal-pointed visibility score over an invented basket of prompts. What actually tracks, and the cheap alternative.
25. Show HN: Context Warp Drive – Deterministic context folding for AI agents (github.com)
Contribute to dogtorjonah/context-warp-drive development by creating an account on GitHub.
26. Show HN: Phobos – A tiny scale-free kernel language with tile-DAG support (www.joa-ebert.com)
29. Giving a domain a hill to climb: benchmarking as data activation (sparsethought.com)
benchmarks as a way to turn messy domain data into a measurable, optimizable substrate for models.