🛠️ Hacker News Tools

Google to pay SpaceX $920M a month for compute capacity at xAI data centers

Transformers are inherently succinct This paper is being published at ICLR 2026 (top AI conference), and was selected as one of three outstanding papers.

Harness engineering: Leveraging Codex in an agent-first world

Show HN: On-device transcriber that's 97% accurate at identifying speakers I’ve spent the last seven months building a tool I wish I’d had in my previous roles. MimicScribe is a macOS menu bar app that fits the &quot;AI notetaker&quot; category. It has accurate on-device speaker identification (a first possibly?), real-time meeting talking points for discovery calls, and a fully keyboard- and voice-driven interface.<p>I believe the accuracy of the speaker ID system is its biggest strength. I used fluid audio’s port of (<a href="https:&#x2F;&#x2F;github.com&#x2F;fluidInference&#x2F;FluidAudio" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;fluidInference&#x2F;FluidAudio</a>) Pyannote&#x27;s community-1 as a base. To improve accuracy, the system uses grammar structure cues from the Parakeet STT to mask by sentence. By taking a second set of samples within that mask for cluster assignment, it leverages the fact that most people don’t finish each other&#x27;s… sandwiches in business meetings. It tends to slightly oversegment, as I’ve found it much easier to merge segments or reassign a speaker than it is to untangle an incorrect merge. <a href="https:&#x2F;&#x2F;github.com&#x2F;MimicScribe&#x2F;benchmarks&#x2F;blob&#x2F;main&#x2F;diarization&#x2F;results&#x2F;RESULTS.md" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;MimicScribe&#x2F;benchmarks&#x2F;blob&#x2F;main&#x2F;diarizat...</a><p>The app provides in-meeting talking points using a prompt tuned for discovery type calls. It can suggest probing questions to help you extract more detail or helps you refocus on the big picture with “magic wand” type questions (e.g. “how would your ideal system work”). Getting low latency models to provide novel, relevant, and totally not hallucinated information is a bit of a reach and it tends to restate the transcript frequently but little gems do come from it sometimes so it’s best to think of it as a source of inspiration and be a vigilant gatekeeper.<p>It’s set up so recording can be started and ended via holding a keyboard shortcut instead of connecting to your calendar service. I prefer this for privacy and to keep transcript history from getting cluttered. Tapping the shortcut shows and hides an always-on-top overlay on your active screen regardless of whether you have other apps full-screen or not. Beyond simple navigation, you can also use voice commands to make post-meeting corrections or additions, for instance, you can simply say &quot;merge this speaker with that speaker&quot; to clean up the transcript.<p>It also has push-to-talk&#x2F;dictate functionality with LLM cleanup - what the app started as but that tool was developer catnip, soo many of them.<p>A developer friend who’s worked in finance reviewed the site and said he’d bounce because the privacy story wasn’t strong enough so I added a completely on-device mode and a bring-your-own-key option. Using cloud models does add a lot to the experience, including context aware speaker merging and fragment cleanup, summary items during meetings, action items attributed, etc. On-device mode is completely free and the speaker identification is still very useful.<p>The privacy story is my biggest worry with the app, particularly since its target audience is more technical people. I’d love to get people&#x27;s thoughts on it and any feedback would be super helpful.

Gov.uk has replaced Stripe with Dutch provider Adyen <a href="https:&#x2F;&#x2F;gds.blog.gov.uk&#x2F;2026&#x2F;06&#x2F;02&#x2F;building-for-the-future-making-change-simple-on-gov-uk-pay&#x2F;" rel="nofollow">https:&#x2F;&#x2F;gds.blog.gov.uk&#x2F;2026&#x2F;06&#x2F;02&#x2F;building-for-the-future-m...</a><p><a href="https:&#x2F;&#x2F;www.adyen.com&#x2F;press-and-media&#x2F;adyen-payments-gov-uk" rel="nofollow">https:&#x2F;&#x2F;www.adyen.com&#x2F;press-and-media&#x2F;adyen-payments-gov-uk</a>

Gemma 4 QAT models: Optimizing compression for mobile and laptop efficiency

[Other] pg_durable: Microsoft open sources in-database durable execution

Looking Forward to Postgres 19: Query Hints

The Quiet Numbers Station: Decoding Nineteen Years of GPS Cryptography <a href="https:&#x2F;&#x2F;lsc-pagepro.mydigitalpublication.com&#x2F;publication&#x2F;?i=865273&amp;p=62&amp;view=issueViewer" rel="nofollow">https:&#x2F;&#x2F;lsc-pagepro.mydigitalpublication.com&#x2F;publication&#x2F;?i=...</a><p>PDF: <a href="https:&#x2F;&#x2F;cdn.coverstand.com&#x2F;61061&#x2F;865273&#x2F;2c88ea662e2b574787232d6b66f1ae8c3cfec2c5.1.pdf" rel="nofollow">https:&#x2F;&#x2F;cdn.coverstand.com&#x2F;61061&#x2F;865273&#x2F;2c88ea662e2b57478723...</a> (article is on page 62)<p>Related: <a href="https:&#x2F;&#x2F;www.404media.co&#x2F;the-u-s-military-quietly-turned-gps-into-a-global-numbers-station-evidence-suggests&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.404media.co&#x2F;the-u-s-military-quietly-turned-gps-...</a>

I Thought I Knew How Electrolysis Worked [video]

[CLI Tool] Show HN: Lowfat – pluggable CLI filter that saved 91.8% of my LLM tokens Hi HN,<p>Not sure if anyone would be interested.<p>But, just wanted to share that I&#x27;ve been maintaining my small tool called &#x27;lowfat&#x27; that helps me filters some of my verbose CLI output.<p>It&#x27;s a single binary, works as an agent hook or a shell wrapper. It has a plugin system to customize filters per command.<p>The idea is pretty simple: agents don&#x27;t need the full kubectl get -o yaml or any 10k-line dump to make decisions. So that lowfat sits in between, strips the noise, and passes through what matters.<p>Here&#x27;s my real report after 2 months of personal use:<p>lowfat history --all<p><pre><code> lowfat plugin candidates ───────────────────────────────────────────────────────── # command runs avg raw cost savings source status 1 kubectl get 101x 14.4K 1.5M 93.9% plugin good 2 grep 103x 13.5K 1.4M 96.2% plugin good 3 git diff 81x 995 80.6K 57.9% built-in good 4 kubectl 90x 485 43.6K 33.6% plugin good 5 docker 127x 5.5K 693.6K 96.1% built-in good 6 ls 489x 117 57.3K 56.2% built-in good 7 find 30x 16.5K 495.0K 95.5% plugin good 8 git show 63x 490 30.9K 38.0% built-in good 9 git 177x 368 65.2K 76.1% built-in good 10 git log 86x 556 47.8K 78.5% built-in good 11 kubectl logs 5x 3.6K 17.8K 43.0% plugin good 12 git status 86x 152 13.1K 58.0% built-in good 13 docker ps 20x 467 9.3K 52.8% plugin good 14 kubectl describe 6x 656 3.9K 1.2% plugin weak 15 docker images 9x 940 8.5K 61.8% built-in good 16 k get 2x 2.1K 4.2K 35.9% plugin good 17 terraform 10x 395 3.9K 32.1% plugin good 18 git commit 32x 77 2.5K 0.0% built-in weak 19 docker build 8x 487 3.9K 37.6% built-in good 20 docker compose 22x 979 21.5K 89.4% built-in good total: 4.4M raw → 4.1M saved (91.8%) </code></pre> My toolset above is kind limited, but it works pretty well for my usecase without any interruption Kinda help me not reaching the token limit for my company Bedrock limit usage and keep optimizing the saving on the go for later usage.<p>But, why not alternatives (<a href="https:&#x2F;&#x2F;github.com&#x2F;zdk&#x2F;lowfat#alternatives" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;zdk&#x2F;lowfat#alternatives</a>) ? The answers are: - My goal is to make the core lightweight but extensible via plugins i.e. not trying to bundle every command in the installed binary so that people own their output filters. - Customizable per usecase via plugin or filter pipelines as I am using my own toolset. - Customizable for non-public CLI tools, for example, some enterprise might have their interal CLI tools that public won&#x27;t have access. - People should own their data. So the design is local-first, No telemetry forever. - I kinda love UNIX-style composible pipes, so lowfat-filter has implemented this style. - Be able to adjust aggressiveness of the filter, so we can control that we won&#x27;t strip something the agent needed.<p>GitHub: <a href="https:&#x2F;&#x2F;github.com&#x2F;zdk&#x2F;lowfat" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;zdk&#x2F;lowfat</a><p>Anyway, if anyone is interested, feedbacks and questions are welcome!<p>Thanks!

Tracing a powerful GNSS interference source over Europe

[Other] ESP32 Bit Pirate, a Hardware Hacking Tool with WebCLI That Speaks Every Protocol

There's no escaping it: an exploration of ANSI codes

Show HN: I Derived a Pancake After 25 years of making other people&#x27;s pancake recipes - always yearning for more tang, more fluff, and more predictability - I decided to derive the pancake recipe from the chemistry.<p>You mark checkboxes for what you have on hand (ricotta, sour cream, kefir, buttermilk, yogurt, cottage cheese, lemon, cream of tartar, etc.) and it computes the best recipe based on targets for acid, fat, salt, sugar, and CO2.<p>My particular favorite are the yeast-raised lemon ricotta kefir pancakes - the best I&#x27;ve ever had.<p>The math is done in a small pure-ESM library: ingredient composition to component masses and acid moles, a stoichiometry layer, and a bisection solver for the target deficits.<p>I&#x27;m not a chemist, so if something is off, tell me and I will fix it!

Unicode Fonts and Tools for X11

[Other] Meta enables ADB on deprecated Portal devices [video] <a href="https:&#x2F;&#x2F;developers.meta.com&#x2F;horizon&#x2F;blog&#x2F;build-apps-for-portal-with-ai&#x2F;" rel="nofollow">https:&#x2F;&#x2F;developers.meta.com&#x2F;horizon&#x2F;blog&#x2F;build-apps-for-port...</a>

Yon – a topos-oriented language with a content-addressed lattice heap Hello everyone. In the last two years I spent, as a dev, part of my free time stretching the limits of my knowledge. Not being a mathematician myself, I discovered that formalizing concepts in mathematical language could nonetheless be useful to improve symbolic reasoning about the concepts themselves. I made use of both books and AI, and I followed the development of the latter, mainly with a critical eye. I have several open projects, and from some observations and explorations on one of them I started asking myself what the current limits of reasoning, of logic, of mathematics itself are. So I explored categories, and topoi, above all starting from Mazzola&#x27;s theory of music. I asked myself whether this could influence type theory in programming, and I ran some experiments. Out of this came this programming language, Yon, inspired by Yoneda and by morphisms. From another project I drew observations on the Leech lattice; from yet another, some experiments with mmap and coordinate-based allocation in a structure that would be advantageous, again, in a topological sense. The language certainly has mistakes here and there and I wrote the documentation in a hurry; the work took 3 weeks in total. It compiles to LLVM for performance reasons, and for now I preferred to avoid a VM and a GC. It contains unusual data structures that turn out to be performant. It&#x27;s worth a look, and I hope it will win some converts, and that someone will want to help me with its development. I&#x27;d love for it to bring fresh stimuli to programming and maybe open a few new frontiers. A few concrete details, for those who want to look under the hood. The compiler is a real pipeline, not an interpreter: an OCaml frontend takes .yon source into a custom MLIR dialect I called &quot;topos&quot;, where the categorical constructs live as first-class operations; its lowering passes take everything down to LLVM IR and from there to a native executable. A single command, yonc, drives the whole chain, and you can stop at any intermediate stage to see what a categorical construct actually becomes on its way to silicon. The runtime is where the Leech lattice observations ended up. The heap is content-addressed over Λ₂₄: every value is mapped to a lattice point and canonicalized under the Conway group Co₀ (via libmmgroup), so the same content always lives at the same address. That buys three things I would now find hard to give up: equality is a single machine comparison no matter how big the value is (string equality benches flat at ~17 ns up to 32,768-character strings, because it compares handles, never bytes); deduplication is global and automatic, with no interning logic in user code; and giving up the GC stopped being a renunciation, since cells are immutable and content-addressed, so there is nothing to trace and nothing to move. Concurrency I kept deliberately simple-minded: no threads, no shared mutable state. A program splits into isolated &quot;Spaces&quot; (separate processes, isolation enforced by the MMU) that talk over shared-memory channels with explicit failure semantics. About what is verified and what is just hope: the ground truth is a regression suite of 112 examples plus a cross-Space scenario suite, with exit codes identical on Linux x86-64 and macOS Apple Silicon (Intel Macs: untested). The book on the site, 21 chapters plus appendices, had every snippet compiled and run before being written down. The benchmarks appendix declares its environment and method; I tried not to publish any number without one. The limits of 1.0 are written down as well, in a baseline document that lists every fixed pool (256 heaps per chain, 64 Spaces, 16 concurrent RPC sessions, and so on), with the rationale that a hard limit that fails loudly is a specification, while a soft limit that degrades silently is a bug. For the license I went with the GCC model: compiler and toolchain are AGPLv3, the runtime is AGPLv3 with an explicit linking exception, so the language itself stays free, and the programs you write in it are entirely yours, under any license you choose.<p>Site + book: <a href="https:&#x2F;&#x2F;yon-lang.org" rel="nofollow">https:&#x2F;&#x2F;yon-lang.org</a> Repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;yon-language&#x2F;yon" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;yon-language&#x2F;yon</a> (tag v1.0.0)<p>Happy to answer anything: the topos dialect, why a lattice rather than a hash, what the categorical constructs lower to, what broke along the way.

[CLI Tool] Open Code Review – An AI-powered code review CLI tool

Man-Computer Symbiosis J. C. R. Licklider (1960)