The best Hacker News stories from Show from the past week

Latest posts:

Show HN: Walrus – a Kafka alternative written in Rust

Show HN: Onlyrecipe 2.0 – I added all features HN requested – 4 years later

Show HN: Onlyrecipe 2.0 – I added all features HN requested – 4 years later

Show HN: Fresh – A new terminal editor built in Rust

I built Fresh to challenge the status quo that terminal editing must require a steep learning curve or endless configuration. My goal was to create a fast, resource-efficient TUI editor with the usability and features of a modern GUI editor (like a command palette, mouse support, and LSP integration).<p>Core Philosophy:<p>- <i>Ease-of-Use:</i> Fundamentally non-modal. Prioritizes standard keybindings and a minimal learning curve.<p>- <i>Efficiency:</i> Uses a lazy-loading piece tree to avoid loading huge files into RAM - reads only what's needed for user interactions. Coded in Rust.<p>- <i>Extensibility:</i> Uses TypeScript (via Deno) for plugins, making it accessible to a large developer base.<p>The Performance Challenge:<p>I focused on resource consumption and speed with large file support as a core feature. I did a quick benchmark loading a 2GB log file with ANSI color codes. Here is the comparison against other popular editors:<p><pre><code> - Fresh: Load Time: ~600ms | Memory: ~36 MB - Neovim: Load Time: ~6.5 seconds | Memory: ~2 GB - Emacs: Load Time: ~10 seconds | Memory: ~2 GB - VS Code: Load Time: ~20 seconds | Memory: OOM Killed (~4.3 GB available) </code></pre> (Only Fresh rendered the ansi colors.)<p>Development process:<p>I embraced Claude Code and made an effort to get good mileage out of it. I gave it strong specific directions, especially in architecture / code structure / UX-sensitive areas. It required constant supervision and re-alignment, especially in the performance critical areas. Added very extensive tests (compared to my normal standards) to keep it aligned as the code grows. Especially, focused on end-to-end testing where I could easily enforce a specific behavior or user flow.<p>Fresh is an open-source project (GPL-2) seeking early adopters. You're welcome to send feedback, feature requests, and bug reports.<p>Website: <a href="https://sinelaw.github.io/fresh/" rel="nofollow">https://sinelaw.github.io/fresh/</a><p>GitHub Repository: <a href="https://github.com/sinelaw/fresh" rel="nofollow">https://github.com/sinelaw/fresh</a>

Show HN: I built a dashboard to compare mortgage rates across 120 credit unions

When I bought my home, the big bank I'd been using for years quoted me 7% APR. A local credit union was offering 5.5% for the exact same mortgage.<p>I was surprised until I learned that mortgages are basically standardized products – the government buys almost all of them (see Bits About Money: <a href="https://www.bitsaboutmoney.com/archive/mortgages-are-a-manufactured-product/" rel="nofollow">https://www.bitsaboutmoney.com/archive/mortgages-are-a-manuf...</a>). So what's the price difference paying for? A recent Bloomberg Odd Lots episode makes the case that it's largely advertising and marketing (<a href="https://www.bloomberg.com/news/audio/2025-11-28/odd-lots-this-is-why-credit-card-rates-are-so-high-podcast" rel="nofollow">https://www.bloomberg.com/news/audio/2025-11-28/odd-lots-thi...</a>). Credit unions are non-profits without big marketing budgets, so they can pass those savings on, but a lot of people don't know about them.<p>I built this dashboard to make it easier to shop around. I pull public rates from 120+ credit union websites and compares against the weekly FRED national benchmark.<p>Features:<p>- Filter by loan type (30Y/15Y/etc.), eligibility (the hardest part tbh), and rate type - Payment calculator with refi mode (CUs can be a bit slower than big lenders, but that makes them great for refi) - Links to each CU's rates page and eligibility requirements - Toggle to show/hide statistical outliers<p>At the time of writing, the average CU rate is 5.91% vs. 6.23% national average. about $37k difference in total interest on a $500k loan. I actually used seaborn to visualize the rate spread against the four big banks: <a href="https://www.reddit.com/r/dataisbeautiful/comments/1pcj9t7/oc_the_high_cost_of_big_banks_i_tracked_daily/" rel="nofollow">https://www.reddit.com/r/dataisbeautiful/comments/1pcj9t7/oc...</a><p>Stack: Python for the data/backend, Svelte/SvelteKit for the frontend. No signup, no ads, no referral fees.<p>Happy to answer questions about the methodology or add CUs people suggest.

Show HN: Fixing Google Nano Banana Pixel Art with Rust

Show HN: Real-time system that tracks how news spreads across 200k websites

I built a system that monitors ~200,000 news RSS feeds in near real-time and clusters related articles to show how stories spread across the web.<p>It uses Snowflake’s Arctic model for embeddings and HNSW for fast similarity search. Each “story cluster” shows who published first, how fast it propagated, and how the narrative evolved as more outlets picked it up.<p>Would love feedback on the architecture, scaling approach, and any ways to make the clusters more accurate or useful.<p>Live demo: <a href="https://yandori.io/news-flow/" rel="nofollow">https://yandori.io/news-flow/</a>

Show HN: Boing

Show HN: Boing

Show HN: Nano PDF – A CLI Tool to Edit PDFs with Gemini's Nano Banana

The new Gemini 3 Pro Image model (aka Nano Banana) is incredible at generating slides, so I thought it would be fun to build a CLI tool that lets you edit PDF presentations using plain English. The tool converts the page you want to edit into an image, sends it to the model API together with your prompt to generate an edited image, then converts the updated image back and stitches into the original document.<p>Examples:<p>- `nano-pdf edit deck.pdf 5 "Update the revenue chart to show Q3 at $2.5M"`<p>- `nano-pdf add deck.pdf 15 "Create an executive summary slide with 5 bullet points"`<p>Features:<p>- Edit multiple pages in parallel<p>- Add entirely new slides that match your deck's style<p>- Google Search enabled by default so the model can look up current data<p>- Preserves text layer for copy/paste and search<p>It can work with any kind of PDF but I expect it would be most useful for a quick edit to a deck or something similar.<p>GitHub: <a href="https://github.com/gavrielc/Nano-PDF" rel="nofollow">https://github.com/gavrielc/Nano-PDF</a>

Show HN: Explore what the browser exposes about you

I built a tool that reveals the data your browser exposes automatically every time you visit a website.<p>GitHub: <a href="https://github.com/neberej/exposedbydefault" rel="nofollow">https://github.com/neberej/exposedbydefault</a><p>Demo: <a href="https://neberej.github.io/exposedbydefault/" rel="nofollow">https://neberej.github.io/exposedbydefault/</a><p>Note: No data is sent anywhere. Everything runs in your browser.

Show HN: Explore what the browser exposes about you

I built a tool that reveals the data your browser exposes automatically every time you visit a website.<p>GitHub: <a href="https://github.com/neberej/exposedbydefault" rel="nofollow">https://github.com/neberej/exposedbydefault</a><p>Demo: <a href="https://neberej.github.io/exposedbydefault/" rel="nofollow">https://neberej.github.io/exposedbydefault/</a><p>Note: No data is sent anywhere. Everything runs in your browser.

Anthony Bourdain's Lost Li.st's

I read through the years about Bourdain's content on the defunct li.st service, but was never able to find an archive of it. A more thorough perusing of archive.org and a pointer from an Internet stranger led me to create this site. Cheers

Show HN: Glasses to detect smart-glasses that have cameras

Hi! Recently smart-glasses with cameras like the Meta Ray-bans seem to be getting more popular. As does some people's desire to remove/cover up the recording indicator LED. I wanted to see if there's a way to detect when people are recording with these types of glasses, so a little bit ago I started working this project. I've hit a little bit of a wall though so I'm very much open to ideas!<p>I've written a bunch more on the link (+photos are there), but essentially this uses 2 fingerprinting approaches: - retro-reflectivity of the camera sensor by looking at IR reflections. mixed results here. - wireless traffic (primarily BLE, also looking into BTC and wifi)<p>For the latter, I'm currently just using an ESP32, and I can consistently detect when the Meta Raybans are 1) pairing, 2) first powered on, 3) (less consistently) when they're taken out of the charging case. When they do detect something, it plays a little jingle next to your ear.<p>Ideally I want to be able to detect them when they're in use, and not just at boot. I've come across the nRF52840, which seems like it can follow directed BLE traffic beyond the initial broadcast, but from my understanding it would still need to catch the first CONNECT_REQ event regardless. On the bluetooth classic side of things, all the hardware looks really expensive! Any ideas are appreciated. Thanks!

Show HN: KiDoom – Running DOOM on PCB Traces

I got DOOM running in KiCad by rendering it with PCB traces and footprints instead of pixels.<p>Walls are rendered as PCB_TRACK traces, and entities (enemies, items, player) are actual component footprints - SOT-23 for small items, SOIC-8 for decorations, QFP-64 for enemies and the player.<p>How I did it:<p>Started by patching DOOM's source code to extract vector data directly from the engine. Instead of trying to render 64,000 pixels (which would be impossibly slow), I grab the geometry DOOM already calculates internally - the drawsegs[] array for walls and vissprites[] for entities.<p>Added a field to the vissprite_t structure to capture entity types (MT_SHOTGUY, MT_PLAYER, etc.) during R_ProjectSprite(). This lets me map 150+ entity types to appropriate footprint categories.<p>The DOOM engine sends this vector data over a Unix socket to a Python plugin running in KiCad. The plugin pre-allocates pools of traces and footprints at startup, then just updates their positions each frame instead of creating/destroying objects. Calls pcbnew.Refresh() to update the display.<p>Runs at 10-25 FPS depending on hardware. The bottleneck is KiCad's refresh, not DOOM or the data transfer.<p>Also renders to an SDL window (for actual gameplay) and a Python wireframe window (for debugging), so you get three views running simultaneously.<p>Follow-up: ScopeDoom<p>After getting the wireframe renderer working, I wanted to push it somewhere more physical. Oscilloscopes in X-Y mode are vector displays - feed X coordinates to one channel, Y to the other. I didn't have a function generator, so I used my MacBook's headphone jack instead.<p>The sound card is just a dual-channel DAC at 44.1kHz. Wired 3.5mm jack → 1kΩ resistors → scope CH1 (X) and CH2 (Y). Reused the same vector extraction from KiDoom, but the Python script converts coordinates to ±1V range and streams them as audio samples.<p>Each wall becomes a wireframe box, the scope traces along each line. With ~7,000 points per frame at 44.1kHz, refresh rate is about 6 Hz - slow enough to be a slideshow, but level geometry is clearly recognizable. A 96kHz audio interface or analog scope would improve it significantly (digital scopes do sample-and-hold instead of continuous beam tracing).<p>Links:<p>KiDoom GitHub: <a href="https://github.com/MichaelAyles/KiDoom" rel="nofollow">https://github.com/MichaelAyles/KiDoom</a>, writeup: <a href="https://www.mikeayles.com/#kidoom" rel="nofollow">https://www.mikeayles.com/#kidoom</a><p>ScopeDoom GitHub: <a href="https://github.com/MichaelAyles/ScopeDoom" rel="nofollow">https://github.com/MichaelAyles/ScopeDoom</a>, writeup: <a href="https://www.mikeayles.com/#scopedoom" rel="nofollow">https://www.mikeayles.com/#scopedoom</a>

Show HN: We built an open source, zero webhooks payment processor

Hi HN! For the past bit we’ve been building Flowglad (<a href="https://flowglad.com">https://flowglad.com</a>) and can now feel it’s just gotten good enough to share with you all:<p>Repo: <a href="https://github.com/flowglad/flowglad" rel="nofollow">https://github.com/flowglad/flowglad</a><p>Demo video: <a href="https://www.youtube.com/watch?v=G6H0c1Cd2kU" rel="nofollow">https://www.youtube.com/watch?v=G6H0c1Cd2kU</a><p>Flowglad is a payment processor that you integrate without writing any glue code. Along with processing your payments, it tells you in real time the features and usage credit balances that your customers have available to you based on their billing state. The DX feels like React, because we wanted to bring the reactive programming paradigm to payments.<p>We make it easy to spin up full-fledged pricing models (including usage meters, feature gates and usage credit grants) in a few clicks. We schematize these pricing models into a pricing.yaml file that’s kinda like Terraform but for your pricing.<p>The result is a payments layer that AI coding agents have a substantially easier time one-shotting (for now the happiest path is a fullstack Typescript + React app).<p>Why we built this:<p>- After a decade of building on Stripe, we found it powerful but underopinionated. It left us doing a lot of rote work to set up fairly standard use cases - That meant more code to maintain, much of which is brittle because it crosses so many server-client boundaries - Not to mention choreographing the lifecycle of our business domain with the Stripe checkout flow and webhook event types, of which there are 250+ - Payments online has gotten complex - not just new pricing models for AI products, but also cross border sales tax, etc. You either need to handle significant chunks of it yourself, or sign up for and compose multiple services<p>This all feels unduly clunky, esp when compared to how easy other layers like hosting and databases have gotten in recent years.<p>These patterns haven’t changed much in a decade. And while coding agents can nail every other rote part of an app (auth, db, analytics), payments is the scariest to tab-tab-tab your way through. Because the the existing integration patterns are difficult to reason about, difficult to verify correctness, and absolutely mission critical.<p>Our beta version lets you:<p>- Spin up common pricing models in just a few clicks, and customize them as needed - Clone pricing models between testmode and live mode, and import / export via pricing.yaml - Check customer usage credits and feature access in real time on your backend and React frontend - Integrate without any DB schema changes - you reference your customers via your ids, and reference prices, products, features and usage meters via slugs that you define<p>We’re still early in our journey so would love your feedback and opinions. Billing has a lot of use cases, so if you see anything that you wish we supported, please let us know!

Show HN: Stun LLMs with thousands of invisible Unicode characters

I made a free tool that stuns LLMs with invisible Unicode characters.<p>Use cases: Anti-plagiarism, text obfuscation against LLM scrapers, or just for fun!<p>Even just one word's worth of “gibberified” text is enough to block most LLMs from responding coherently.

Show HN: I built an interactive HN Simulator

Hey HN! Just for fun, I built an interactive Hacker News Simulator.<p>You can submit text posts and links, just like the real HN. But on HN Simulator, all of the comments are generated by LLMs + generate instantly.<p>The best way to use it (IMHO) is to submit a text post or a curl-able URL here: <a href="https://news.ysimulator.run/submit" rel="nofollow">https://news.ysimulator.run/submit</a>. You don't need an account to post.<p>When you do that, various prompts will be built from a library of commenter archetypes, moods, and shapes. The AI commenters will actually respond to your text post and/or submitted link.<p>I really wanted it to feel real, and I think the project mostly delivers on that. When I was developing it, I kept getting confused between which tab was the "real" HN and which was the simulator, and accidentally submitted some junk to HN. (Sorry dang and team – I did clean up after myself).<p>The app itself is built with Node + Express + Postgres, and all of the inference runs on Replicate.<p>Speaking of Replicate, they generously loaded me up with some free credits for the inference – so shoutout to the team there.<p>The most technically interesting part of the app is how the comments work. You can read more about it here, as well as explore all of the available archetypes, moods, and shapes that get combined into prompts: <a href="https://news.ysimulator.run/comments.html" rel="nofollow">https://news.ysimulator.run/comments.html</a><p>I hope you all have as much fun playing with it as I did making it!

Ask HN: How are Markov chains so different from tiny LLMs?

I polished a Markov chain generator and trained it on an article by Uri Alon and al (<a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC7963340/" rel="nofollow">https://pmc.ncbi.nlm.nih.gov/articles/PMC7963340/</a>).<p>It generates text that seems to me at least on par with tiny LLMs, such as demonstrated by NanoGPT. Here is an example:<p><pre><code> jplr@mypass:~/Documenti/2025/SimpleModels/v3_very_good$ ./SLM10b_train UriAlon.txt 3 Training model with order 3... Skip-gram detection: DISABLED (order < 5) Pruning is disabled Calculating model size for JSON export... Will export 29832 model entries Exporting vocabulary (1727 entries)... Vocabulary export complete. Exporting model entries... Processed 12000 contexts, written 28765 entries (96.4%)... JSON export complete: 29832 entries written to model.json Model trained and saved to model.json Vocabulary size: 1727 jplr@mypass:~/Documenti/2025/SimpleModels/v3_very_good$ ./SLM9_gen model.json </code></pre> <i>Aging cell model requires comprehensive incidence data. To obtain such a large medical database of the joints are risk factors. Therefore, the theory might be extended to describe the evolution of atherosclerosis and metabolic syndrome. For example, late‐stage type 2 diabetes is associated with collapse of beta‐cell function. This collapse has two parameters: the fraction of the senescent cells are predicted to affect disease threshold . For each individual, one simulates senescent‐cell abundance using the SR model has an approximately exponential incidence curve with a decline at old ages In this section, we simulated a wide range of age‐related incidence curves. The next sections provide examples of classes of diseases, which show improvement upon senolytic treatment tends to qualitatively support such a prediction. model different disease thresholds as values of the disease occurs when a physiological parameter ϕ increases due to the disease. Increasing susceptibility parameter s, which varies about 3‐fold between BMI below 25 (male) and 54 (female) are at least mildly age‐related and 25 (male) and 28 (female) are strongly age‐related, as defined above. Of these, we find that 66 are well described by the model as a wide range of feedback mechanisms that can provide homeostasis to a half‐life of days in young mice, but their removal rate slows down in old mice to a given type of cancer have strong risk factors should increase the removal rates of the joint that bears the most common biological process of aging that governs the onset of pathology in the records of at least 104 people, totaling 877 disease category codes (See SI section 9), increasing the range of 6–8% per year. The two‐parameter model describes well the strongly age‐related ICD9 codes: 90% of the codes show R 2 > 0.9) (Figure 4c). This agreement is similar to that of the previously proposed IMII model for cancer, major fibrotic diseases, and hundreds of other age‐related disease states obtained from 10−4 to lower cancer incidence. A better fit is achieved when allowing to exceed its threshold mechanism for classes of disease, providing putative etiologies for diseases with unknown origin, such as bone marrow and skin. Thus, the sudden collapse of the alveoli at the outer parts of the immune removal capacity of cancer. For example, NK cells remove senescent cells also to other forms of age‐related damage and decline contribute (De Bourcy et al., 2017). There may be described as a first‐passage‐time problem, asking when mutated, impair particle removal by the bronchi and increase damage to alveolar cells (Yang et al., 2019; Xu et al., 2018), and immune therapy that causes T cells to target senescent cells (Amor et al., 2020). Since these treatments are predicted to have an exponential incidence curve that slows at very old ages. Interestingly, the main effects are opposite to the case of cancer growth rate to removal rate We next consider the case of frontline tissues discussed above.</i>

Show HN: Forty.News – Daily news, but on a 40-year delay

This started as a reaction to a conversational trope. Despite being a tranquil place, even conversations at my yoga studio often start with, "Can you believe what's going on right now?" with that angry/scared undertone.<p>I'm a news avoider, so I usually feel some smug self-satisfaction in those instances, but I wondered if there was a way to satisfy the urge to doomscroll without the anxiety.<p>My hypothesis: Apply a 40-year latency buffer. You get the intellectual stimulation of "Big Events" without the fog of war, because you know the world didn't end.<p>40 years creates a mirror between the Reagan Era and today. The parallels include celebrity populism, Cold War tensions (Soviets vs. Russia), and inflation economics.<p>The system ingests raw newspaper scans and uses a multi-step LLM pipeline to generate the daily edition:<p>OCR & Ingestion: Converts raw pixels to text.<p>Scoring: Grades events on metrics like Dramatic Irony and Name Recognition to surface stories that are interesting with hindsight. For example, a dry business blurb about Steve Jobs leaving Apple scores highly because the future context creates a narrative arc.<p>Objective Fact Extraction: Extracts a list of discrete, verifiable facts from the raw text.<p>Generation: Uses those extracted facts as the ground truth to write new headlines and story summaries.<p>I expected a zen experience. Instead, I got an entertaining docudrama. Historical events are surprisingly compelling when serialized over weeks.<p>For example, on Oct 7, 1985, Palestinian hijackers took over the cruise ship Achille Lauro. Reading this on a delay in 2025, the story unfolded over weeks: first they threw an American in a wheelchair overboard, then US fighter jets forced the escape plane to land, leading to a military standoff between US Navy SEALs and the Italian Air Force. Unbelievably, the US backed down, but the later diplomatic fallout led the Italian Prime Minister to resign.<p>It hits the dopamine receptors of the news cycle, but with the comfort of a known outcome.<p>Stack: React, Node.js (Caskada for the LLM pipeline orchestration), Gemini for OCR/Scoring.<p>Link: <a href="https://forty.news" rel="nofollow">https://forty.news</a> (No signup required, it's only if you want the stories emailed to you daily/weekly)