Show HN: Memory Lane – bootstrap your naive Claude instances with their history https://ift.tt/ABe9FIx

Show HN: Memory Lane – bootstrap your naive Claude instances with their history https://ift.tt/a6SCVe2 December 1, 2025 at 04:04AM

Show HN: I Built Tinyfocus – A Minimal Tool to Help Solo Founders Focus https://ift.tt/8rciGgZ

Show HN: I Built Tinyfocus – A Minimal Tool to Help Solo Founders Focus Hi HN, I just launched Tinyfocus, a small productivity tool designed specifically for solo founders and builders. The goal is simple: help you focus on what matters and get more done in less time. Here’s what Tinyfocus does: Lets you track your top tasks and prioritize efficiently. Provides micro dashboards to keep your daily focus in check. Lightweight, no distractions, no fluff. I built it entirely by myself, iterating in public, and I wanted to share it with the community to get feedback. It’s been crazy seeing how a simple tool can make such a difference in daily focus, especially when you’re juggling multiple projects as a solo founder. Check it out here: tinyfoc.us I’d love to hear your thoughts – any feedback, feature ideas, or bugs you notice. Thanks! https://ift.tt/1YEbjq7 December 1, 2025 at 01:05AM

Show HN: Unmarker.it – Client-Side Tool to Disrupt Invisible AI Watermarks https://ift.tt/1FhYvMe

Show HN: Unmarker.it – Client-Side Tool to Disrupt Invisible AI Watermarks I built a browser-only tool that disrupts invisible AI watermarks using Canvas, geometry, noise, and JPEG recompression. No backend, no uploads, no tracking. Pipeline: Shake – Random rotation (±0.5°) + slight zoom Stir – Low-amplitude RGB noise via getImageData Crush – JPEG recompression at ~0.85 quality Tested with SynthID (Google Gemini AI watermarking), and it remained undetected in all tests. Pipeline improvements? What would you add/change? Github: https://ift.tt/OViq0Gr https://ift.tt/jSCwloR December 1, 2025 at 01:37AM

Show HN: No Environment Setups Anymore https://ift.tt/VZQpUg0

Show HN: No Environment Setups Anymore Hi everyone, for last 7 months, I have been learning all the attempts made to eliminate codebase environment setups. Here's my product which is a leap in the same direction and will help you run any codebase on relevant machine. Check it out on gitarsenal.dev/ and we got ranked 6th on Product Hunt as well. https://ift.tt/VcwAU3n November 30, 2025 at 02:57AM

Show HN: Zero-power photonic language model–code https://ift.tt/ZJGEco3

Show HN: Zero-power photonic language model–code The model uses a 1024-dimensional complex Hilbert space with 32 layers of programmable Mach–Zehnder meshes (Reck architecture) and derives token probabilities directly via the Born rule. Despite using only unitary operations and no attention mechanism, a 1024×32 model achieves coherent TinyStories generation after < 1.8 hours of training on a single consumer GPU. This is Part 1 - the next step is physical implementation with $50 of optics from AliExpress. https://zenodo.org/records/17764289 November 30, 2025 at 01:45AM

Show HN: Lifetime Black Friday Deals (Mega List) https://ift.tt/v4xjuhf

Show HN: Lifetime Black Friday Deals (Mega List) https://ift.tt/wDbmzaG November 29, 2025 at 11:23PM

Show HN: Auth Agent – the first agent-native auth flow for websites. Check out https://ift.tt/NQm8x26

Show HN: Auth Agent – the first agent-native auth flow for websites. Check out https://ift.tt/gywcMUB November 29, 2025 at 11:27PM

Show HN: Oblit – A zero-dependency, binary-first protocol for Node.js (Show HN https://ift.tt/xMmJH7g

Show HN: Oblit – A zero-dependency, binary-first protocol for Node.js (Show HN November 29, 2025 at 12:17AM

Show HN: Research Papers as Memes https://ift.tt/EkFsqRf

Show HN: Research Papers as Memes https://near.tl/tech November 28, 2025 at 11:19PM

Show HN: I built a free astro and tailwind static site for GitHub pages https://ift.tt/YgLHFEG

Show HN: I built a free astro and tailwind static site for GitHub pages Using my GitHub pro+ with vs code setup This is a demonstration of how good of a site can I build essentially 100% for free + free hosting (if coded manually without a 50$ subscription) And I went completely overboard on purpose its 99% useless for a real production deployment im sure but for mini blogs probably might be useful idk I dont even use the new GitHub spark or whatever to slow compared to 1k+ line edits every couple minutes im obviously working on a ton of other things I won't make public yet but will in the future https://tariqdude.github.io/Github-Pages-Project-v1/ November 28, 2025 at 05:17AM

Show HN: Whole-home VPN router with hardware kill switch (OpenWrt and WireGuard) https://ift.tt/jApuSng

Show HN: Whole-home VPN router with hardware kill switch (OpenWrt and WireGuard) With internet censorship and surveillance on the rise, ie; UK Online Safety Bill (July 2025) and Australia's social media legislation (Dec 2025) introducing mandatory age verification (read: initial step on the pathway to social credit), I wanted a privacy-first solution that protects browsing history from ISPs and third-party verification services, but not one that requires you to be an Einstein to deploy. This stack turns a Raspberry Pi (or any OpenWrt-compatible device) into a network-wide VPN gateway. Key features: - Hardware kill switch: VPN down = no internet (not a software rule that can leak) - AmneziaWG obfuscation for DPI-resistant connections - Optional AdGuard Home for DNS filtering - Works for all devices including smart TVs and IoT that can't run VPN apps Not a techie? The README is optimized for AI-assisted deployment. Feed it to your LLM of choice (Claude, GPT, etc.) and it can walk you through the entire setup for your specific hardware. Mullvad-focused but works with any WireGuard provider. MIT license. Docker deploy in testing (coming soon) https://ift.tt/9YHcW5C November 28, 2025 at 05:50AM

Show HN: No Black Friday – A directory of fair-price brands https://ift.tt/hLpMBms

Show HN: No Black Friday – A directory of fair-price brands The idea came from noticing how many brands inflate prices only to discount them later. Some companies refuse to do that, and I wanted a place to highlight them. If you know a company that doesn’t participate in Black Friday or similar discount events, please add it or share it here. I’d love to grow the list with help from the community. Manuel https://ift.tt/8fZR7xV November 28, 2025 at 04:20AM

Show HN: ZigFormer – An LLM implemented in pure Zig https://ift.tt/R8PSJV9

Show HN: ZigFormer – An LLM implemented in pure Zig Hi everyone, I've made an early version of ZigFormer, a small LLM implemented in Zig with no dependencies on external ML frameworks like PyTorch or JAX. ZigFormer is modelled after a textbook LLM (like GPT-2 from OpenAI) and can be used as a Zig library as well as a standalone application to train a model and chat with it. This was mainly an educational project. I'm sharing it here in case others find it interesting or useful. Link to the project: https://ift.tt/HgcwM14 November 27, 2025 at 11:19PM

Show HN: Fixing Google Nano Banana Pixel Art with Rust https://ift.tt/Xd8PUKs

Show HN: Fixing Google Nano Banana Pixel Art with Rust https://ift.tt/XaM1nvW November 26, 2025 at 10:46PM

Show HN: A New Mnemonic Scheme – Seedless, Passphrase-Sealed, Builtin Multichain https://ift.tt/SZt6OJG

Show HN: A New Mnemonic Scheme – Seedless, Passphrase-Sealed, Builtin Multichain TL;DR: Today’s mnemonics are bare keys with no cryptographic protection. MSCIKDF adds passphrase-sealing, seedless operation, and multi-curve support in a single derivation scheme. The user's exposure to compromise/leakage risk can be periodically reset, marking a significant advancement in cryptographic security. In the first half of 2025, more than $1.7B in crypto assets were stolen, and roughly 70% of those incidents involved mnemonic-compromise pathways. The core problem is structural: today’s mnemonics are bare assets—whoever sees them immediately owns everything, and there is no cryptographic-level protection, no rotation model, and no way to safely evolve toward PQC. I built MSCIKDF to directly solve this. MSCIKDF is a cryptographic primitive that introduces passphrase-sealed mnemonics, curve-isolated derivation, and rotatable secrets at the KDF layer. It ensures that: - The seed is never stored on disk, and never kept in memory—it only exists for ~20 microseconds during signing or verification. - A mnemonic and its passphrase can be rotated unlimited times without changing any addresses and without migrating assets. - One mnemonic supports all major elliptic-curve families (Ed25519, Secp256k1, sr25519, P-256, etc.), covering essentially all chains. - The algorithm is pluggable, allowing smooth PQC upgrades in the future while keeping the same mnemonic and the same addresses. - Bonus: it supports UNICODE (Chinese / Japanese / Korean / Arabic / Emoji) as passphrases. Why these properties are possible? Under the hood, MSCIKDF was designed around: - Single-root → multi-context isolation (each chain, wallet, device, agent, or application gets a mathematically isolated stream). - Zero-persistence secret handling (the derived seed is never kept in long-term memory or disk). - Rotatable passphrase sealing, allowing unlimited secret rotation with stable public identities. - Curve-agnostic, multi-algorithm derivation, supporting both signatures and encryption (Ed25519, X25519, Secp256k1, sr25519, ECDSA, etc.). - PQC compatibility, meaning post-quantum KDF modules can be plugged in without breaking identities or requiring wallet migrations. In short: MSCIKDF turns mnemonics from “bare private keys” into cryptographically protected, renewable, multi-curve identity roots. Paper (arXiv): https://ift.tt/PWC2YEn Playground: https://ift.tt/xpmI3T2 https://ift.tt/xpmI3T2 November 26, 2025 at 11:46PM

Show HN: Superglue – OSS integration tool that understands your legacy systems https://ift.tt/0V1phXA

Show HN: Superglue – OSS integration tool that understands your legacy systems If you've ever worked in a large company, you've probably encountered "shadow infrastructure": scripts nobody understands or custom connectors written once and never touched again. This glue layer isn't documented, isn't owned by anyone, and tends to break when systems are upgraded or someone leaves. It's also the part everybody dreads working on, because it's hard to understand, painful to work with, and full of unknown unknowns. We built superglue so that engineers stop wasting time on deciphering legacy APIs and documentation. superglue ingests existing glue code, SQL, configs, docs, OpenAPI specs and reverse-engineers what the system is actually doing. It then maps dependencies and regenerates everything as clean javascript code that can run directly or be exposed via MCP or SDK. It also monitors API changes and schema drift, and automatically repairs integrations when upstream systems change. In short: It turns legacy integrations into code you can easily understand, test, and update. So that engineers can do more exciting feature work, and companies can migrate and upgrade systems faster. Think of it as: a context engine + code generator + integration runtime for legacy glue. What we'd love feedback on - How do you deal with "nobody knows what this script does" situations? - What would you want to know about your legacy systems? OSS/community version: https://ift.tt/0wuSV3p More info: https://superglue.ai Happy to go deeper on the technical details. https://superglue.ai November 25, 2025 at 11:28PM

Show HN: SafeShare – Clean tracking params locally (PWA and bookmarklets https://ift.tt/5lm9kb7

Show HN: SafeShare – Clean tracking params locally (PWA and bookmarklets SafeShare is a tiny web app + bookmarklets that remove tracking params (utm_, gclid, fbclid, etc.) and unwrap common redirects – all locally in the browser (no accounts, no servers). • What it does: clean URLs, unwrap t.co/google redirectors, team whitelist for allowed params. • How it works: client-only JS, PWA + service worker; no network calls to our servers. • Try it: https://j-ai-71.github.io/Supersystem/quickstart.html • Feedback welcome (edge cases, params/redirectors to add). https://j-ai-71.github.io/Supersystem/ November 26, 2025 at 12:45AM

Show HN: We built an open source, zero webhooks payment processor https://ift.tt/F9DfeNv

Show HN: We built an open source, zero webhooks payment processor Hi HN! For the past bit we’ve been building Flowglad ( https://flowglad.com ) and can now feel it’s just gotten good enough to share with you all: Repo: https://ift.tt/qnMDskO Demo video: https://www.youtube.com/watch?v=G6H0c1Cd2kU 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. 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. 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). Why we built this: - 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 This all feels unduly clunky, esp when compared to how easy other layers like hosting and databases have gotten in recent years. 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. Our beta version lets you: - 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 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! https://ift.tt/qnMDskO November 26, 2025 at 12:33AM

Show HN: I built an interactive map of jobs at top AI companies https://ift.tt/eboDs1M

Show HN: I built an interactive map of jobs at top AI companies I built a live interactive map that shows where top AI companies hire around the world. I collected this data for a hackathon project. Many ATS providers have a public API that you can hit with the slug of the companies to get open jobs. The hardest part was finding the companies. I tried Firecrawl but it returned around 200 companies per provider which wasn’t enough for me. Then, I tried SERPAPI but it was expensive. I ended up using SearXNG to discover companies by ATS type and fetch their job postings. This produced a large dataset of 200k+ jobs (I only use a subset as it would have taken too much time processing). A few days ago, I decided to build a visualization of the data as I didn’t know what to do with it and wanted people to benefit. I kept catching myself wanting to ask simple questions like “show only research roles in Europe” or “filter for remote SWE positions” (and had plenty of free ai credits) so I added a small LLM interface that translates natural language into filters on the map. The map is built with Vite + React + Mapbox. Live demo: https://map.stapply.ai GitHub (data): https://ift.tt/8hEWnkD Would love feedback, ideas for improvement, or contributions. https://map.stapply.ai November 25, 2025 at 01:08AM

Show HN: Sphere-Base-One– A Python Kernel for Integer-Based Physics Optimization https://ift.tt/yFAwoOh

Show HN: Sphere-Base-One– A Python Kernel for Integer-Based Physics Optimization https://ift.tt/KdZSJgR November 24, 2025 at 11:53PM

Show HN: Pg-aiguide – Write better PostgreSQL code with AI https://ift.tt/6J0vmRY

Show HN: Pg-aiguide – Write better PostgreSQL code with AI Hi HN, I built a suite of tools to help ai generate better PostgreSQL code. The most interesting part is an opinionated set of skills to help it design better Postgres schemas. Also includes search over the manual. Deployeable as both an MCP server and as a Claude Code Plugin. I want to also include ecosystem docs and skills. Timescale (where I work) is already included. Looking for help with PostGIS and pgvector. Full open source. https://ift.tt/fHXah0K November 24, 2025 at 10:14PM

Show HN: OhNiceRepo – Easily discover trending GitHub gems and repos https://ift.tt/1TLz4w6

Show HN: OhNiceRepo – Easily discover trending GitHub gems and repos https://ohnicerepo.pages.dev November 24, 2025 at 01:41AM

Show HN: Genesis DB now provides a full gRPC API alongside HTTP https://ift.tt/m3cHjNS

Show HN: Genesis DB now provides a full gRPC API alongside HTTP The Protobuf definition is published openly: https://ift.tt/Qdv6Y1m Genesis DB also exposes gRPC Server Reflection, so clients can introspect the service without needing the .proto file locally (useful for tools like grpcurl, kreya, or dynamic client generation). https://ift.tt/Tu86vBp November 24, 2025 at 12:57AM

Show HN: Makefiles, Metalanguages, Matrioshka Automata https://ift.tt/GzeUuTd

Show HN: Makefiles, Metalanguages, Matrioshka Automata Immediately buried last time, so reposting for your lazy Sunday. This project is a strange labor of love, practically guaranteed to inspire horror and delight. It's also tough to summarize. Partly it's very practical and involves familiar tools, but part of it is also a new programming language with esolang roots. I'll start with the practical and move towards the peculiar. The one-sentence summary: compose.mk brings docker-fluency, polyglots, and a capable standard library to Makefiles. A more in-depth elevator pitch from the main landing page is below, and some related links at the end in footnotes 1-5. > Meet compose.mk, a tool / library / framework for Makefile-based automation, scripting, and lightweight orchestration. Native support for docker, docker-compose, workflow primitives, JSON IO, TUI elements, and more, all provided by a single file with no dependencies beyond what's already in your development environment. Typical use-cases include general project automation, especially decoupling your CI/CD from different kinds of platform lock-in. Other superpowers include the ability to quickly incorporate foreign tools and foreign code as first-class objects, which provides unique and powerful capabilities for quickly assembling console applications, systems prototyping, and component-oriented design experiments in general. Definitely not the Makefiles of your ancestors. Here's where it starts to get more weird and fun. Building the ideal environment for zero-dependency automation and pesky "glue code" moves in a certain direction. So it happens that compose.mk moonlights as an interpreter / compiler / packaging tool for a new kind of programming language. CMK-lang (or just CMK) is multiparadigm with diverse influences, from functional to concatenative, ultimately specializing in things like extensibility, interoperability, DAGs, and dispatch. CMK is a superset of Makefile that can be transpiled to vanilla Makefile. And it is what is known as a matrioshka language. Paraphrasing the definition from esolangs-wiki: > A matrioshka language is formed by bundling one or more meta-languages with one or more language descriptions. They can be identified by their program forms, which have multiple, distinct 'phases' with different syntactic and semantic rules. There are often two phases; the first gives a set of rules, and the second provides objects on which those rules are to be applied. In CMK-lang, matrioshka "objects" are things like container-runtimes or foreign interpreters, and "rules" are DAGs in the form of tasks, task-groups, or foreign code. For those interested in matrioshkas and topics in PLT, I suggest the alternate landing pages at footnotes 6-9. Love it or hate it, I think you'll agree that compose.mk is easily the biggest, baddest, most highly powered mutant Makefile the world has ever seen. If it helps you can think of CMK-lang as a PoC that's waiting for another back-end implementation ;) Playing around with it has convinced me though that the gap is real, and the world really needs containers-first matrioshka languages that work locally, and aren't tightly coupled to bulky remote platforms or infrastructure. It also needs languages that are capable of aggressively reusing and recombining existing code and existing tools. [1]: https://robot-wranglers.github.io/compose.mk/standard-lib [2]: https://robot-wranglers.github.io/compose.mk/bridge [3]: https://robot-wranglers.github.io/compose.mk/container-dispa... [4]: https://robot-wranglers.github.io/compose.mk/demos/polyglots [5]: https://robot-wranglers.github.io/compose.mk/json [6]: https://robot-wranglers.github.io/compose.mk/matrioshka [7]: https://robot-wranglers.github.io/compose.mk/language [8]: https://robot-wranglers.github.io/compose.mk/compiler [9]: https://robot-wranglers.github.io/compose.mk/demos/packaging https://robot-wranglers.github.io/compose.mk/ November 23, 2025 at 11:25PM

Show HN: Curious about tones in Chinese? An extension for language learners https://ift.tt/bdI2pqw

Show HN: Curious about tones in Chinese? An extension for language learners I made a simple Chrome extension that: * color codes Chinese text based on the tone of each syllable * provides a popup dictionary * gives example sentences, useful links, and AI analysis while you browse * integrates with Anki Connect for 1-click flashcard creation It can be installed here: https://ift.tt/seAvrDU... https://ift.tt/GO4p6Ii November 23, 2025 at 11:54PM

Show HN: Santamon – Lightweight macOS threat detection agent https://ift.tt/2zKEXOH

Show HN: Santamon – Lightweight macOS threat detection agent a lightweight macOS detection agent that taps into Santa’s Endpoint Security telemetry, runs CEL detection rules locally on-device, and only ships high-signal alerts to a tiny backend. basically a poor man’s macOS EDR for home labs and small fleets! https://ift.tt/Xx2s73d November 23, 2025 at 12:41AM

Show HN: Onlymaps, a Python Micro-ORM https://ift.tt/reslDNH

Show HN: Onlymaps, a Python Micro-ORM https://ift.tt/4YKL2vb November 22, 2025 at 09:54PM

Show HN: Mint – an open-source photo editor and digital compositor for the web https://ift.tt/OsVzhl9

Show HN: Mint – an open-source photo editor and digital compositor for the web A friend and I built a small multipurpose digital compositing tool meant to provide most of what you'd need for day-to-day meme-making, image markups, and collage creation in a static web app. Combines what I love about web platforms like Canva with a lower barrier to entry than heavyweight tools like the beloved Photopea. Basic mobile support included. Built in Svelte, with a dirt simple Canvas-powered rendering engine. Feature requests/bug reports/PR's strongly encouraged @ https://ift.tt/4ym5XUo https://mint.photo/ November 22, 2025 at 11:42PM

Show HN: Transcribe Your Voice in Terminal Locally https://ift.tt/TgPdrXR

Show HN: Transcribe Your Voice in Terminal Locally Use hns, a speech-to-text CLI tool to transcribe your voice from your microphone directly to clipboard. Integrate hns with Claude Code, Ollama, LLM, and more CLI tools for powerful workflows. hns transcribes your voice 100% locally using faster-whisper. The whisper model is downloaded automatically on first run and after that, hns can be used completely offline. After transcription, the text is displayed in the terminal (written to stdout) as well as automatically copied to your clipboard, ready to be pasted anywhere with Ctrl+V or Cmd+V. GitHub: https://ift.tt/TxzyW7t https://hns-cli.dev/ November 22, 2025 at 12:04AM

Show HN: Wealthfolio 2.0- Open source investment tracker. Now Mobile and Docker https://ift.tt/myX24DE

Show HN: Wealthfolio 2.0- Open source investment tracker. Now Mobile and Docker Hi HN, creator of Wealthfolio here. A year ago, I posted the first version. Since then, the app has matured significantly with two major updates: 1. Multi-platform Support: Now available on Mobile (iOS), Desktop (macOS, Windows, Linux), and as a Self-hosted Docker image. (Android coming soon). 2. Addons System: We added explicit support for extensions so you can hack around, vibe code your own integrations, and customize the app to fit your needs. The core philosophy remains the same: Always private, transparent, and open source. https://ift.tt/iwK6WgB November 21, 2025 at 11:34PM

Show HN: MCP Traffic Analysis Tool https://ift.tt/JamsVzD

Show HN: MCP Traffic Analysis Tool https://ift.tt/QOjRbI4 November 18, 2025 at 01:06AM

Show HN: Tangent – Open-source security data pipeline https://ift.tt/ZdKY8z0

Show HN: Tangent – Open-source security data pipeline Hi HN! We’re Ethan and Danny, the authors of Tangent ( https://ift.tt/gtfc6Z5 ), a Rust-based log pipeline where all normalization, enrichment, and detection logic runs as WASM plugins. We kept seeing the same problems in the OCSF ( https://ocsf.io ) community: 1) Schemas change constantly. Large companies have whole teams dedicated to keeping vendor→OCSF mappings up to date. 2) There’s no shared library of mappings, so everyone recreates the same work. 3) Writing mappers is tedious, repetitive work. 4) Most pipelines use proprietary DSLs that are hard to share and hard for tools/LLMs to generate. Tangent takes a different approach: no DSLs – mappings and enrichments are just normal code compiled to WASM, shareable plugins – we maintain a community library ( https://ift.tt/7zKbE0N ), interoperability – we can run other engines’ DSLs (e.g., Bloblang) inside WASM for easy migration, full flexibility – plugins can validate schemas, call external APIs ( https://ift.tt/foK9POV... ), or perform complex transforms ( https://ift.tt/bl2Oaeo... ). Here's an example Python transformation plugin to drop all fields from a log except `message`: import json from typing import List from wit_world.imports import log # `log.Logview` is Tangent's zero-copy JSON accessor type. def process_logs(self, logs: List[log.Logview]) -> bytes: out = bytearray() for lv in logs: msg = lv.get("msg") value = msg.value if msg is not None else "" out.extend(json.dumps({"message": value}).encode() + b"\n") return bytes(out) We have plenty more examples in the repo. Because plugins are just Go/Python/Rust, LLMs can create new mappers with ease. For example, I asked: Generate a mapper from AWS Security Hub Finding to OCSF and only had to make a few minor tweaks. ( https://ift.tt/LAMY2Xg... ) Performance-wise, a 16-core Amazon Linux box processes ~480 MB/s end-to-end (TCP → Rust-WASM transform → sink) on ~100-byte JSON logs. The CLI includes tooling to scaffold, test, and benchmark plugins locally. Here's a deep dive into how we are able to get this performance: https://ift.tt/ZRltYBP . We’d love to get your feedback! What do you think? https://ift.tt/gtfc6Z5 November 20, 2025 at 11:41PM

Show HN: Rapid-rs – Zero-config web framework for Rust https://ift.tt/ZjYpk20

Show HN: Rapid-rs – Zero-config web framework for Rust I built rapid-rs to eliminate the hours of boilerplate when starting a new Rust web service. One command gets you: - Auto-configured DB, logging, CORS - OpenAPI/Swagger UI at /docs - Request validation - Production-ready observability Built on Axum. Early benchmarks show ~50K req/s with 10-20MB RAM. This is v0.1 - feedback welcome! Crates: https://ift.tt/J6c3Gvm https://ift.tt/J6c3Gvm November 20, 2025 at 11:15PM

Show HN: Chrome Store–featured extension that writes X replies via DOM observers https://ift.tt/dvc2wpg

Show HN: Chrome Store–featured extension that writes X replies via DOM observers Hi HN, A couple of months ago I posted an early version of this Chrome extension. Since then I’ve refined it, fixed a bunch of stability issues, and it was recently featured on the Chrome Web Store’s “Featured” section, which was a nice surprise. What the extension does: – Detects the active tweet or thread directly in the browser – Collects relevant context (parent tweet, author info, thread shape) – Formats a prompt and sends it to the OpenAI API – Inserts the generated reply straight into Twitter’s native reply box All of this happens inside the X.com DOM, without storing any user data. Technical bits: – Uses MutationObserver to track X.com’s constantly changing DOM – Handles dynamically inserted tweet nodes, shadow DOM, and thread expansions – Debounces context extraction to avoid unnecessary re-runs – Simulates native input events to inject the reply so it feels built-in – Avoids backend state; everything is read client-side except the final API call Challenges: – X.com changes UI structure often, so selectors break unpredictably – Preventing duplicate injections when the DOM re-renders – Keeping prompt size small enough for fast generation – Reducing overhead so the extension doesn’t slow down the page Recent improvements: – More stable tweet/thread detection – Better context selection logic – Cleaner UI in the reply popup – Small performance fixes and race-condition fixes Chrome Store page: https://ift.tt/Va5dWqs... Would appreciate feedback from people who’ve built browser extensions or dealt with X.com’s DOM patterns. Happy to discuss any details. https://ift.tt/YGP9UZp November 20, 2025 at 10:28PM

Show HN: I made a down detector for down detector https://ift.tt/UZJsd9M

Show HN: I made a down detector for down detector After down detector went down with the rest of the internet during the Cloudflare outage today I decided to build a robust, independent tool which checks if down detector is down. Enjoy!! https://ift.tt/rfSRpAo November 19, 2025 at 07:05AM

Show HN: DNS Benchmark Tool – Compare and monitor resolvers https://ift.tt/p42EeLI

Show HN: DNS Benchmark Tool – Compare and monitor resolvers I built a CLI to benchmark DNS resolvers after discovering DNS was adding 300ms to my API requests. v0.3.0 just released with new features: compare: Test single domain across all resolvers top: Rank resolvers by latency/reliability/balanced monitor: Continuous tracking with threshold alerts 1,400+ downloads in first week. Quick start: pip install dns-benchmark-tool dns-benchmark compare --domain google.com CLI stays free forever. Hosted version (multi-region, historical tracking, alerts) coming Q1 2026. GitHub: https://ift.tt/OusPfLV Feedback: https://forms.gle/BJBiyBFvRJHskyR57 Built with Python + dnspython. Open to questions and feedback! https://ift.tt/OusPfLV November 20, 2025 at 12:52AM

Show HN: ChunkBack – A Fake LLM API server for testing apps without paying https://ift.tt/duQr3VS

Show HN: ChunkBack – A Fake LLM API server for testing apps without paying Hi HN, I've been working with LLMs in production for a while both as a solo dev building apps for clients and working at an AI startup. The one thing that always was a pain was to pay OpenAI/Gemini/Anthropic a few dollars a month just for me to say "test" or have a CI runner validate some UI code. So I built this server called ChunkBack, that mocks the popular llm provider's functionality but allows you to type in a deterministic language: `SAY "cheese"` or `TOOLCALL "tool_name" {} "tool response"` I've had to work in some test environments and give good results for experimenting with CI, but it's still an early project so would love feedback and more testers on. https://ift.tt/IFdKN9D November 19, 2025 at 11:12PM

Show HN: Copus – Internet gem marketplace for bookmark collectors (x402-powered) https://ift.tt/IEZodrB

Show HN: Copus – Internet gem marketplace for bookmark collectors (x402-powered) Hey HN! We’re a small team of artists, developers, and coffee lovers who’ve watched a lot of websites we love shut down over the years. We’ve been looking for a way to support them with income and exposure. We see that more people are interacting with the web through AI instead of visiting sites directly, so the ad-based model is breaking. The open web needs a new business model. Our take is to incentivize people (and, in the future, AI agents) to find and share valuable content (links), with both the finder and the original creator rewarded. Along the way we were inspired by discussions like: Pocket shut down: https://ift.tt/VHeZ9fS x402 protocol: https://ift.tt/rCJwSBf “To survive the AI age, the web needs a new business model”: https://ift.tt/MdU8kvu Key features Social bookmarking It’s like a decentralized Digg or a Pinterest-for-websites. You can share (curate) any URI (URL) through the site or the browser extension. Others can collect and build on your collections. Pay-to-visit Finding valuable content is valuable. You can set a stablecoin price for visiting a link you shared. Payments are powered by the x402 protocol. Support sites/content you love Half of the pay-to-visit revenue goes to the author of the original content, claimable after they opt into x402 or register a Copus account. Permanent storage Your collections (bookmarks) are automatically stored on the Arweave blockchain. We pay the storage fees so you’ll never lose them. Other features we have in mind Spaces Like Pinterest boards, to organize your collections and collaborate with others. Weave If a link reminds you of another link, you can “weave” them together in a “you may also like” section. It’s a bit like a collective Obsidian graph where standalone websites become a connected map and every site is a rabbit hole. AI agent support You can train agents to curate and purchase for you. Social features Follow accounts with great taste. Who we imagine this is for If you’ve been bookmarking over the years, you already have tons of internet gems in hand! Please pick the best ones to share with the world. They’re valuable for both readers and original creators. Were you a Pocket user? Save your best bookmarks here and never lose them. (We plan to support putting a copy of the whole website on-chain once the project scales. Right now we put the link, category info, and your recommendation notes on-chain for free.) Some other things Copus is open source, with the frontend built using Claude Code. We plan to launch a governance token to put ownership of the project into the hands of the people who use it. We don’t mess with rights and privacy. Aside from some essential terms needed to keep the project running, your rights remain yours. Copus has a Chinese version (Copus.io), which is a haven for around 150k Chinese fan-fiction lovers rn. We might merge the two sites once the English content reaches scale or we might not. How we plan to make money We’re still figuring it out. The first idea is: Take a 10% fee on each payment. Put unclaimed creator earnings into low-risk investments (similar to how stablecoins earn yield). Hope you enjoy Copus, and thank you in advance for trying it out early! https://ift.tt/XViGznL November 19, 2025 at 01:18AM

Show HN: Guts – convert Golang types to TypeScript https://ift.tt/j1gcVLu

Show HN: Guts – convert Golang types to TypeScript https://ift.tt/p4U0fzI November 19, 2025 at 12:55AM

Show HN: Optimizing LiteLLM with Rust – When Expectations Meet Reality https://ift.tt/10HbzSO

Show HN: Optimizing LiteLLM with Rust – When Expectations Meet Reality I've been working on Fast LiteLLM - a Rust acceleration layer for the popular LiteLLM library - and I had some interesting learnings that might resonate with other developers trying to squeeze performance out of existing systems. My assumption was that LiteLLM, being a Python library, would have plenty of low-hanging fruit for optimization. I set out to create a Rust layer using PyO3 to accelerate the performance-critical parts: token counting, routing, rate limiting, and connection pooling. The Approach - Built Rust implementations for token counting using tiktoken-rs - Added lock-free data structures with DashMap for concurrent operations - Implemented async-friendly rate limiting - Created monkeypatch shims to replace Python functions transparently - Added comprehensive feature flags for safe, gradual rollouts - Developed performance monitoring to track improvements in real-time After building out all the Rust acceleration, I ran my comprehensive benchmark comparing baseline LiteLLM vs. the shimmed version: Function Baseline Time Shimmed Time Speedup Improvement Status token_counter 0.000035s 0.000036s 0.99x -0.6% count_tokens_batch 0.000001s 0.000001s 1.10x +9.1% router 0.001309s 0.001299s 1.01x +0.7% rate_limiter 0.000000s 0.000000s 1.85x +45.9% connection_pool 0.000000s 0.000000s 1.63x +38.7% Turns out LiteLLM is already quite well-optimized! The core token counting was essentially unchanged (0.6% slower, likely within measurement noise), and the most significant gains came from the more complex operations like rate limiting and connection pooling where Rust's concurrent primitives made a real difference. Key Takeaways 1. Don't assume existing libraries are under-optimized - The maintainers likely know their domain well 2. Focus on algorithmic improvements over reimplementation - Sometimes a better approach beats a faster language 3. Micro-benchmarks can be misleading - Real-world performance impact varies significantly 4. The most gains often come from the complex parts, not the simple operations 5. Even "modest" improvements can matter at scale - 45% improvements in rate limiting are meaningful for high-throughput applications While the core token counting saw minimal improvement, the rate limiting and connection pooling gains still provide value for high-volume use cases. The infrastructure I built (feature flags, performance monitoring, safe fallbacks) creates a solid foundation for future optimizations. The project continues as Fast LiteLLM on GitHub for anyone interested in the Rust-Python integration patterns, even if the performance gains were humbling. Edit: To clarify - the negative performance for token_counter is likely in the noise range of measurement, suggesting that LiteLLM's token counting is already well-optimized. The 45%+ gains in rate limiting and connection pooling still provide value for high-throughput applications. https://ift.tt/WYhO2AE November 18, 2025 at 11:32PM

Show HN: UpBeat – an AI-Enhanced RSS/Atom Reader that only shows you good news https://ift.tt/HVl1bJp

Show HN: UpBeat – an AI-Enhanced RSS/Atom Reader that only shows you good news Hey everyone, I'm Sean, and I've built UpBeat. Why did I build this? Well, the world is more complex than ever, and every stream, device and social feed screams for our attention whilst telling us that everything is awful. While it's important to know what's going on in the world - do we really need to be bombarded with negativity 24/7? Absolutely not! It's bad for our mental health, it's bad for our attention spans, and it's bad for society as a whole. So that's why I built UpBeat - My friends, loved ones, and I needed a break from the doom cycle. So, here it is :) Some technical details, it's a macOS app built with Go using the Wails.io framework and it (currently) uses the Distilbert model which runs on the Apple Neural engine, so inference takes ~40ms. https://ift.tt/TcJyrpn November 18, 2025 at 12:51AM

Show HN: Kalendis – Scheduling API (keep your UI, we handle timezones/DST) https://ift.tt/lJZvRx9

Show HN: ToolHop – Fast, simple utilities for every workflow https://ift.tt/EPjULg1

Show HN: ToolHop – Fast, simple utilities for every workflow ToolHop is your all-in-one browser toolbox with 200+ fast-loading calculators, converters, generators, color labs, and dev helpers. Use global search or curated categories to jump straight into the right utility, run it client-side for instant feedback, and deep-link results to your team. Whether you’re formatting copy, validating data, checking DNS, or exploring palettes, ToolHop keeps your core workflows a single tab away, and it’s entirely free, no account required. --- I built ToolHop because I was sick of the usual “free tool” bait-and-switch. Every time I needed to convert an image, compress a file, check some text, or run a quick calculation, I’d end up hitting some arbitrary limit like “10 uses per week” or a forced signup wall. It’s ridiculous how something as basic as converting a JPG to a PNG can turn into a subscription pitch. So ToolHop started as a personal frustration project: I wanted a single place with a ton of genuinely useful tools that didn’t nag, lock you out, or throttle you. Over time that grew into 200+ handcrafted tools, all fast, simple, and actually free. No trickery, no timers, no limits. As I built it, the process became about consistency and quality. I wanted the tools to feel seamless, not slapped together. That meant focusing on speed, clean UI, accurate results, and making sure each tool works instantly without friction. The goal was always the same: a site that respects people’s time. Something you can rely on whenever you just need a tool to work. If ToolHop saves someone even a few minutes of hassle, then the project did its job. https://toolhop.app November 17, 2025 at 10:58PM

Show HN: Octopii, a framework for building distributed applications in Rust https://ift.tt/MkCuIr8

Show HN: Octopii, a framework for building distributed applications in Rust it won't let me put url for some reason, here it is: https://ift.tt/0q8NplX November 17, 2025 at 11:45PM

Show HN: ResendForward – OS server and UI for use with Resend.com inbound https://ift.tt/mUExwal

Show HN: ResendForward – OS server and UI for use with Resend.com inbound With Resend's new inbound feature I wanted to build a simple application that handles processing webhook events and forwarding emails for multiple applications. Right now Resend requires you to implement that logic in each new application. repo - https://ift.tt/jFGmBkJ live - https://ift.tt/xKfSqic Built with react + pocketbase, extremely simple to self host. https://ift.tt/jFGmBkJ November 17, 2025 at 02:27AM

Show HN: Minivac 601 Simulator - a 1961 Relay Computer https://ift.tt/YXiIcPr

Show HN: Minivac 601 Simulator - a 1961 Relay Computer Hey HN! I'm very proud of sharing this project with you all, after ~2 years of starts and stops, and about 5 different attempts at making it. Context/history: In 1961, the Minivac 601 [0], an educational electronics kit - somewhat similar to those "300 circuits in one" you may have had growing up as I did - was created by none other than Claude Shannon. The Minivac is disarmingly simple: it consists roughly speaking of 6 relays, 12 lights, 6 buttons, and a motorized wheel. You'd think that it couldn't really do much. Well, amazingly, it can do a lot. You can wire up the components in a way that will make the Minivac play tic-tac-toe, or OCR-detect 10 digits... The sample "demo" circuit I chose for the homepage shows a binary counter that counts up to 7. Another amazing thing about the Minivac is definitely its manuals [1]. Their spirit is what I hope to capture in the coming (years?) as I keep improving this project. The manuals are generous and well-written and are not only an amazing gradual introduction to relay-based logic - they touch on computing at large. With amazing 1960s graphics/cartoons, of course. That's probably what got me to work on the Minivac. I learned about the device a bit before going to the Recurse Center, fell in love with the manuals, and was frustrated that I couldn't try out the circuits or play around with the device! I thought that creating a JavaScript-based emulator would be an "easy" way to get there. Turns out that correctly simulating electricity isn't "easy". :-) But I'm very proud that it now seems to be doing the right thing for most circuits that I've tested from the book. Yes, this Minivac Simulator has a TypeScript testing suite! Looking forward to hearing from you all. Cheers [0] https://ift.tt/HEAFQRk [1] https://ift.tt/g9aWuR2... Repo: https://ift.tt/kHto25Z https://ift.tt/yDk63Z8 November 16, 2025 at 10:24PM

Show HN: An Apache Beam batch processing clone in Rust https://ift.tt/dGFgi5b

Show HN: An Apache Beam batch processing clone in Rust I've been experimenting with Apache Beam as of late at work and found that it can be slow in Python, and more complicated to use in Java where performance is better. I decided to experiment with JetBrains' AI Assistant and build an Apache Beam clone in Rust. I appreciate any commentary or feedback! https://ift.tt/wxrJeiI November 16, 2025 at 01:46AM

Show HN: DeepClause – A Neurosymbolic AI System Built on WASM and Prolog https://ift.tt/tA8aEcR

Show HN: DeepClause – A Neurosymbolic AI System Built on WASM and Prolog Hi HN, Today I'd like to present the results of my weekend project of the last year or so. Given there are many posts on HN about LLMs and Prolog, I thought that this would be of interest. DeepClause is my own (possibly misguided :-) attempt at combining LLMs with Logic Programming, ultimately hoping to establish a foundation for building more reliable agents, that produce reproducible and fully traceable result. At the heart of DeepClause is a DSL called "DeepClause Meta Language" (DML) which can be used to encode agent behaviors as executable logic programs. DML is executed by a meta-interpreter implemented in Prolog and thus natively supports things like constraint logic programming, knowledge graphs, symbolic reasoning, ... The DML interpreter itself runs inside the SWI Prolog WASM module, thus allowing for a secure and sandboxed execution environment for AI agents. The project is still rough around a lot of edges, but I'd love to get some feedback and comments. https://ift.tt/BKv3bCW November 15, 2025 at 08:53PM

Show HN: ByteSync – Open-source hybrid file sync (LAN and remote, E2EE) https://ift.tt/eElSVuY

Show HN: ByteSync – Open-source hybrid file sync (LAN and remote, E2EE) Hi everyone, I've been developing ByteSync, an open-source file synchronization, backup and deduplication tool designed to bridge the gap between local and remote sync. In spirit, it's somewhat closer to FreeFileSync, but with an integrated networking layer and end-to-end encryption — which means you can synchronize files between computers on the same LAN or across the internet without VPNs or firewall setup. Everything works transparently through the same interface. The synchronization model is based on DataNodes (which represent repositories, such as servers or NAS devices) and DataSources (the folders or files inside them). A session can include multiple participants, each with one or several DataNodes, and ByteSync handles all comparisons and transfers automatically. To optimize performance, the engine uses a two-stage inventory process: an initial indexation followed by comparisons limited to items that actually changed. This keeps synchronization fast even with large datasets. There's also a flat mode, useful when structure doesn't matter and you just want to compare or align files by name. Currently, ByteSync is focused on interactive synchronization — it's not yet automated or daemon-based (CLI integration is planned). But it's already fully functional for discovering and managing differences between repositories, both local and remote. ByteSync runs on Windows, macOS, and Linux, and the entire codebase is available on GitHub: https://ift.tt/Yj5NF4B You can also download binaries and read the documentation here: https://ift.tt/OTPz0px I'd really appreciate feedback and contributors — whether on usability, architecture, or ideas for future features. The goal is to make a solid, privacy-respectful alternative for hybrid file synchronization that remains simple to use and open for everyone. November 14, 2025 at 09:02PM

Show HN: spymux – Spy on your tmux panes https://ift.tt/gKQw8xb

Show HN: spymux – Spy on your tmux panes I had motivation for writing this after I kept switching back and forth between agents to see if they've finished what they were working on (and I couldn't find a similar tool out there). I'd imagine it can be useful for other scenarios as well, e.g. tracking multiple build/test jobs over multiple windows. Still a work in progress, but I thought I'd share :) https://ift.tt/AxrqswV November 15, 2025 at 12:09AM

Show HN: TalkiTo – enabling voice and Slack for Claude Code and Codex CLI https://ift.tt/T29CNUR

Show HN: TalkiTo – enabling voice and Slack for Claude Code and Codex CLI Hey everyone, here is an open source project I've been working on to add voice input/output to terminal based coding agents. One thing about the new terminal coding agents I really like is being able to multi-task but right now it's a bit like a Tesla on autopilot needing your hands still on the wheel. You need to be checking often if your input is required or if it's going off the rails. To be able to go fully hands free I wanted to add TTS and ASR. Then I added slack and WhatsApp hooks to TalkiTo as well. It's fully open source with a BYOK philosophy and it's configured to work with any of the major ASR/TTS providers. It also supports local whisper and kokoro/kittentts if you want a decent free/private option. It works by wrapping the coding agent and capturing the input/output. It does have an MCP server running but thats mainly for configuration - I found that using MCP to speak or listen was too slow. The upshot of the MCP server is you can type (or say) "talkito disable ASR" or "talkito change tts to kokoro". Here is a demo video I made here: https://www.youtube.com/watch?v=pf8jFt0smqs I like to think of it as similar to SuperWhisper but with TTS, the focus on coding agents and configurability. Really curious to get feedback. Thanks! https://ift.tt/fH8PRzA November 14, 2025 at 10:47PM

Show HN: Fine-tune open-source LLMs quickly and easily (early access) https://ift.tt/tIPsNFR

Show HN: Fine-tune open-source LLMs quickly and easily (early access) https://www.tinytune.xyz/ November 14, 2025 at 02:03AM

Show HN: What Can Happen When You Code While Overtired https://ift.tt/ZcmNizI

Show HN: What Can Happen When You Code While Overtired Generative art has been a long-standing interest of mine. I’ve seen a lot over the years — but this generation was absolutely pure. https://number-garden-story.netlify.app/ November 14, 2025 at 12:38AM

Show HN: A game of higher or lower using GitHub stars https://ift.tt/xAONJUd

Show HN: A game of higher or lower using GitHub stars Its unpolished v0 so I am releasing it to get general feedback.Built extensively with the aid of AI https://ift.tt/QYB48jf November 13, 2025 at 08:39PM

Show HN: Shadowfax AI – an agentic workhorse to 10x data analysts productivity https://ift.tt/g0uwi8D

Show HN: Shadowfax AI – an agentic workhorse to 10x data analysts productivity Hi HN, We built an AI agent for data analysts that turns the soul crushing spreadsheet & BI tool grind into a fast, verifiable and joyful experience. Early users reported going from hours to minutes on common real-world data wrangling tasks. It's much smarter than an Excel copilot: immutable data steps, a DAG of SQL views, and DuckDB for instant crunching over millions of rows. Our early agent prototype ranked #1 on the Spider2-DBT bench. https://spider2-sql.github.io Try it out and we'd love your feedback! Thanks, Di Wu & the Shadowfax team P.S. Shadowfax is Gandalf's horse from LOTR. There's a hidden easter egg site with 3 different triggers, see if you can find them. https://ift.tt/HxhEuNI November 13, 2025 at 08:04PM

Show HN: JavaScript Engines Zoo https://ift.tt/FrSW1JQ

Show HN: JavaScript Engines Zoo https://ift.tt/3c4bunM November 12, 2025 at 11:02PM

Show HN: Cancer diagnosis makes for an interesting RL environment for LLMs https://ift.tt/mAxRM1N

Show HN: Cancer diagnosis makes for an interesting RL environment for LLMs Hey HN, this is David from Aluna (YC S24). We work with diagnostic labs to build datasets and evals for oncology tasks. I wanted to share a simple RL environment I built that gave frontier LLMs a set of tools that lets it zoom and pan across a digitized pathology slide to find the relevant regions to make a diagnosis. Here are some videos of the LLM performing diagnosis on a few slides: ( https://www.youtube.com/watch?v=k7ixTWswT5c ): traces of an LLM choosing different regions to view before making a diagnosis on a case of small-cell carcinoma of the lung ( https://youtube.com/watch?v=0cMbqLnKkGU ): traces of an LLM choosing different regions to view before making a diagnosis on a case of benign fibroadenoma of the breast Why I built this: Pathology slides are the backbone of modern cancer diagnosis. Tissue from a biopsy is sliced, stained, and mounted on glass for a pathologist to examine abnormalities. Today, many of these slides are digitized into whole-slide images (WSIs)in TIF or SVS format and are several gigabytes in size. While there exists several pathology-focused AI models, I was curious to test whether frontier LLMs can perform well on pathology-based tasks. The main challenge is that WSIs are too large to fit into an LLM’s context window. The standard workaround, splitting them into thousands of smaller tiles, is inefficient for large frontier LLMs. Inspired by how pathologists zoom and pan under a microscope, I built a set of tools that let LLMs control magnification and coordinates, viewing small regions at a time and deciding where to look next. This ended up resulting in some interesting behaviors, and actually seemed to yield pretty good results with prompt engineering: - GPT 5: explored up to ~30 regions before deciding (concurred with an expert pathologist on 4 out of 6 cancer subtyping tasks and 3 out of 5 IHC scoring tasks) - Claude 4.5: Typically used 10–15 views but similar accuracy as GPT-5 (concurred with the pathologist on 3 out of 6 cancer subtyping tasks and 4 out of 5 IHC scoring tasks) - Smaller models (GPT 4o, Claude 3.5 Haiku): examined ~8 frames and were less accurate overall (1 out of 6 cancer subtytping tasks and 1 out of 5 IHC scoring tasks) Obviously, this was a small sample set, so we are working on creating a larger benchmark suite with more cases and types of tasks, but I thought this was cool that it even worked so I wanted to share with HN! November 13, 2025 at 12:01AM

Show HN: SQL++ – 5x faster than Prisma (Rust) https://ift.tt/rNuwqeM

Show HN: SQL++ – 5x faster than Prisma (Rust) I built SQL++, a type-safe SQL library for Rust using PostgreSQL's binary protocol. Benchmarks vs Prisma (5,000-10,000 queries): - Simple queries: 1.5x faster - Complex aggregations: 19.9x faster - Batch inserts: 5.6x faster - Average: 5x faster One benchmark didn't finish in Prisma (crashed), SQL++ completed in 2.5min. Why faster: 1. No runtime query building - validates once, caches forever 2. Zero ORM overhead - direct struct mapping 3. Binary protocol - implemented PostgreSQL wire protocol from scratch Currently supports: - Full SQL (CTEs, window functions, JOINs, subqueries) - DDL (CREATE/ALTER/DROP TABLE, indexes) - ~60% of SQL spec Limitations: - PostgreSQL only - v0.1 (expect bugs) - No ORM relationships (by design) Built as a high school project. Feedback welcome! GitHub: https://ift.tt/opOUvNs Benchmarks: https://ift.tt/cspvkt9 https://ift.tt/opOUvNs November 12, 2025 at 11:33PM

Show HN: Open-Source LaTeX OCR, Alternative to Mathpix/SimpleTex https://ift.tt/N17m8QP

Show HN: Open-Source LaTeX OCR, Alternative to Mathpix/SimpleTex https://texocr.netlify.app/ November 12, 2025 at 11:01PM

Show HN: dspx — Serverless-friendly DSP for Node.js (native C++ + Redis state) https://ift.tt/0sjoRhr

Show HN: dspx — Serverless-friendly DSP for Node.js (native C++ + Redis state) https://ift.tt/ZLYmxTW November 11, 2025 at 09:58PM

Show HN: Tracking AI Code with Git AI https://ift.tt/239jyU0

Show HN: Tracking AI Code with Git AI Git AI is a side project I created to track AI-generated code in our repos from development, through PRs, and into production. It does not just count lines, it keeps track of them as your code evolves, gets refactored and the git history gets rewritten. Think 'git blame' but for AI code. There's a lot about how it works in the post, but wanted to share how it's been impacting me + my team: - I find I review AI code very differently than human code. Being able to see the prompts my colleagues used, what the AI wrote, and where they stepped in to override has been extraordinarily helpful. This is still very manual today, but hope to build more UI around it soon. - “Why is this here?” — more than once I’ve giving my coding agent access to the past prompts that generated code I’m looking at, which lets the Agent know what my colleague was thinking when they made the change. Engineers talk to AI all day now…their prompts are sort of like a log of thoughts :) - I pay a lot of attention to the lines generated for every 1 accepted ratio. If it gets up over 4 or 5 it means I’m well outside the AI’s distribution or prompting poorly — either way, it’s a good cause for reflection and I’ve learned a lot about collaborating with LLMs. This has been really fun to build, especially because some amazing contributors who were working on similar projects came together and directed their efforts towards Git AI shine. We hope you like it. https://ift.tt/PRcafoi November 11, 2025 at 12:26AM

Show HN: Tiny Diffusion – A character-level text diffusion model from scratch https://ift.tt/BKQT3FS

Show HN: Tiny Diffusion – A character-level text diffusion model from scratch https://ift.tt/4VXmMeR November 10, 2025 at 10:13PM

Show HN: Trilogy Studio, open-source browser-based SQL editor and visualizer https://ift.tt/0MY7XD9

Show HN: Trilogy Studio, open-source browser-based SQL editor and visualizer SQL-first analytic IDE; similar to Redash/Metabase. Aims to solve reuse/composability at the code layer with modified syntax, Trilogy, that includes a semantic layer directly in the SQL-like language. Status: experiment; feedback and contributions welcome! Built to solve 3 problems I have with SQL as my primary iterative analysis language: 1. Adjusting queries/analysis takes a lot of boilerplate. Solve with queries that operate on the semantic layer, not tables. Also eliminates the need for CTEs. 2. Sources of truth change all the time. I hate updating reports to reference new tables. Also solved by the semantic layer, since data bindings can be updated without changing dashboards or queries. 3. Getting from SQL to visuals is too much work in many tools; make it as streamlined as possible. Surprise - solve with the semantic layer; add in more expressive typing to get better defaults;also use it to wire up automatic drilldowns/cross filtering. Supports: bigquery, duckdb, snowflake. Links [1] https://ift.tt/2f30Dqv (language info) Git links: [Frontend] https://ift.tt/f89JCk3 [Language] https://ift.tt/pca8G4e Previously: https://ift.tt/0Nq46Oo (significant UX/feature reworks since) https://ift.tt/dsp6FM0 https://ift.tt/kDFQUl0 November 10, 2025 at 06:26AM

Show HN: I'm a pastor/dev and built a 200M token generative Bible https://ift.tt/SKcXniL

Show HN: I'm a pastor/dev and built a 200M token generative Bible https://ift.tt/rD70ntq November 10, 2025 at 03:11AM

Show HN: Every-few-days satellite timeline for any spot, Sentinel-2 SR https://ift.tt/Dx4ZyFJ

Show HN: Every-few-days satellite timeline for any spot, Sentinel-2 SR I built anicha.earth because I kept needing a fast, no-frills way to see how a place changes over time — not once a year, but every week or so. Recently I worked on super-resolution for Sentinel-2 (about 8–10× upscaling) for an agriculture project. Along the way I realized two things: (1) this could be useful beyond ag, and (2) I couldn’t find a tool that lets you pick any area and quickly scrub through years of imagery. So I made one that’s as simple and as fast as I can make it. Under the hood it uses Copernicus Sentinel-2 L2A (10 m/pixel). With S2A+B the nominal revisit is ~5 days (depends on clouds; with Sentinel-2C the cadence is tighten further). For any area you select, the app gathers all available scenes since 2018 and shows them on the map and in a time strip for easy scrubbing. There’s AI-Enhanced view: a super-resolution model that makes it toward ~1–2 m. The model was trained on millions of satellite/aerial images, primarily open NAIP data. This is an early beta and a bit rough. I’m most curious about now is where this is actually useful? https://anicha.earth https://mzoom.space November 9, 2025 at 09:20PM

Show HN: I built a website to visualize company financial data https://ift.tt/B3NjA0h

Show HN: I built a website to visualize company financial data Hi HN, I built a website myfinsight.com that aims to make complicated company financials easy to understand. The problem: The go-to place for financial data such as revenue, sales, net income is Yahoo finance. However, their data is usually wrong and very limited. The numbers are hard to digest to get insight quickly. There are also numerous websites that provide much better data for a very expensive monthly fee. Solution: a website that provides free diagrams and charts that visualize important financial data, such as income growth rate by date, revenue breakdown etc. It is free because the financial data process is highly automated without manual input and correction. I used to send the finance infographics to friends and family. I found it easier just to make a website and they can grab the data from it. Next steps: there is a long tail of companies that don’t file their reports correctly. I am trying to make it more accurate somehow, and maybe add live stock prices to the website. I am also looking for feedback! Please play around with it and let me know if something is wrong. https://myfinsight.com/ November 9, 2025 at 04:30AM

Show HN: Easily reduce GitHub Actions costs with Ubuntu-slim migration https://ift.tt/SnK04ga

Show HN: Easily reduce GitHub Actions costs with Ubuntu-slim migration Hi, HN! I've been running GitHub Actions workflows for a while, and when GitHub announced ubuntu-slim runners as a cheaper alternative to ubuntu-latest, I wanted to migrate. (Blog: https://ift.tt/Cdi84gT... ) But manually checking which workflows can safely migrate is tedious—you need to check for Docker usage, services, containers, execution times, and missing commands. So I built gh-slimify, a GitHub CLI extension that automates this. It scans your workflows, detects migration candidates, checks for incompatible patterns, identifies missing commands, and can safely update workflows with one command. Try it: gh extension install fchimpan/gh-slimify gh slimfy # Scan workflows gh slimfy fix # Update safe jobs only Open source (MIT). I'd love feedback on how to improve it or what edge cases I might have missed. https://ift.tt/znxkJSi November 8, 2025 at 11:49PM

Show HN: I combine Htmx, LiveView and SolidJS for interactive server components https://ift.tt/G6B8EFH

Show HN: I combine Htmx, LiveView and SolidJS for interactive server components I like htmx, LiveView, React and Solid. They are great at different points, so I try to combine them in Solv (Stateless Offline-capable LiveView) and write a prototype to show the benefits. Solv's main idea is that stateless servers keep client's state in a volatile cache. It enables server components that are also interactive, which is best of both worlds between LiveView and htmx. Then fine-grained reactivity is added to achieve efficient DOM updates + minimal payload size. This provides: - SSR with close-to-zero rehydration cost. - No API endpoints, server can just read from DB then render & update clients directly. - Server components that are interactive. - Minimal payload for updates from server. - Stateless servers that can handle stateful-like request/response. - Avoid consistent connection to server, clients can work offline after page load, update local - state, keep pending server requests and sync later (can also use a sync engine like InstantDB to simplify some part of the page). Repo: https://ift.tt/HRWTIEe Demos deployed to: https://solv-03.phucvin.workers.dev/ (this uses a free plan of Cloudflare Workers) You can also run it yourself online: https://ift.tt/yJpSo9l Details: - Counter 01: simple counter work entirely at client. - Counter 02: 2 counters; increasing is client-side; reseting is a server action. - Counter 03: multiple counters; adding a new counter is a server action that also renders the component server-side (note that client handles the loading effect when button is clicked). More details in the repo. Thanks for reading, and please let me know if this is a good idea to continue. https://ift.tt/HRWTIEe November 8, 2025 at 11:34PM

Show HN: I built an HTTP client that perfectly mimics Chrome 142 https://ift.tt/drMpV5J

Show HN: I built an HTTP client that perfectly mimics Chrome 142 BoringSSL and nghttp2. Matches JA3N, JA4, and JA4_R fingerprints. Supports HTTP/2, async/await, and works with Cloudflare-protected sites. Not trying to compete with curl_cffi - just a learning project that turned into something functional. https://ift.tt/k7Zyr3H November 8, 2025 at 10:28PM

Show HN: Three Emojis, a daily word puzzle for language learners https://ift.tt/4bS9nep

Show HN: Three Emojis, a daily word puzzle for language learners I'm in the process of learning German and wanted to play a German version of the NYT’s Spelling Bee. It was awful, I was very bad at it, it was not fun. So I built my own version of Spelling Bee meant for people like me. Three Emojis is a daily word game designed for language learners. You get seven letters and a list of blanked-out words to find. When you discover shorter words, they automatically fill into longer ones—like a crossword—which turns out to be really useful for languages like German. Each word also gets three emojis assigned to it as a clue, created by GPT-5 to try and capture the word’s meaning (this works surprisingly well, most of the time). If you get stuck, you can get text/audio hints as well. It supports German and English, with new puzzles every day. You can flag missing words or suggest additions directly in the game. The word lists include slang, abbreviations, and chat-speak—because those are, in my opinion, a big part of real language learning too (just nothing vulgar, too obscure or obsolete). Every word you find comes with its definition and pronunciation audio. If you want infinite hints or (coming soon) archive access, you can upgrade to Pro. Feedback is very welcome, it's my first game and I'm certainly not a frontend guy. Happy spelling! https://ift.tt/n1M8Hbv November 8, 2025 at 02:36AM

Show HN: A Lightweight Kafka Alternative https://ift.tt/tGR5qrN

Show HN: A Lightweight Kafka Alternative https://ift.tt/pqFJQiI November 7, 2025 at 08:58PM

Show HN: DIY accessibility mouse helps people even with complete paralysis https://ift.tt/iqX3UNS

Show HN: DIY accessibility mouse helps people even with complete paralysis This is a DIY, open-source alternative to expensive solutions like the MouthPad, eye-trackers, or even complex systems like Neuralink. Everyone deserves access to assistive technology. https://ift.tt/S8Z5Hty November 7, 2025 at 01:31AM

Show HN: TabPFN-2.5 – SOTA foundation model for tabular data https://ift.tt/j4pzyk1

Show HN: TabPFN-2.5 – SOTA foundation model for tabular data I am excited to announce the release of TabPFN-2.5, our tabular foundation model that now scales to datasets of up to 50,000 samples and 2,000 features - a 5x increase from TabPFN v2, published in the Nature journal earlier this year. TabPFN-2.5 delivers state-of-the-art predictions in one forward pass without hyperparameter tuning across classification and regression tasks. What’s new in 2.5 : TabPFN-2.5 maintains the core approach of v2 - a pretrained transformer trained on more than hundred million synthetic datasets to perform in-context learning and output a predictive distribution for the test data. It natively supports missing values, cateogrical features, text and numerical features is robust to outliers and uninformative features. The major improvements: - 5x scale increase: Now handles 50,000 samples × 2,000 features (up from 10,000 × 500 in v2) - SOTA performance: TabPFN-2.5 outperforms tuned tree-based methods and matches the performance of a complex ensemble (AutoGluon 1.4), that itself includes TabPFN v2, tuned for 4 hours. Tuning the model improves performance, outperforming AutoGluon 1.4 for regression tasks. - Rebuilt API: New REST interface along with Python SDK with dedicated fit & predict endpoints, making deployment and integration more developer-friendly - A distillation engine that converts TabPFN-2.5 into a compact MLP or tree ensemble while preserving accuracy and offer low latency inference. There are still some limitations. The model is designed for datasets up to 50K samples. It can handle larger datasets but that hasn’t been our focus with TabPFN-2.5. The distillation engine is not yet available through the API but only through licenses (though we do show the performance in the model report). We’re actively working on removing these limitations and intend to release newer models focused on context reasoning, causal inference, graph networks, larger data and time-series. TabPFN-2.5 is available via API and a package on Hugging Face. Would love for you to try it and give us your feedback! Model report: https://ift.tt/ByQ9wve... Package: https://ift.tt/hLvXux0 Client: https://ift.tt/ajBFrl0 Docs: https://ift.tt/at0vXzn https://ift.tt/FiuDjhA November 7, 2025 at 01:26AM

Show HN: ShellAI – Local Terminal Assistance with SLM https://ift.tt/n7A0exf

Show HN: ShellAI – Local Terminal Assistance with SLM https://ift.tt/EliMZ3g November 6, 2025 at 11:27PM

Show HN: I was in a boring meeting so I made an encyclopedia https://ift.tt/cr7DWUq

Show HN: I was in a boring meeting so I made an encyclopedia https://ift.tt/hmiItrp November 5, 2025 at 09:58PM

Show HN: Strange Attractors – Visualized with Easylang https://ift.tt/a5hBEwf

Show HN: Strange Attractors – Visualized with Easylang This post about strange attractors isn't as cool as this recent one: https://ift.tt/YTconIf . My visualization runs with my programming language, which is interpreted by WASM code and uses JS and Canvas – without WebGL. https://ift.tt/Iu5xmSw November 5, 2025 at 09:25PM

Show HN: Agor → Figma for AI Coding (Open Source) https://ift.tt/0djz1p5

Show HN: Agor → Figma for AI Coding (Open Source) https://agor.live November 4, 2025 at 08:59PM

Show HN: Pion/rtwatch – Watch video in sync with friends, pause/seek on back end https://ift.tt/f43zgBs

Show HN: Pion/rtwatch – Watch video in sync with friends, pause/seek on back end https://ift.tt/9QxwGMy November 5, 2025 at 12:18AM

Show HN: Nallely – a modular reactive Python system for custom MIDI instruments https://ift.tt/V5Rx8NC

Show HN: Nallely – a modular reactive Python system for custom MIDI instruments Hi HN! I'm Vince. I built Nallely, a modular reactive Python framework for creating custom MIDI instruments by patching signal-processing modules together, like a modular synthesizer for controls systems. Nallely focuses on real-time, thread-isolated, reactive behavior, letting you experiment with emergent behaviors. Demo video: https://www.youtube.com/watch?v=rbMnKAdqAVI building a patch from scratch and hot-debugging a running instance near the end. Key features: * Visual patching interface for connecting reactive modules (neurons), * Extensible via Python API, WebSocket, and/or code generation, * Integrates any input source (MIDI, webcam, ...) to control synthesizers. # Yes, but why? Existing software/libraries that proposes MIDI manipulation are powerful but not friendly to live experimentation. They are low-level, hard to rewire on the fly, and often heavy for embedded or headless setups. I wanted a system that could also evolve dynamically where modules could be patched, hot-swapped, and debugged in real time. # Architecture The system is built around a reactive threading model with no shared data: each neuron lives in its own thread and communicates by sending messages through channels. No more CC,... , at the neuron level, everything is a signal (a simple int/float value through time). No global tick, each neuron works on its own time. Each neuron being reactive, they are sleeping the majority of the time. The system takes heavy inspiration from the "Systems as Living Things" philosophy and Smalltalk by treating each thread as a small living entity more than a processing unit. Here is how to code a simple Sample&Hold module: class SampleHold(VirtualDevice): input_cv = VirtualParameter(name="input", range=(0, 127)) trigger_cv = VirtualParameter(name="trigger", range=(0, 1), conversion_policy=">0") @on(trigger_cv, edge="rising") def hold_value(self, value, ctx): return self.input The control layer uses a small WebSocket protocol that the react-based web UI uses to control and introspect sessions. A WebSocket-bus neuron lets external application auto-register to it to send/receive signals: another neuron in the network can serve signals captured from any source. They're useful to distribute computation loads on different machines. # What have I learned so far A simple threading model can be powerful in a MIDI/music context: * you can stop/resume a thread, stopping a part of the processing chain seamlessly; * overflown neurons can mitigate the pressure without impacting the whole session; * if a thread crashes, it is paused to give you the ability to debug the instance, and resume it; * simple websockets have an acceptable throughput. I was expecting a system entirely based on Python threads to be really ineffective, but it's surprisingly reasonable. Empirically I see ~1-2 % CPU per thread. A 20 threads classical session (~45 patches) uses roughly 21% CPU and 45MB RAM on CPython 3.13 GIL. CPython 3.14 no-GIL shows similar CPU but ~65MB RAM. Feedback loops raise usage (~38 %). Interestingly, on CPython 3.13 the load spreads across multiple cores, I suppose that the threads are sleeping enough to release often the GIL. # Try it! You can grab a precompiled PyInstaller built binary in the latest github actions artifacts. Doc is linked in the README, and deep-dive posts are available here: https://dr-schlange.github.io/nallely-midi/posts . # I would love feedback * What could be improved to make it easier to get familiar with? * Are there blind spots or design choices that could be problematic long-term? * Although it's MIDI-oriented, the system is really signal-agnostic, any idea for non-audio use-case? (e.g. visuals, etc) https://ift.tt/1bgTazR November 4, 2025 at 11:20PM

Show HN: I built a Raspberry Pi webcam to train my dog (using Claude) https://ift.tt/aPBDUm6

Show HN: I built a Raspberry Pi webcam to train my dog (using Claude) Hey HN! I’m a Product Manager and made a DIY doggy cam (using Claude and a Raspberry Pi) to help train my dog with separation anxiety. I wrote up a blog post sharing my experience building this project with AI. https://ift.tt/hbKOJl5 November 3, 2025 at 07:04AM

Show HN: Give your coding agents the ability to message each other https://ift.tt/Chk9Tou

Show HN: Give your coding agents the ability to message each other I submitted this earlier but it didn’t get any traction. But it’s blowing up on Twitter, so I figured I would give it another shot here. The system is quick and easy to setup and works surprisingly well. And it’s not just a fun gimmick; it’s now a core part of my workflow. https://ift.tt/n3GST02 November 3, 2025 at 04:39AM

Show HN: Carrie, for what Calendly can't do https://ift.tt/x7VyrJH

Show HN: Carrie, for what Calendly can't do Hey everyone, Through my career, I've spent too many hours and too much mental load on busywork like scheduling and following up on people's availabilities. So, I built Carrie. You simply cc her into your emails, and she sorts out meeting times across time zones, finds what works best for everyone, confirms the meeting and sends the invite. She handles scenarios beyond what Calendly can handle and it’s been freeing me up from the back-and-forth of juggling different meeting requests. I’ve been testing this with a beta group of users and am now looking to expand the user pool (please feel free to join the waitlist if you're interested). Would also love feedback on whether this seems useful and what seems to be missing to make this part of your workflow. Thanks! https://getcarrie.com/ November 2, 2025 at 09:40PM

Show HN: I built a smart blocker after destroying my dopamine baseline https://ift.tt/QFj1hku

Show HN: I built a smart blocker after destroying my dopamine baseline I'm a solo dev. A few years ago, I got addicted to Reddit. Spent months in that loop. Being a programmer, I thought I'd be clever. Redirected Reddit's domain to nowhere in my DNS file. Worked great until I'd just... open the file and undo it 20 minutes later. So I made it irreversible. Locked the DNS file so it can't be edited unless I boot my Mac in safe mode. And if I do that, there's a script that instantly locks it again. Haven't used Reddit since last year. Problem solved, right? Wrong. I just replaced Reddit with Twitch and YouTube. Started keeping streams running in the background while I coded. This went on for almost a year. It killed my ability to focus. If you know about dopamine, you know your brain releases it when it wants you to repeat an activity. The constant background streams destroyed my dopamine baseline. When I tried to code without anything running one day, it felt genuinely weird. Hard problems that used to be interesting just felt like grinding. I tried blocking Twitch and YouTube the same way I blocked Reddit. But I actually need YouTube for learning. I watch programmers on Twitch I learn from. I couldn't just nuke them entirely. So I built something smarter. The first version was terrible. Blocked things it shouldn't, let through things it should've blocked. Really buggy and annoying. Then I added AI. I tell it what I'm working on, and it blocks anything unrelated to that task. This was the breakthrough. I need YouTube for tutorials, but I don't need 3-hour video essay rabbit holes. The extension knows the difference now. It reminds me in the moment. Not after I've already wasted an hour. Right when I'm about to click into the distraction, it stops me and makes me think: "Is this what I'm supposed to be doing right now?" The result: I actually enjoy hard problems again. Turns out I wasn't burned out, I'd just wrecked my brain's reward system. Then I had to market this thing, so I started using Twitter. And oh boy, Twitter is addicting. You post something and wait for the notification to light up. I had to install my own extension on my Twitter Chrome profile. It's wild how effective it is when something reminds you "you're here to market, not scroll" right as you're about to fall into the feed. It's still hard sometimes. Your brain will try to disable it. But having something that catches you in the moment before you lose an hour makes all the difference. It's a Chrome extension, currently at beta v1.0.43: https://ift.tt/SpOsXEU... It's free, no signup, no payment. Just install it. Fair warning: it's still in beta. There will be bugs. But it works well enough that I use it daily, and it's helped me get my focus back. Built this to fix my own problem. Figured other devs might be in the same boat. Question for HN: Anyone else dealt with this? The programming with streams thing destroyed my focus for almost a year before I realized what was happening. What worked for you? https://ift.tt/OVf8dQb November 3, 2025 at 12:47AM

Show HN: Just vibe coded a HN TV dashboard https://ift.tt/OhJStZ1

Show HN: Just vibe coded a HN TV dashboard https://ift.tt/6BAt1jV November 2, 2025 at 01:41AM

Show HN: Proxmox-GitOps: Container Automation Framework https://ift.tt/5h8bpZ1

Show HN: Proxmox-GitOps: Container Automation Framework By encapsulating infrastructure within an extensible monorepository - recursively resolved from Git submodules at runtime - Proxmox-GitOps provides a comprehensive Infrastructure-as-Code (IaC) abstraction for an entire, automated, container-based infrastructure. Core Concepts: - Recursive Self-management: Control plane seeds itself by pushing its monorepository onto a locally bootstrapped instance, triggering a pipeline that recursively provisions the control plane onto PVE. - Monorepository: Centralizes infrastructure as comprehensive IaC artifact (for mirroring, like the project itself on Github) using submodules for modular composition. - Single Source of Truth: Git represents the desired infrastructure state. - Loose coupling: Containers are decoupled from the control plane, enabling runtime replacement and independent operation. https://ift.tt/rOfpSk1 November 2, 2025 at 12:49AM

Show HN: Why write code if the LLM can just do the thing? (web app experiment) https://ift.tt/IFg0d4x

Show HN: Why write code if the LLM can just do the thing? (web app experiment) I spent a few hours last weekend testing whether AI can replace code by executing directly. Built a contact manager where every HTTP request goes to an LLM with three tools: database (SQLite), webResponse (HTML/JSON/JS), and updateMemory (feedback). No routes, no controllers, no business logic. The AI designs schemas on first request, generates UIs from paths alone, and evolves based on natural language feedback. It works—forms submit, data persists, APIs return JSON—but it's catastrophically slow (30-60s per request), absurdly expensive ($0.05/request), and has zero UI consistency between requests. The capability exists; performance is the problem. When inference gets 10x faster, maybe the question shifts from "how do we generate better code?" to "why generate code at all?" https://ift.tt/x73qSLG November 2, 2025 at 12:45AM

Show HN: Jod – Conversational observability with MCP, no more dashboard juggling https://ift.tt/ui1F8fG

Show HN: Jod – Conversational observability with MCP, no more dashboard juggling Hi, Prastik & Gaurab here. We’re building Jod( https://jodmcp.com ), a project that grew out of our own frustration with current troubleshooting and observability workflows. Jod lets you chat with your logs to debug issues, generate on-demand dashboards, analyze service health, and monitor with ease, all from a single chat window. You can ask things like: “Why did latency spike last night?” “Show me 5xx errors from the payments service.” “Create a time series graph showing error counts for the last 6 hours.” ..and Jod will pull, summarize, or even visualize the answers for you. Before Jod, we spent countless hours digging through CloudWatch and deployment logs, juggling 10+ dashboards just to trace one issue. It often took as much time as writing the actual code. During incidents, things got even worse, too much noise, endless context switching, and a lot of repetitive work. We figured we couldn’t be the only ones feeling that pain, so we decided to build something that could make the process a little easier. Right now, Jod connects to CloudWatch through an MCP server, which streams responses to the backend over SSE, and the client displays everything in a conversational interface. You can ask questions about your logs, request visualizations with the @Graph annotation, or dig deeper into errors and trends. We’ve actually debugged and fixed multiple issues in Jod’s own codebase using Jod itself. That said, it’s still early days, and there’s a lot we want to improve. On our short-term roadmap, we plan to: - Add support for metrics and traces, not just logs. - Expand to other providers like Azure and GCP. - Release a standalone MCP server so developers can plug it into their own AI clients. If any of this resonates with you, we’d love for you to try it out: https://jodmcp.com . It’s free to get started! We’d really appreciate your feedback, bug reports, and suggestions on this. Thank you. https://jodmcp.com November 1, 2025 at 06:23PM