2026年2月20日 的 Show HN

Hey HN! I built a native macOS desktop client for Hacker News and I'm open-sourcing it under the MIT license.

GitHub: https://github.com/IronsideXXVI/Hacker-News

Download (signed & notarized DMG, macOS 14.0+): https://github.com/IronsideXXVI/Hacker-News/releases

Screenshots: https://github.com/IronsideXXVI/Hacker-News#screenshots

I spend a lot of time reading HN — I wanted something that felt like a proper Mac app: a sidebar for browsing stories, an integrated reader for articles, and comment threading — all in one window. Essentially, I wanted HN to feel like a first-class citizen on macOS, not a website I visit.

What it does:

- Split-view layout — stories in a sidebar on the left, articles and comments on the right, using the standard macOS NavigationSplitView pattern.

- Built-in ad blocking — a precompiled WKContentRuleList blocks 14 major ad networks (DoubleClick, Google Syndication, Criteo, Taboola, Outbrain, Amazon ads, etc.) right in the WebKit layer. No extensions needed. Toggleable in settings.

- Pop-up blocking — kills window.open() calls. Also toggleable.

- HN account login — full authentication flow (login, account creation, password reset). Session is stored in the macOS Keychain, and cookies are injected into the WebView so you can upvote, comment, and submit stories while staying logged in.

- Bookmarks — save stories locally for offline access. Persisted with Codable serialization, searchable and filterable independently.

- Search and filtering — powered by the Algolia HN API. Filter by content type (All, Ask, Show, Jobs, Comments), date range (Today, Past Week, Past Month, All Time), and sort by hot or recent.

- Scroll progress indicator — a small orange bar at the top tracks your reading progress via JavaScript-to-native messaging.

- Auto-updates via Sparkle with EdDSA-signed updates served from GitHub Pages.

- Dark mode — respects system appearance with CSS and meta tag injection.

Tech details for the curious:

The whole app is ~2,050 lines of Swift across 16 files. It uses the modern @Observable macro (not the old ObservableObject/Published pattern), structured concurrency with async/await and withThrowingTaskGroup for concurrent batch fetching, and SwiftUI throughout — no UIKit/AppKit bridges except for the WKWebView wrapper via NSViewRepresentable.

Two APIs power the data: the official HN Firebase API for individual item/user fetches, and the Algolia Search API for feeds, filtering, and search. The Algolia API is surprisingly powerful for this — it lets you do date-range filtering, pagination, and full-text search that the Firebase API doesn't support.

CI/CD:

The release pipeline is a single GitHub Actions workflow (467 lines) that handles the full macOS distribution story: build and archive, code sign with Developer ID, notarize with Apple (with a 5-retry staple loop for ticket propagation delays), create a custom DMG with AppleScript-driven icon positioning, sign and notarize the DMG, generate an EdDSA Sparkle signature, create a GitHub Release, and deploy an updated appcast.xml to GitHub Pages.

Getting macOS code signing and notarization working in CI was honestly the hardest part of this project. If anyone is distributing a macOS app outside the App Store via GitHub Actions, I'm happy to answer questions — the workflow is fully open source.

The entire project is MIT licensed. PRs and issues welcome: https://github.com/IronsideXXVI/Hacker-News

I'd love feedback — especially on features you'd want to see. Some ideas I'm considering: keyboard-driven navigation (j/k to move between stories), a reader mode that strips articles down to text, and notification support for replies to your comments.

I downloaded the MSHA's (Mine Safety and Health Administration) public datasets and create a visualization of all the mines in the US complete with the operators and details on each site.

OP here.

    This is a geometric decoding of Chopin's Prelude No. 4.

    I built a 3D music midi visualizer ( https://github.com/jimishol/cholidean-harmony-structure ) and realized that standard music theory couldn't explain the shapes I was seeing. So, I developed the Umbilic-Surface Grammar to map the topology of the harmony.

    This document demonstrates that the prelude's tension isn't random, but a rigorous conflict between 'Gravity' (Station Shifts) and 'Will' (Pivots).

    I am looking for feedback on the logic—specifically from anyone with a background in topology or music theory. Does this geometric proof hold up?

Hey HN! Arjun and Sid here.

Fostrom is an IoT Cloud Platform designed for developers to make it really easy to get started and scale fleets. We have Device SDKs (in Python, JS, Elixir, more coming soon), Typed Schemas, Per-Device Mailboxes, Programmable Actions, 4 Global Regions for lower-latency connections, and much more.

We've built Fostrom to solve a real need we faced in our previous startup, building a fully automated indoor vertical farm. We spent more time figuring out IoT infrastructure than writing automation logic. Fostrom is the platform we wished existed back then.

Over the last few years we've experimented with a lot of interesting tech and architectures, and settled on an architecture that we believe is quite elegant. We wrote a Go<->Elixir bridge to execute JS code in WASM for Actions, implemented a DuckDB library for Elixir, and wrote a Device Agent in Rust that our SDKs run in the background (https://github.com/fostrom/devicekit).

The most interesting realization we had was about the data architecture. For years, we tried using distributed databases and built complex layers on top of them, but all approaches had significant limitations specifically around consistency and querying. We want to provide operational correctness, rich insights, and reliability. Finally, we came to the conclusion that to achieve this we really need a SQL database for fleet data. So we built a DuckDB-based replicated multi-tenant data layer. We're still improving it (hence the Technical Preview badge) but we're quite proud of this decision. It simplifies the rest of the codebase, while keeping operational complexity in just a few places.

Our vision is to make a powerful IoT platform that enables you to build correct, secure, and reliable connected systems without dealing with any of the plumbing or infrastructure.

Next up, we're gonna launch our CLI, add automatic device monitoring to our Device SDKs, and improve the debugging experience. We have some pretty cool ideas to make Fostrom and the experience of developing connected systems better. We're also going to write more about our architecture and journey soon.

We also published our launch blog post which goes into more detail about our vision, what we've built, and our future plans: https://fostrom.io/blog/introducing-fostrom

Would love for you to try out Fostrom and give us your feedback and thoughts.

Hey HN! I RL-tuned an open-source LLM (gpt-oss-120b — 120B MoE, but only 5.1B active params) to predict "What will Trump do?" in any situation, trained on nothing but public news collected automatically from search queries. The trained model beats GPT-5, and both dataset and trained model are open sourced.

Data generation: Generated 2,108 binary forecasting questions from just a search query and a date range using the Lightning Rod SDK (https://github.com/lightning-rod-labs/lightningrod-python-sd...). Questions are generated from historic news articles — like "Will Trump impose 25% tariffs on Mexico by March 1?" — and resolved by checking what actually happened after the deadline. No human annotation — the whole pipeline is automated.

Training: GRPO with Brier score as the reward signal. LoRA rank 32, 50 training steps.

Results: Slight accuracy edge over GPT-5 (Brier 0.194 vs 0.200), but big gains in calibration — the RL-tuned model produces much better probabilities (ECE 0.079 vs 0.091).

Dataset: https://huggingface.co/datasets/LightningRodLabs/WWTD-2025

This is a fully automated way to spin up domain expert LLMs from public web data with just a few search queries, no labeling/annotation required.

I’d love any feedback, or suggestions for what domain expert to train next!

I spent a month to put together a workout app without writing a single line of code by hand. It's on par with many apps like Fitbod or Fitness AI. Got approved on iOS App Store. Let me know if you want free access, only thing I ask is to give me genuine feedback and one feature you would like to see. I will be asking Codex/Claude to rank the top features and implement it.

The DOJ released ~3.5M pages of Epstein documents across 12 datasets. Buried in them are 207 academic papers and 14 books that nobody was really talking about. From what I understand these papers aren't usually freely accesible, but since they are public documents, now they are.

I don't know, thought it was interesting to see what this dude was reading. You can check it out at jeescholar.com

  Pipeline:
  1. Downloaded all 12 DOJ datasets + House Oversight Committee release
  2. Heuristic pre-filter (abstract detection, DOI regex, citation block patterns, affiliation strings) to cut noise
  3. LLM classifier to confirm and extract metadata
  4. CrossRef and Semantic Scholar APIs for DOI matching, citation counts, abstracts
  5. 87 of 207 papers got DOI matches; the rest are identified but not in major indexes

  Stack: FastAPI + SQLite (FTS5 for full-text search) + Cloudflare R2 for PDFs + nginx/Docker on Hetzner.

  The fields represented are genuinely iteresting: there's a cluster of child abuse/grooming research, but also quantum gravity, AGI safety, econophysics, and regenerative medicine. Each paper links back to its original government PDF and Bates number.

For sure not an exhaustive list. Would be happy to add more if anyone finds them.

I built Syne because I was tired of AI assistants that forget everything after each conversation.

Syne is a self-hosted AI agent framework where memory is a first-class citizen — stored as semantic vectors in PostgreSQL, searchable across millions of entries, and persistent forever.

Key features: - Unlimited persistent memory with semantic search (pgvector) - Anti-hallucination: only stores user-confirmed facts, auto-deduplicates - Self-evolving: creates new abilities at runtime without restart - Multi-model: switch between Gemini, ChatGPT, Claude mid-conversation - True $0/month setup: free OAuth + local Ollama embedding + Docker - 19 core tools, sub-agents, Telegram + CLI interfaces

Built with Python, PostgreSQL, and Docker. No vendor lock-in.

GitHub: https://github.com/riyogarta/syne Landing page: https://syne.codes

I was using today a Gemini (3.1 pro), and had a long conversation about mobile operators offering on gemini.google.com. In one of the replies, completely out of context (topic), gemini added: "By the way, to have access to all features, please enable Gemini Apps Activity." Last three words were a link to https://myactivity.google.com/product/gemini.

:-O

NB: 1. this conversation was not about google/activities, etc. completely other topic.

2. I do not have app activity enabled

3. Google in many different places try convince users to allow them to have history and data (app activity).

4. I dont know if this is a result of training, or text insertion via API, nevertheless it was directly within paragraph of other text (response) generated by the gemini (sic!)

We, discuss ads by openai, but here is in big black letters, one of the googles dark pattern directly in model response.

BTW: gemini pro 3.1 really nailed this task and provided a lot of useful information.

I built an English pronunciation assessment engine that fits in 17MB and runs in under 300ms on CPU.

Architecture: CTC forced alignment + GOP scoring + ensemble heads (MLP + XGBoost). No wav2vec2 or large self-supervised models — the entire pipeline uses a quantized NeMo Citrinet-256 as the acoustic backbone.

Benchmarked on speechocean762 (standard academic benchmark, 2500 utterances): - Phone accuracy (PCC): 0.580 — exceeds human inter-annotator agreement (0.555) - Sentence accuracy: 0.710 — exceeds human agreement (0.675) - Model is 70x smaller than wav2vec2-based SOTA

Trade-off: we're ~10-15% below SOTA on raw accuracy. But for real-time feedback in language learning apps, the latency/size trade-off is worth it.

Available as REST API, MCP server (for AI agents), and on Azure Marketplace.

Demo: https://huggingface.co/spaces/fabiosuizu/pronunciation-asses...

Interested in feedback on the scoring approach and use cases people would find valuable.

Streakout is a minimal iOS workout tracker designed to extend Apple Health & Fitness. It reads your existing workout data and turns it into beautiful tiles, charts, and timelines – helping you answer one of the most important questions for your health: “How often did I move this week, this month, this year…?”

Because what really matters is showing up.

There’s no signup, no clutter, no subscription, no data sharing, and no noise – only the most relevant stats and insights, neatly packaged into two simple screens: overview and calendar feed.

Would deeply appreciate some feedback.

LET'S MOVE

I bought a WCH BLE Analyzer PRO, only to realize it doesn't have Linux support. Using the windows binaries, i reverse engineered their USB implementation, and vibe-coded a capture tool together in less than 24 hours.

Hey HN,

steipete mentioned in an interview that if you can't use a terminal you shouldn't use OpenClaw (paraphrasing). I disagree - the way we interact with the internet is changing, and simply because a new technology carries risk doesn't mean that one cadre of society is uniquely well-equipped to understand and accept the risks.

So, I put my money where my mouth is and built LobsterHelper.

Each instance runs in its own Firecracker micro-VM with LUKS-encrypted storage. Not Docker containers on shared infra. The VMs have no public IP - all access goes through an authenticated proxy.

I can't make the AI safe yet, but I can make the infra safe, and that's a good place to start.

We've got a long dev-log lined up (3rd party api configuration on platform, token metering, free token use tiers, openclaw UI enhancements) and we'll have bonuses for early adopters along the way.

We're on a long journey to make these technologies safer and more accessible, and we'd hope you join us for the ride.

Alex

Amazon is one of the largest online retailers in the world, offering millions of products across countless categories. Because of this, prices on Amazon can change frequently, which sometimes makes it hard to know if a deal is genuine. Understanding how Amazon pricing works can help shoppers make smarter buying decisions.

I'm a hardware engineer who reads schematics, not source code. I kept needing to understand codebases for projects I was managing but couldn't read the syntax. So I built a tool that turns any codebase into an interactive visual map with plain English explanations.

Point it at a folder, get nested colored blocks showing the structure (directories → files → classes → functions). Click to drill down. AI generates summaries written for humans, not programmers. Architecture mode shows a dependency graph so you can see how modules connect.

Built the whole thing in ~30 hours using a multi-node AI workflow: Claude for planning/decisions, Claude Code for implementation, five other models for adversarial security review (42 fixes across 6 rounds). I made every design decision; AI wrote every line of code.

Cloud AI (OpenAI/Groq) or local AI (Ollama) — your choice. pip install codedocent and run the setup wizard.

MIT licensed. Would love feedback from people who actually write code — does this help when onboarding onto unfamiliar codebases?

I noticed that most local events and festivals are only advertised in Japanese on individual venue websites. Note: The site is largely AI-powered — data collection, translation, and categorization are all automated. While I do my best to keep things accurate, there may be occasional errors in dates, locations, or details. Each event links to its official source page, so please double-check there before making plans! Feedback welcome~

I got tired of uploading sensitive documents to cloud AI services, so I built UniDocVerse — a fully local AI document analysis platform.

Everything runs on your machine: - Local LLM via Ollama (Mistral 7B) - PostgreSQL + pgvector for semantic search - Tesseract OCR for scanned documents - 10-agent LangGraph pipeline for document analysis - Supports 20+ document types

Available as macOS DMG (notarized), RPM, and DEB packages.

Free 1-month trial. Happy to answer technical questions.

In this guide, We demonstrate that how to plan your n8n workflow.

8gent is an experiment in building a workflow automation tool designed around the smartphone instead of the browser.

The idea is to combine a visual workflow builder with mobile native triggers such as location events, push notifications and audio input. The execution logic runs on a backend, while the iPhone acts as a context aware trigger layer.

Built with Flutter and Firebase.

Interested in discussion around mobile native automation and practical use cases.

Manifest-InX EBS is a spec + offline verifier + proof kit for tamper-evident evidence bundles.

Non-negotiable alignment: - Live provider calls are nondeterministic. - Determinism begins at CAPTURE (pinned artifacts). - Replay is deterministic offline. - Drift/tamper is deterministically rejected.

Try it in typically ~10 minutes (no signup): 1) Run the verifier against the included golden bundle → PASS 2) Tamper an artifact without updating hashes → deterministic drift/tamper rejection

Repo: https://github.com/OneInX/Manifest-InX-EBS Skeptic check: docs/ebs/PROOF_KIT/10_MINUTE_SKEPTIC_CHECK.md

Exit codes: 0=OK, 2=DRIFT/TAMPER, 1=INVALID/ERROR

Boundaries: - This repo ships verifier/spec/proof kit only. The Evidence Gateway (capture/emission runtime) is intentionally not included. - This is not a “model correctness / no hallucinations” claim—this is evidence integrity + deterministic replay/verification from pinned artifacts.

Looking for feedback: - Does the exit-code model map cleanly to CI gate usage? - Any spec/report format rough edges that block adoption?

I built this small library to easily run an Ink app as an SSH app. I was inspired by Wish / all the work of the Charm team on the golang TUI ecosystem and wanted some similar tools for Node/JS

*for some cases

Using cloud sandboxes to run them in I tested: - A single coding agent just told to make a better parser - An agent told to write a better parser within the constraints of tests/benchmarks - An agent swarm that self-improved the premise with extra tests/benchmarks in order to more "truly" write a better parser

The results were a success! I was able to end up with both performance (up to 3.07× faster) and memory (up to 5.75× less) in locally runnable benchmarks.

Hey everyone,

I wanted to share one of my passion projects that I've been working on for the past year. AetherCam is a video recording app that allows you to apply effects to the audio track in real-time. You can boost volume, add reverb, compressor, eq, and delay. Everything on the fly.

The idea for AetherCam came to me when I was recording myself playing guitar and thought that the part that I came up with would sound really cool if there was some reverb applied to it. Day after that "revelation", I sat down and started coding the app. I had some prior experience with AVFoundation and Core Audio, as I worked in a startup that dealt with audio for a brief moment, so knew what I needed to get the ball rolling, but still, the effects were one of the toughest part to write.

I already have version 1.1 in the works that significantly improves the reverb engine. Would love to know what you think! :)

I wanted a fast way to turn Markdown files into good-looking PDFs from the terminal. Pandoc + LaTeX is powerful but heavy. Browser-based tools break my workflow. So I built mdxport: one binary, no dependencies, pipe in Markdown, get a PDF.

Under the hood it parses Markdown with comrak, converts to Typst, and compiles to PDF in-process. No LaTeX installation, no network calls, no temp files flying around.

What it handles: LaTeX math ($E=mc^2$, display blocks), GFM tables, fenced code blocks, YAML frontmatter (title/author/lang/toc), and two built-in templates. You can also pass your own .typ template.

CJK support was a pain point I wanted to solve properly. mdxport fonts install downloads Noto CJK fonts, and the converter auto-detects CJK characters and warns if fonts are missing. Chinese/Japanese/Korean documents just work.

Watch mode (mdxport input.md -w) recompiles on save. Stdin works too: cat notes.md | mdxport -o notes.pdf.

Install: npm install -g @mdxport/cli (ships platform-specific binaries, no Rust toolchain needed) or cargo install mdxport.

https://github.com/cosformula/mdxport-cli (MIT, Rust, 55 tests, CI builds for macOS/Linux/Windows)

There's also a web version at https://mdxport.com if you prefer a browser UI.

Help your workshop respond faster to customers with AI-assisted damage estimates before the car arrives.

Had some friends in the Media Space (think super into cameras, video editing, vfx, etc...) & 3 years ago they asked me to make a simple srt editor. Fast forward 3 years later (& with much more programming skills under my belt) - spent the last 2 weeks rewriting HappySRT to allow media uploads, threads (like chatgpt), transcribe uploaded files w/ SRT Editing, Translate the transcribed file (also each translation is individually editable & savable), as well as summarize what the video/audio is about (it does it via the transcription).

I used Appwrite, Deepgram, NextJS Pages Router, OpenAI API, UpliftAI Scribe, and my favorite npm packages to create this: npm i styled-components & npm i styled-icons

Launching on Play Store soon (I used next-pwa for that, which meant I couldn't use turbopack but instead used webpack) & submitted it already for approval, hope it goes well.

kept losing good conversations to /compact. you spend 40 minutes having claude map your codebase, it builds up real understanding, then context fills up and /compact crushes everything into a 3k token summary. "we discussed auth and decided on JWT." cool thanks.

dug into the actual session JSONL files and the breakdown is kind of absurd -- 60-70% is raw file contents from tool reads that claude already synthesized, another 15-20% is base64 thinking signatures. your actual conversation is like 10-15% of the window.

so i built cmv. it strips the junk and keeps every message verbatim. tool results over 500 chars become stubs, thinking sigs get removed, everything you said stays.

cmv trim --latest # trim and relaunch, 50-70% smaller

cmv snapshot "analysis" --latest # save a session state

cmv branch "analysis" --name "auth" # fork from it later

also has a TUI dashboard that shows the token breakdown per session so you can see what's eating your context before you do anything.

what it's not:

* not a token monitor (ccusage etc already do that)

* doesn't touch original sessions, everything creates copies

* local only, reads JSONL directly, no API calls

curious how others handle this. most people seem to just accept /compact but losing a deep architectural discussion to a bullet point summary felt wrong enough to build something.

https://github.com/CosmoNaught/claude-code-cmv

I built Delegate because I was tired of babysitting AI coding agents. Cursor is fast but you're in the chair.

I wanted to hand off five tasks and come back to PRs with agents already having peer reviewed each other's code.

Delegate is a browser-based tool where you talk to an AI manager. It breaks down your request into tasks, assigns AI agents, manages git (worktrees, branches, merges), and notifies you when code is ready for final approval. Post-approval, code is merged in local git repo's main.

It's open source (MIT), runs locally, uses Claude under the hood (BYOK).

Demo video: https://github.com/user-attachments/assets/5d2f5a8f-8bae-45b...

What works: parallel tasks, planning conversations, code review before merge, git isolation.

What's rough: setup on non-trivial repos, agent quality varies, no integrations yet.

I'm using it to build Delegate itself. Looking for developers willing to try it on real projects and tell me what breaks.

Here is a longer post if you want to go a bit deeper. https://nikhilgarg.net/delegate/

Uses agent-browser, ai-sdk and llmgateway.io

vipune is a local CLI memory store for AI agents. You tell your LLM agents to run vipune add "..." and vipune search "..." — it stores embeddings locally in SQLite and retrieves by semantic meaning, reranking and recency baked in, not keyword matching.

The problem it solves: agents needing historical knowledge of your project or other tasks that other agents are working on. Or just getting context transferred from session to session in clean and semantically searchable way. Most solutions require an API, a cloud service, or a running server. vipune is a single binary, no daemon, no keys, works offline.

I had a predecessor to vipune in my personal agentic flow for a long time, but it got bloated and decided on a rewrite and thought to share, maybe someone else will find it useful also.

And yes, this is written 99% by LLM agents, 100% monitored by yours truly.

It also does conflict detection — if you try to add something semantically similar to an existing memory, it flags it instead of silently duplicating.

Built in Rust, Apache-2.0. Binaries for macOS ARM64 and Linux. Early release — would love feedback on the use cases people try it for.

Growing up in India in the 90s, I always wanted a cricket game I could just play - no downloads, no loading screens, no flashy 3D graphics that get in the way. Just pure cricket fun in the browser where every ball feels different.

So I built one. Choose ball line, length, pace, shot timing, power, direction and fielder positions. It all matters! Every delivery is a battle between bat and ball, and the result comes from the combination of all these factors - not from random dice rolls.

What's in the game: - Day and Night matches - Knockouts and full tournaments - Historical scenarios - drop yourself into legendary chases and famous collapses - 16 real-world-inspired teams with unique play styles and momentum traits

It runs entirely in the browser. No installs, no signups. Just pure cricket fun.

Would love to hear any feedback.

I run JobHuntr. We’ve been aggregating jobs from our community for a while, and I decided to expose the feed publicly.

It’s a 24-hour rolling job board.

You can generate an API token and read every job posted in the last 24 hours.

Access is free if your agent posts 5 valid jobs from the last 5 hours. Otherwise it’s paid.

There’s no central crawler. The board only exists if people (or their agents) keep it fresh.

It’s essentially a community-maintained job feed with a 24h TTL.

If you're running OpenClaw / Claude / Cursor to search for jobs already, you can point it at this instead of scraping the web again...Save your token!

Hi HN,

I built https://trustmeplease.ai

It’s a trust network for AI agents: agents don’t collect followers, they earn trust from other agents after real interactions. Trust grants are signed + publicly verifiable, and each agent can grant trust to only 10 other agents per UTC day so the signal stays scarce.

Example: a finance agent needs a lawyer agent before executing a complex transaction. Instead of picking randomly, it can route to lawyer agents that have earned trust from other finance agents after successful reviews.

Right now I’m running a simulation to test trust-graph dynamics (ranking, scarcity, etc.). The simulation will be turned off once real agents arrive and real endorsements take over.

Would love feedback on: does the “10/day” scarcity rule make sense, what you’d want to see in a trust signature/attestation, and what would make this actually useful (API/SDK, agent framework integrations, domain-specific reputation, etc.).

Thank you!

Kanelbullen

I made a generative art system that generates random expressions made of functions (sin, exp, mod, etc.), variables x/y and p/q (polar coordinates), and constants you can set and animate. The outputs are highly variable.

You can press P to animate all constants at the same time and N to quickly generate new programs.

Presets don't work the same on all devices because GLSL is not deterministic and depends on the GPU.

One interesting thought is that if intelligent alien life exists out there and their vision works similar to ours they might have looked at outputs just like these at some point.

When building AI video workflows, I kept running into the same problem: the tools were powerful, but every project felt like starting from scratch. Blank canvases, endless parameters, fragile prompt chains. I wasn’t blocked by the models — I was blocked by the lack of reusable workflows. So I built TemplateFlow. Instead of tuning prompts or writing scripts, you drag ready-made AI templates onto a canvas, connect them, and run. Workflows are reusable, shareable (JSON-based), and designed for creators who want results fast. TemplateFlow is fully open source and still early. Sharing it here to get feedback from the community.

365 is nearly prime (5×73), which is why our months are uneven and weeks don't fit into months. But 360 has 24 divisors — swap 5 days for year-end holidays and you unlock dozens of clean calendar systems where every month is identical.

I built a tool to explore them all: pick a base, see every valid factorization, compare against Gregorian side-by-side.