每日 Show HN

Upvote0

2025年12月31日 的 Show HN

35 条
395

Use Claude Code to Query 600 GB Indexes over Hacker News, ArXiv, etc. #

exopriors.com faviconexopriors.com
141 评论7:47 AM在 HN 查看
Paste in my prompt to Claude Code with an embedded API key for accessing my public readonly SQL+vector database, and you have a state-of-the-art research tool over Hacker News, arXiv, LessWrong, and dozens of other high-quality public commons sites. Claude whips up the monster SQL queries that safely run on my machine, to answer your most nuanced questions.

There's also an Alerts functionality, where you can just ask Claude to submit a SQL query as an alert, and you'll be emailed when the ultra nuanced criteria is met (and the output changes). Like I want to know when somebody posts about "estrogen" in a psychoactive context, or enough biology metaphors when talking about building infrastructure.

Currently have embedded: posts: 1.4M / 4.6M comments: 15.6M / 38M That's with Voyage-3.5-lite. And you can do amazing compositional vector search, like search @FTX_crisis - (@guilt_tone - @guilt_topic) to find writing that was about the FTX crisis and distinctly without guilty tones, but that can mention "guilt".

I can embed everything and all the other sources for cheap, I just literally don't have the money.

132

Circuit Artist –Circuit simulator with propagation animation and rewind #

github.com favicongithub.com
19 评论4:12 PM在 HN 查看
Hello,

Circuit Artist is a game about drawing digital circuits as pixel art — like MS Paint, but pixels are wires and little triangles are NANDs. Hit play and the simulation runs in real time. It's fully white-box: every wire's state is visible.

I launched it last year with a unit-delay simulation that ran until convergence on each update. It worked, but I realized it had some fundamental problems:

(i) It hid what was actually happening. Players couldn't tell the difference between a ripple carry adder and a look-ahead adder — everything just resolved instantly.

(ii) Design bugs became "game bugs." When players had non-converging wires (like A = not(A)), the game couldn't explain why or where the error was.

(iii) Layout didn't matter. A NAND on one side of the map would instantly propagate to the opposite side, making spatial decisions meaningless. Unit delay probably works better for schematic-based simulators like Logisim, not a layout-focused game like this.

(iv) Propagation visualization couldn't be added on top of the unit-delay engine — there was no notion of direction or distance.

So I rebuilt the engine with variable-delay event-based simulation using an adaptation of Elmore delay. The delay for each wire depends on distance and fanout — longer wires are slower, higher fanout adds delay, branching wires are faster.

For the implementation: I compute Elmore delay over wire trees. Each pixel has resistance and capacitance, with NAND inputs as nodes to account for fanout. For non-tree wire graphs, I build a Dijkstra-based spanning tree (driver as root) and interpolate for edges outside the tree. I calculate the delay for each node and project it back to the image so every pixel has a distance.

This lets me animate propagation with a glow effect during simulation. The shader knows when each wire was triggered, the current time, and the Elmore distance of each pixel, so it resolves state per-pixel in real time.

On top of that:

- Time rewind: the simulation is now delta-based, so players can freeze and scrub backwards to debug. Super useful for cyclic circuits.

- Layers: since timing matters now, I added up to 3 layers. NANDs only go on the bottom layer, upper layers propagate faster (lower capacitance). Lets you route wires above dense NAND areas.

The game has a campaign for learning from scratch, though it's still incomplete and in progress. I'm working toward Steam Workshop support so people can create their own campaigns to teach different concepts or cover more advanced topics.

GitHub (full source): https://github.com/lets-all-be-stupid-forever/circuit-artist

Steam: https://store.steampowered.com/app/3139580/Circuit_Artist/

40

I built a universal clipboard that syncs realtime on multiple devices #

quickclip.space faviconquickclip.space
79 评论10:48 AM在 HN 查看
I’m Gopal, the guy behind QuickClip.

I built this out of pure frustration. Copying items between my phone and laptop was very painful. Sending notes and links on WhatsApp. Saving random drafts I’d forget about. It was total waste of time.

So I made QuickClip for myself first. A dead simple way to move text, links and images between devices instantly. No setup drama. No thinking. Fully encrypted

I use it every day. Shipping it publicly now to see if anyone else has the same problem.

Would honestly love to hear, how you move stuff between devices today, what’s broken or slow and what would make this actually useful for you

Happy to answer anything and take suggestions. Thanks for checking it out.

24

Shadowlight, a voice-driven murder mystery and heist inside Minecraft #

playshadowlight.com faviconplayshadowlight.com
3 评论11:43 AM在 HN 查看
We built Shadowlight because we wanted NPCs that feel alive in Minecraft, and because we think conversation can be a real skill loop, not just flavor text.

Shadowlight is a solo Minecraft experience where talking is the main mechanic. NPCs are LLM-powered with different personalities and communication styles, and you progress by earning trust through good listening, calibrated tone, and smart questioning.

Your goal is simple: interview the manor’s characters, gather clues, and figure out who is guilty.

At the end of each run, you get feedback on your conversations. The goal is to reflect back what worked and what didn’t in how you communicated, so you can adjust your approach on the next run.

Try it for free by joining our Minecraft server: mc.playshadowlight.com

Trailer video: https://www.youtube.com/watch?v=6867RsSxhSE&t=42s

Gameplay Video: https://www.loom.com/share/dff09df831de4a588d9fee9a055cfec1

We would love feedback on what resonate with you and what you would like to see in this kind of experience.

17

Magic CSV – Transform CSVs with plain English, no formulas #

magiccsv.app faviconmagiccsv.app
3 评论4:55 AM在 HN 查看
I built a tool that lets you describe CSV transformations in plain English and get clean data back. Upload a CSV, type something like "split address into street, city, state, zip" or "standardize all dates to YYYY-MM-DD", and it does it. I built this because I kept seeing people struggle with the same repetitive spreadsheet tasks—date formatting, splitting columns, cleaning up phone numbers—and reaching for Python or formulas when they just wanted the data fixed. It's free to try: https://magiccsv.app Would love feedback on what transformations people actually need. What's the messiest CSV problem you've dealt with?
6

FuseCells – 2,500 handcrafted levels logic puzzle game with leaderboard #

igodia.dev faviconigodia.dev
3 评论12:04 AM在 HN 查看
Hi everyone, I built FuseCells, a minimalistic logic puzzle game where every level is handcrafted (no procedural generation). It started as a personal challenge: design a clean rule-set and scale it to thousands of puzzles without losing difficulty balance.

What’s unique: - 2,500 handcrafted levels across multiple grid sizes - Deterministic logic, no guessing required - A rule system inspired by constraint-solving and path-finding concepts - Daily Challenge + leaderboards (I added these to give the game a “reason to come back” without turning it into a grind) - Built end-to-end as a solo dev project

A small data point: after adding Daily Challenge + leaderboards, the game quickly reached ~400 active players (depending on how you define active), and I’m trying to understand what makes them return: competition, habit, or simply puzzle pacing.

Technical notes (if you’re curious): - My internal tools validate solvability using a custom constraint solver - Difficulty is approximated via solver step count - Optimized to run smoothly on low-end devices - Designed for iOS, fully adapted for iPad

I’d love feedback from puzzle lovers, game designers, and anyone into handcrafted logic design - especially on difficulty curve, onboarding clarity, and whether daily + leaderboards add value or feel distracting.

4

Client-side encrypted AI detector using model ensembling #

veredictlabs.com faviconveredictlabs.com
0 评论2:18 AM在 HN 查看
Hi HN, I’m Oscar, a Year 8 student from Australia who enjoys messing around with computers and AI.

I recently built an AI detector to build on my skills in computer science. I entered the prototype into the Oliphant Science Awards which is a local science competition (writing a 4000-word report on the methodology) and ended up winning, so I decided to polish it into a real web service that the world can make use of.

I noticed that schools and businesses are rushing to use AI detectors, but most commercial tools require you to send full, plaintext documents to cloud servers. For researchers or IP-sensitive work, sending data unencrypted to a third party (who might use it for training) is a major privacy risk. Additionally, current commercial AI detectors aren't very transparent and are unable to tell you why they come to a conclusion. I wanted to build something that helps people make informed decisions based on as much information as possible, not tell them a simple percentage with nothing to back it up.

I built Veredict to be secure and private: 1. The browser generates a one-time AES-256 key and encrypts the text locally using the Web Crypto API. 2. This AES key is encrypted using the server’s RSA public key. 3. The encrypted payload is sent to my backend (Python/FastAPI running on Modal serverless GPUs). 4. We decrypt in memory only for the split-second of inference. The plaintext is never saved to a database.

The detection logic uses an ensemble of 4 models (including statistical analysis of perplexity/burstiness and a fine-tuned BERT model) to output a confidence score.

A note on the login: I know HN prefers demos without sign-ups. Since I am a student paying for the GPU compute out of a limited budget, I strictly require Google Auth to prevent bots from draining my credits. I hope you understand.

The app provides a free daily quota (250 words) so you can test the architecture.

Link: https://veredictlabs.com

I'd really appreciate feedback on basically anything regarding my project :)

4

A Prompt-Injection Firewall for AI Agents and RAG Pipelines #

3 评论9:01 PM在 HN 查看
We built SafeBrowse — an open-source prompt-injection firewall for AI systems.

Instead of relying on better prompts, SafeBrowse enforces a hard security boundary between untrusted web content and LLMs.

It blocks hidden instructions, policy violations, and poisoned data before the AI ever sees it.

Features: • Prompt injection detection (50+ patterns) • Policy engine (login/payment blocking) • Fail-closed by design • Audit logs & request IDs • Python SDK (sync + async) • RAG sanitization

PyPI: pip install safebrowse

Looking for feedback from AI infra, security, and agent builders.

3

Running Analytics on 22 GB of Hacker News in SQLite #

philippdubach.com faviconphilippdubach.com
0 评论3:00 PM在 HN 查看
First of all please go check out the amazing project this analysis is based on: https://news.ycombinator.com/item?id=46435308

For my paper [1] I analyzed HN attention dynamics using 72k temporal snapshots from December 2025 to model decay curves, preferential attachment, and engagement survival. Today I’ve cross-referenced these findings with the recent 22GB HackerBook SQLite export to validate my early-engagement prediction models on this 22GB historical subset.

[1] Preprint available on SSRN: https://dx.doi.org/10.2139/ssrn.5910263

3

I built an AI tool to automate property tax appeals for $29 #

appealpropertytaxonline.com faviconappealpropertytaxonline.com
2 评论9:28 PM在 HN 查看
Hi HN, I built this tool to help homeowners generate official appeal letters without hiring expensive lawyers or spending hours on research. It analyzes your property data to write personalized arguments and offers a certified mail option via Lob. It supports 3,000+ US counties. Would love to hear your feedback on the workflow or the pricing model.
3

Perfetto2LLM - A tool to pass system traces to an LLM #

perfetto-to-llm.vercel.app faviconperfetto-to-llm.vercel.app
0 评论5:38 AM在 HN 查看
I work with perfetto traces for ML/GPU optimization a lot. Copy pasting trace info to LLM is not easy. As usually the traces are gzip compressed, have very large file sizes. Also there is no way to select a certain section and sending it to LLM easily (can maybe write a query to do this but not ergonomic). So vibe coded this tool to quickly select the kernels/slices/threads I want to ask about and one click to get a text/json/markdown in my clipboard to paste for LLMs.

I think an MCP server for this might also be useful (tried a few but all miss certain things i wanted so chose to just build this quickly)

3

Fleet / Event manager for Star Citizen MMO #

0 评论7:18 PM在 HN 查看
I built an open-source org management platform for Star Citizen, a space MMO where player orgs can have 50K+ members managing fleets worth millions.

https://scorg.org

The problem: SC's official tools won't launch until 2026, but players need to coordinate now - track 100+ ship fleets, schedule ops across timezones, manage alliances, and monitor voice activity during battles.

Interesting challenges solved:

1. Multi-org data isolation - Users join multiple orgs, so every query needs scoping.

2. Canvas + Firebase Storage CORS - Couldn't export fleet layouts as PNG. Solution: fetch images as blobs, convert to base64 data URLs, then draw to canvas. No CORS config needed.

3. Discord bot - Built 4 microservices (VoiceActivityTracker, EventNotifier, ChannelManager, RoleSync) sharing Firebase state. Auto-creates channels for ops, cleans up when done.

Features: role-based access, event calendar with RSVP, LFG matchmaking, drag-and-drop fleet builder, economy tools, alliance system, analytics dashboard, mobile-responsive.

~15 pages, fully functional. Custom military-inspired UI (monospace, gold accents).

3

qq-pwn – A browser-based idler game #

qq-pwn.com faviconqq-pwn.com
0 评论9:41 PM在 HN 查看
Yes, the URL is qq-pwn. No, it's not malware.

It is supposed to be a browser-based incremental roguelike game.

The Twist: The code is 100% AI-generated. I am running this as an experiment to see if an LLM can handle game creation and balancing without human intervention (apart from guidance).

Human feedback on the progression is appreciated!

2

Eimi – Turn any information into daily cards #

eimi.tech faviconeimi.tech
0 评论3:56 PM在 HN 查看
Hi HN,

I’m the developer behind Eimi (https://eimi.tech/).

The Problem: I found myself constantly distracted. I wanted to learn new words and keep up with specific news, but every time I opened an app, I ended up doomscrolling for 20 minutes. The signal-to-noise ratio was too low.

The Solution: I built Eimi to "calm down" my consumption. It’s a card-based app that delivers content depends on your personal needs. Instead of an infinite feed, you get curated highly customizable cards.

Use Cases:

Language Learning: Instead of flashcards without context, it generates cards with usage examples for words I'm trying to learn.

News/Market Watch: I hook up my interests (e.g., tech news, specific stocks), and it generates a summary card. No clickbait, just the gist.

It's currently in Public Beta. I’d love to hear your thoughts on the UX and what other "card types" you’d find useful.

Thanks!

2

A web-based lighting controller built because my old became a brick #

github.com favicongithub.com
0 评论4:46 PM在 HN 查看
I’m a student and I built this because my old lightning controller (DMX) became a brick after the vendor’s control software was deprecated in 2025.

My focus was entirely on developing a robust backend architecture to guarantee maximum performance. Everything is released under GPLv3.

The current frontend is just a "vibecoded" dashboard made with plain HTML and JavaScript to keep rendering latency as low as possible. In earlier versions Svelte was used. Svelte added too much complexity for an initial mvp.

Video: https://streamable.com/95kd4r Repo: https://github.com/Arian-Ott/hyperion

Technical Details: The system uses a distributed architecture where a FastAPI server manages the state in a Redis. State changes are pushed via WebSockets to Raspberry Pi gateways, which then independently maintain the constant 44Hz binary stream to the lights. This "push model" saves massive amounts of bandwidth and ensures low latency. In a stress test, I processed 10 universes (5,120 channels) at 44Hz with zero packet loss (simulated). An OTP-based pairing makes the setup extremely simple (plug-and-play).

I’m looking forward to your feedback on the architecture and the Redis approach!

Happy New Year!

1

DBcooper – Open-source database client for macOS built with Tauri #

github.com favicongithub.com
0 评论11:05 AM在 HN 查看
Hey HN! I built DBcooper, a native macOS database client using Tauri (Rust) + React.

Supports PostgreSQL, SQLite, Redis, and ClickHouse in one app.

Features: - Schema visualizer with interactive ER diagrams - AI-powered SQL generation (bring your own OpenAI key) - Command palette for keyboard-first navigation - SSH tunnel support for remote connections - Fast and lightweight (~15MB) thanks to Tauri

This started as a side project because I wanted something simpler than the existing options. Still in early development, but it's usable for daily work.

Built with: Rust + Tauri v2, React + TypeScript, shadcn/ui

Would love feedback from fellow developers. What features would you want in a database client?

1

Chinese Learning Site – I made a free online version of the HSK books #

learnchinese.ai faviconlearnchinese.ai
0 评论9:04 PM在 HN 查看
I made a website for learning Chinese. There are many already, but I think there is always room for new entries into the Chinese language learning ecosystem.

The Site: - Course is an online version of the HSK course books (HSK 1-2 done so far, 3-4 coming soon) including all the vocabulary, texts, and all the content/activities from the coursebook and workbooks. - SRS Flashcards - SRS Sentence speaking - Character writing tool (Stroke data is from MakeMeAHanzi) - Dictionary is from the MBDG Chinese Dictionary - I build it using Rails and StimulusJS (with AI help) - Uses a lot of TurboFrames/TurboStreams to make it interactive - Goal is to be mobile friendly and responsive. - I got AdSense approval so it will be a completely free ad supported website.

AI: - I use AI to build the site, transcribe the course, enhance the images, transcribe speech, generate audio, and for some minor supplemental content and formatting. - 99%+ of the course and content is not AI generated and is from the HSK courses. - I am exploring additional ways to use AI, but I am very cautious about the quality and accuracy of AI generated content.

I would appreciate any feedback about the site, whether you are learning Chinese or not.

1

Real-Time Interview Assistant #

voicemeetai.com faviconvoicemeetai.com
0 评论11:02 AM在 HN 查看
Hi HN! I built VoiceMeetAI, a Browser Extension for interviews and meetings.

The idea is simple: when someone asks you a question, you can record the audio (tab or mic), it gets transcribed, and the extension generates an answer in real time.

1

A small AI tool I built to speed up outfit changes in product photos #

aiclotheschanger.net faviconaiclotheschanger.net
0 评论11:00 AM在 HN 查看
I'm an Amazon seller, not a designer.

One thing that kept slowing me down was making small, repetitive changes to product photos—mainly swapping outfits or variants. Photoshop can do it, but for this kind of work it felt heavy. A simple change often turned into a lot of clicking, masking, and exporting.

Like most people, I'd already tried using ChatGPT and other image tools for this, but they didn't fit well into a repeatable product photo workflow.

I started experimenting with image models on the side and built a small internal tool to handle these edits more directly. After some trial and error, the results became consistent enough that I actually started using it instead of Photoshop for certain tasks.

I cleaned it up a bit and made it public. It's intentionally narrow and not trying to do everything—just focused on speed and reducing friction in a workflow I deal with almost every day.

Link: https://aiclotheschanger.net

Happy to hear feedback, especially from people who spend a lot of time dealing with product images.

1

S3Broker – CF Worker library to protect your S3 storage from ransomware #

github.com favicongithub.com
0 评论5:39 AM在 HN 查看
Hi HN Community! I use S3 providers (e.g. Cloudflare R2) along with software like rclone and Mountain Duck for my personal media storage. Some S3 providers do not include built-in versioning / deletion protection. I find my data vulnerable to accidental deletion or ransomware attacks in case my device is compromised.

To tackle this issue, I built a "middleware" on Cloudflare Workers to proxy S3 requests with configurable guardrail policies. The client uses an alternative secret key to connect to this middleware, and the middleware would reject with 403 if it determines the S3 request is dangerous. For example, I could configure that all files in `/important/.*` path older than 1 week could not be deleted or overwritten. Doing so, I could restrict ransomware's abilities on the API level.

The code has been packed into a library and published to npm to be used on CF workers. This is an active work in progress, and I have already deployed it for my personal use (it works). Let me know what you think!