2025年8月8日 的 Show HN

Yes, Python.

I can already hear the screams from the rafters telling me how terrible of a choice Python is - but in my case, I valued modularity, extensibility, hackability over raw performance. (It was also a challenge to myself to see how far I can get without referencing existing implementations)

Synchrotron processes nodes: simple Python classes with typed I/O and a render() method for processing. It can be as concise as 5 lines:

    class IncrementNode(Node):
        input: StreamInput
        output: StreamOutput

        def render(self, ctx):
            self.out.write(self.a.read(ctx) + 1)

Nodes can then be spawned and linked programmatically or in the graphical editor. Synchrotron handles the rest at runtime. Besides the web UI, you can also interact with the engine via Python, REST, DSL, or standalone TUI.

Currently you can build synths, FX chains, MIDI instruments, arpeggiators, controllers, or just mess about with sound :>

Editor: https://synchrotron.thatother.dev/ Source: https://github.com/ThatOtherAndrew/Synchrotron

It's still experimental (and my first ever shipped project), but I'd love feedback from people who tinker with audio/DSP/live coding. Docs are terrible currently, but that's my next big goal!

LLM from URL —— A free AI chat completion service directly from URL

Usage:

In the address bar of any web browser, type your question after https://818233.xyz/ and hit Enter to get the instant answer. You know the best part of this? Whitespace in the url is supported in most web browsers!

You can also use curl or Wget to retrieve the appended url by replacing any whitespace with a '+' character. If you need to have an actual '+' character in your question, just use '++'.

Example:

The url <https://818233.xyz/h> in any web browser will return the same answer as if you send "hi there" to an AI chatbot.

curl command: curl https://818233.xyz/hi+there

wget command: wget -qO- https://818233.xyz/hi+there

Limit: No chat history

Fair use policy: Abuse of the service will result in an IP ban.

Contact: For questions, suggestions or bug reports, feel free to drop me an email ([email protected]).

Three years ago, I posted here about hacking together a fast e-ink laptop from a T480 because I was tired of spending all day on LCDs. I liked e-ink’s comfort, but it was too slow for day-to-day use.

https://news.ycombinator.com/item?id=26245563

That post drew in people, which grew into a community experimenting with ways to make e-ink usable for everyday computing. That project later turned into a company and a multi-year project to make e-ink fast and open.

We built our own FPGA-based controller, Caster, and went through multiple iterations to push past e-ink’s usual limits, slow refresh, ghosting, and proprietary controllers.

Now, after three years, we’ve launched the Modos Paper Developer Kit and Monitor: the fastest open-hardware e-ink display, with 75 Hz refresh and sub-100 ms latency.

It works with 6" to 13.3" mono or color panels over HDMI or USB-C, supports multiple grayscale modes, and has a C API for low-level control.

The hardware, firmware, and schematics are on our GitHub.

https://github.com/Modos-Labs/Glider

Our goal is to make e-ink fast and open enough that anyone can build on it, for hacking, research, or daily use.

Thanks, HN, for being part of the journey.

A simple calculator that estimates how many concurrent requests your GPU can handle for a given LLM, with shareable results.

Agents and workflows as data instead of code.

Julep is a backend-as-a-service to build agents and workflows. Main features:

1. Experimentation Velocity:

With most agent frameworks, agents are code that need to be deployed. To iterate on the prompts or update the agent, you need to make changes to the code and deploy it to your server. This makes agents hard to iterate on quickly. But with agents and tasks declared as data, all that happens in the background is make an api call to update or make a new entry in the database. This increases the experimentation velocity.

2. Scalability and long-running background tasks:

Scaling is a hard software engineering problem and the solution is to define large scale systems to be distributed, stateful and scalable. Agents as data is a very helpful paradigm because you have to solve for this problem only once at the orchestrator level. Julep can handle 100s of thousands of executions paralely.

3. Updating an agent / User Personalization:

Tools such as create agent, update agent etc. can update the agent recursively. Let’s assume an agent is chatting with a user and mentions that they only like to communicate in a professional tone. The agent can call a tool to 'update the agent' and modify/update the instructions to include that the user only likes to communicate in a professional tone.

This was the reason to create a backend when the market was flooded with agent frameworks. In Julep, you can declare a workflow in YAML, and it can automatically scale Millions of users.

After years on HN, I've built SmartResearchAI to solve what PhDs struggle with: finding answers in their mountain of PDFs. Today we add *free GPT-5 access* to our AI assistant.

*Core Capability* Upload your research files (PDF/DOCX) → ask questions → get answers with *inline citations* pinpointing exact sources.

*GPT-5 Powered Features* 1. *Document Retrieval Mode* - "Find conflicting methodology claims in these 3 papers" → Answers with [Source 1, p.12], [Source 2, p.4] - Extracts data from tables/figures with source links

2. *Smart Synthesis* - "Summarize Author X's theory using only my uploaded chapters" - "Compare these 5 studies' conclusions about CRISPR risks"

3. *Drafting + Integrity Tools* - Write with auto-citations (APA/MLA/Chicago) - Built-in plagiarism/AI detection (98.3% accuracy)

*Why GPT-5 Excels Here* - 41% better at cross-document reasoning vs. GPT-4 (our benchmarks) - Handles technical jargon in STEM/humanities - Traces claims to your specific uploads - no hallucinations

*Try It* https://smartresearch-ai.com/?ref=HN

*HN: Challenge Us* - How can we improve source tracing for math-heavy papers? - Should model choice (GPT-5/Claude/Gemini) be task-automated? - What safeguards would make you trust this for thesis work?

I built Plonky - a browser-based physics puzzle game that runs entirely in your browser with no downloads required. Key features: 50+ levels with innovative ragdoll physics mechanics Responsive design works on desktop and mobile Completely ad-free and school/workplace friendly Unique wobbly character mechanics create unpredictable gameplay Intuitive controls with keyboard and touch support Tech stack: Built with Next.js, TypeScript, and Tailwind CSS. The game itself is embedded via iframe from a CDN. Why I built this: I wanted to create an accessible physics game that anyone could enjoy instantly without barriers. Most physics games require downloads or have complex setups, but Plonky brings the joy of ragdoll physics to everyone through a simple web browser. The game has already received positive feedback from thousands of players who appreciate the unique physics mechanics and creative puzzle design. Try it out and let me know what you think! https://plonky.top

Hey HN – I built a framework called Aegis to govern AI-assisted software development.

The core idea is that AI-generated code should follow the same rules as human code: versioned, validated, observable. Aegis enforces this through blueprint-based development, drift detection, and runtime compliance systems. It’s designed for teams using tools like Copilot, Kilo, or Lovable to build production systems with confidence.

This isn’t a library — it’s a way to architect AI-native engineering workflows.

Would love feedback, questions, and critiques. Especially curious if others are facing similar issues with AI output governance or system reliability in their workflows.

Happy to dive into internals or philosophy if there's interest.

If you’ve run InfluxDB at scale, you know the pain: Retention policies mean throwing away history, keeping everything means huge hardware & license costs.

We built ExyData Historian to fix that.

What it does?

- Automatically exports old InfluxDB 1.x/2.x data to compressed Parquet in S3 or MinIO - Keep recent data hot in InfluxDB, move the rest to cheap storage - Run fast SQL on archived data via Apache Arrow + DuckDB - Query it all through one interface and / API. No hot/cold boundary for the user

Why it matters - 70–80% lower storage costs - Historical queries that are as fast (or faster) than InfluxDB itself - No manual exports, no query rewrites, no downtime

Who’s using it right now? InfluxDB Enterprise Customers and Huge instances of OSS, telcos and logistics companies are trying this right now.

We help you to reduce your Enterprise licensing cost, cause you are going to shrink your InfluxDB cluster.

You keep your existing InfluxDB running, Historian works alongside it, moving history to cheap storage while giving you more analytics power.

We’d love feedback from anyone managing large InfluxDB deployments.

Hey HN! I built this Wordle-inspired web game over the past ~6 hours. It’s a quick project, but I think it’s fun to play and would love any feedback, especially on the game mechanics. Visually, it’s definitely a bit rough and could use some polish, but I wanted to share it early and see what people think!

Hi! My name is Andrii and I build an app that's pretty much a "Cursor for everything".

I've started working on a similar project in November 2024, but the idea was to make it look more like an extension for macOS Terminal app. As I often forget some bash command, like resetting last Git commit, and end up Googling how to do it, which is annoying.

Then I thought why not to make it work with other apps too, like a standalone popup, so I build Inset :)

The popup itself is inspired by the one that's in Cursor. And it allows you to switch between models, save presents for prompts you use often.

It has different modes - quick question and insert. The quick question mode helps you answer any question about the current app (you can toggle Camera icon to send screenshots of it to AI provider), basically if you're in Photoshop, you can ask it for step by step instructions to do something and it just works.

The Insert mode is useful as it inserts the output texts directly in the input field of your app. So like you can select some text in Notion, ask Inset to reformat it and it will do it directly in the app.

It works on any application or website. I tried to make it fast, lightweight and privacy-first. At the moment the app is on macOS only.

I have a ton of ideas how to make this app even better. I'd love to add an Agent mode in there too so that it would be able to move mouse cursor, type something, switch apps, etc. for you!

You can find the app at https://useinset.com It costs $10/mo, and there is a free trial available!

Any feedback is much appreciated!

Hi guys, ten months ago I left my investment banking, software engineering, job to build software products for mortgage originators (morfi.com). I mostly build white-label products that mortgage companies can deploy to their consumer-facing websites, or use in their back offices to improve operations.

I recently completed this AI Loan Estimate Analyzer product for a client. It uses AI to analyze a borrower's loan estimates or closing disclosures, and then uses an intelligent recommendation engine to suggest various ways you could improve your loan terms (saving you money).

Some months ago, another developer posted their version of an AI Loan Estimate Analyzer, but a lot of the comments were about privacy, security and fraud concerns. In part, it was concerning that they were collecting and analyzing consumers' loan estimates without being a licensed mortgage company.

In our case, we've partnered with reputable, licensed, mortgage companies (white-labeling our software), so borrowers can feel confident that their data isn't used in fraudulent ways. The analysis is done between the user and a licensed mortgage company, and the data is secure.

Can you guys try it out and offer feedback? If you happen to be currently a mortgage, it'll likely help you save money.

AI LE Analyzer: oxford.loan-estimate-analysis.morfi.com

In this case, I white-labeled the tool for my client "Oxford Home Lending" by using subdomains and a site configuration database, so Oxford's version of the AI LE Analyzer is styled to match their existing brand.

Oxford is a reputable mortgage lender based in Columbus, OH and they are a subsidiary of Lower.com which is an even larger digital mortgage lender that's based in Columbus, OH.

Thanks!

For context - this is the other AI LE Analyzer post: https://news.ycombinator.com/item?id=42149044

I’ve been a freelance data analyst/developer for a while and I’d regularly get hired to build custom data pipelines and dashboards in Google Sheets/Looker Studio. Custom data integrations came with a large upfront cost, and possible maintenance but clients told me they were reluctant to use the major data connector platforms as they were complicated and expensive over time, so I decide to fix this. I recently built SyncRange to be dead simple data connector to Google Sheets with a generous free plan and simple pricing.

So far I’ve built out the following connectors based on the needs of my clients and early users:

- Shopify - Facebook/Instagram ads - LinkedIn Ads - Google Ads - Google Analytics - Google Search Console

I plan on adding more connectors as needed by my clients/users, most of which are in the ecommerce/marketing space so I will focus my time there.

After years of building projects that I thought we interesting, but were not validated, I have really noticed the difference building something customers actually want.

Please let me know if you have any feedback.

Want to share my project with the community, this is the project I have been building to learn Elixir, while also scratching an itch that I've had for a long time. This is a little package manager written in elixir. Some of how it does things is similar to home-brew. It is also pretty early days.

Right now, I only develop and test on my Mac, but would like to ensure it supports other platforms in the future. It also currently works a little more like a ports tree. So, like macports. It doesn't installed pre-compiled binaries, but instead it follows a 'spell definition' to pull sources, build and then install the build artifacts.

What I am looking for when sharing this. First, if there is interest in an alternative system package manager. Second, and more important, code reviews, these would be much appreciated.

A few links:

https://spell-book.run

https://github.com/spellbookpm/spellbook

https://github.com/spellbookpm/StandardBookOfSpells

I built a mobile fax app from scratch and submitted it to the App Store in just 60 hours. It includes: - A React Native mobile app - Backend - Landing page - App Store application

My app was approved on the first try.

Send Fax Pro: https://apps.apple.com/tr/app/send-fax-pro-secure-faxing/id6...

Landing page: https://sendfax.pro

Also I added fax testing page at https://sendfax.pro/receive-online-free-fax

Thanks in advance for your feedback!

With all the new study modes and learning-targeted AI models, we have been building fiig specifically to reduce chatting and encourage actual source exploration and thinking.

We love feedback. Fiig has been built alongside higher-education students, professional researchers, and teachers. With fiig, you can find scholarly sources, cite the work, and our AI agent can highlight and annotate any found or uploaded PDF.

Other cool features: Tool Tabs: Each agent has a set of tools that can be used in chat while you work. Drag & Drop: Organize and contextualize your workflow by dropping elements straight into chat. @reference Anything: Use @ to reference notes and files directly in your prompts. Studio Interface: Open multiple PDFs, search for sources, and generate text in one unified platform.

LMK what you think <3

hi! i built an app that takes the pain out of invoicing so you can send them faster and worldwide without a headache. i've always found invoicing to be a waste of time, switching between templates, calculating taxes, tracking different currencies, and keeping files organized. so i made FiscalBud :)

the idea from tools like stripe inspired me, but for invoices. it lets you create, customize, and send professional invoices to clients anywhere in the world in just minutes. it supports 8 currencies, 77 languages (you can choose the output data language and ui language separately), and works in 248 countries, so you can bill confidently on a global scale.

it comes with smart templates, automatic tax/subtotal/total calculations, localized csv exports, and cloud storage to keep everything organized. (coming soon) you can automate recurring invoices, payment reminders, and follow-ups.

it's built to be secure and privacy-focused, with encryption and compliance baked in. you can even send invoices directly via email using your own smtp settings, with automatically signed pdfs.

i've got plenty of ideas for making it even better, like deeper automation and more integrations with other tools you already use (including Stripe which is on the roadmap).

any feedback is much appreciated! :)

Hello HN!

Build your next CLI in Rust with Clap and Deeb for data persistence!

I have made this starter repo that takes care of everything from simple commands, data persistence, logging, and environment variable setup!

Let me know what you think!

You can export as image, share annotations as URL or have a collaborative annotation session with others.

Source code available at: https://github.com/surfly/annotateweb.com

Hi all, I’m the maker of YooAI.

I built this because I’ve always loved trying out the latest AI models — but I kept hitting the same frustration: subscriptions everywhere. Sometimes I’d pay for a full month just to use a model for 2 days. Other times, I had to subscribe to multiple platforms just to try something for a single afternoon. It felt wasteful and limiting.

So I created YooAI — a platform that brings together cutting-edge models like Veo3, Kling, and more. Most features are free and don’t even require login. Some advanced models cost a small fee due to usage costs, but there’s no subscription, just pay-as-you-go.

My goal is to make top AI tools easier, cheaper, and more flexible to access. And this is just the beginning — more powerful models will be added soon.

Would love to hear your feedback and ideas. Thanks for checking it out!