[1]: https://byte.tsundoku.io/#198502-381 [2]: https://news.ycombinator.com/item?id=17683184
To truly understand how operating systems and network protocols work, I decided to combine two classic learning tools: the xv6 teaching OS and a from-scratch TCP/IP stack.
I'm excited to share the result: my own from-scratch TCP/IP networking stack running directly inside the xv6-riscv (https://github.com/pandax381/xv6-riscv-net) kernel.
The project uses a modern virtio-net driver, allowing it to run seamlessly in QEMU and communicate with the host machine.
Key features:
- From-Scratch Stack: The core is powered by microps (https://github.com/pandax381/microps), a TCP/IP stack I originally wrote to run in user-space as a personal project to learn the low-level details of networking.
- Kernel Integration: This project ports microps from user-space into the xv6-riscv kernel.
- Socket API: Implements standard system calls (socket, bind, accept, etc.) to enable network application development.
- User-level Tools: Comes with a simple ifconfig command, plus tcpecho and udpecho servers to demonstrate its capabilities.
This has been a fantastic learning experience. My goal was to demystify the magic behind network-aware operating systems by building the components myself.
I'd love to hear your feedback and answer any questions!
We built Smooth because existing browser agents were slow, expensive, and unreliable. Even simple tasks could take minutes and cost dollars in API credits.
We started as users of Browser Use, but the pain was obvious. So we built something better. Smooth is 5x faster, 7x cheaper, and more reliable. And along the way, we discovered two principles that make agents actually work.
(1) Think like the LLM (https://x.com/karpathy/status/1937902205765607626).
The most important thing is to put yourself in the shoes of the LLM. This is especially important when designing the context. How you present the problem to the LLM determines whether it succeeds or fails. Imagine playing chess with an LLM. You could represent the board in countless ways - image, markdown, JSON, etc. Which one you choose matters more than any other part of the system. Clean, intuitive context is everything. We call this LLM-Ex.
(2) Let them write code (https://arxiv.org/pdf/2401.07339)
Tool calling is limited. If you want agents that can handle complex logic and manipulate objects reliably, you need code. Coding offers a richer, more composable action space. Suddenly, designing for the agent feels more like designing for a human developer, which makes everything simpler. By applying these two principles religiously, we realized you don't need huge models to get reliable results. Small, efficient models can get you higher reliability while also getting human-speed navigation and a huge cost reduction.
How it works:
1. Extract: we look at the webpage and extract all relevant elements by looking at the rendered page.
2. Filter and Clean: then, we use some simple heuristics to clean up the webpage. If an element is not interactive, e.g. because a banner is covering it, we remove it.
3. Recursively separate sections: we use several heuristics to represent the webpage in a way that is both LLM-friendly and as similar as possible to how humans see it.
We packaged Smooth in an easy API with instant browser spin-up, custom proxies, persistent sessions, and auto-CAPTCHA solvers. Our goal is to give you this infrastructure so that you can focus on what's important: building great apps for your users.
Before we built this, Antonio was at Amazon, Luca was finishing a PhD at Oxford, and we've been obsessed with reliable AI agents for years. Now we know: if you want agents to work reliably, focus on the context.
Try it for free at https://zero.circlemind.co/developer
Docs are here: https://docs.circlemind.co
Demo video: https://youtu.be/18v65oORixQ
We'd love feedback :)
Our approach differs significantly from traditional SDK generators in that we use structured pattern matching queries to create and update the code. Other SDK generators use templates, which overwrite custom changes and produce code that looks machine generated.
We’ve mixed in LLM codegen by creating this workflow: Run deterministic codegen to establish the SDK structure. Let LLMs enhance specific components where adaptability adds value and include agent rules files that enforce consistency and correctness with type checking and integration tests against mock servers. The system will retain the LLM edits, while the rest of the SDK is automatically maintained by the deterministic generator (keeping it in sync with the API). LLMs can edit most the files (see python rules and typescript rules).
You can try it out from your terminal: Install: npm install -g @sideko/cli Login: sideko login Initialize: sideko sdk init Prompt: “Add a new function that…”
Check out the repo for more details: https://github.com/Sideko-Inc/sideko We’d love to hear your thoughts!
I'm a C++ algorithms engineer, and today I'd like to share my first full-stack web project: PrivyDrop.
This project was born from two ideas:
First, I wanted to solve a daily annoyance of mine: I needed a simple, AirDrop-like way to send text, links, or screenshots between my phone and PC. I tried many tools, but they either required me to sign up, had various limitations, or uploaded my data to their servers, which I was never comfortable with.
Second, this was a personal experiment. Last year, with all the talk about "AI replacing programmers," I got curious. As a developer with no web background, I wanted to see how long it would take to learn full-stack development from scratch (an area I'm really interested in) and build a complete application, using AI as my primary coding partner and mentor.
PrivyDrop is the result of that learning and experiment.
It's a peer-to-peer (P2P) file-sharing tool based on WebRTC. Its core principle is that *your data belongs only to you*: all files and text are transferred directly between your browser and the recipient's, fully end-to-end encrypted, without ever passing through an intermediate server.
*Here are some of its key features:*
* Completely free and open-source, with no ads. * No sign-up or installation required—just open your browser. * Direct P2P connection for privacy, security, and speed. * No limits on file size or type. * Support for entire folder transfers. * Built-in resumable transfers.
The whole process has been a fantastic learning journey. I'm sharing it now in the hope that it can solve the same pain point for others. I'm really eager to hear any feedback, ideas, or even harsh criticism from the HN community!
* *Live Demo:* https://www.privydrop.app * *GitHub Repo:* https://github.com/david-bai00/PrivyDrop
Thanks, everyone!
Howdy - I’m Kevin, co-founder of Firebender, and we built the first background coding agent in android studio / Jetbrains!
Why not just use Cursor background agents or OpenAI Codex?
Both of these require setting up a cloud container and cloning your existing developer environment, and maintaining it. Then when you want to iterate on changes as AI inevitably makes a mistake, you either throw away the work, or have to pull down the branch and clean it up. This feels really clunky. With Firebender, background agents run locally in a lightweight git worktree/IDE tab. This means when the agent is done, you can easily clean up the changes and run code with a few clicks.
Under the hood, the agent behaves similarly to claude code (didn’t want to reinvent the wheel), but also leverages all of the hooks into IntelliJ sdk like go-to-definition, find usages, auto-imports for accuracy, and it gives a cleaner visual UI for reviewing changes and merging them. You can use any frontier model like gpt-5/sonnet-4 as the base.
We’ve had to do quite a bit of reverse engineering of the IntelliJ codebase to cleanly set up and manage the isolated environment, and I think you’ll appreciate the simple UX of hitting cmd+enter to run a background agent anywhere.
get started docs: https://docs.firebender.com/get-started/background-agents
download the plugin: https://firebender.com
Would love to get your feedback to help us improve the tool for you! Thanks!
We needed to build this for inxm.ai, and realised this was the perfect time to give back to the community.
Enterprise MCP Bridge is Open Source and solves Auth, Multi User, and REST apis by wrapping your existing MCPs.
Why? Not every place has reliable internet, and many people don’t want their financial data stored on third-party servers. CashLedger works completely offline and keeps data private — stored locally on your device.
Features: - Add transactions quickly (text or voice input) - Spending insights with charts and analytics - Generate professional PDF statements - Customizable dashboard and transaction cards - Choose your flow: go straight to your account after login, or start with a customizable home screen
The UX twist: My first tester (my father) wanted to skip the dashboard entirely. I wanted one. So I added an option: users can choose their preferred flow.
Demo: https://cashflow-friend-pwa.vercel.app/
I’d love your feedback: - Is this useful for anyone beyond me? - What features would you expect from a privacy-first cash tracker? - Any pitfalls you see in making this offline-first?
Key features:
- Quick to use (~20 LOC to create a functioning linter/checker) - Works with different file formats (as in the description) - Supports multiple ignore/noqa directives (per-line ignores, range ignores, whole file, rules choice during runs) - Single check can span all elements of the file or even all files - Works directly with Python's AST - Tutorials and API reference to make users life easier
All in all seems to be quite functional for my private (yet) linters, so I decided to make it into a FOSS project.
Have fun, would love to hear your feedback and ideas regarding it!
What’s different • Host is strictly constrained to Yes/No/Unknown to keep the reasoning with the player. • 500+ curated puzzles across mystery/logic/psych/absurd categories; growing set. • Multilingual (English/Chinese to start). • Solo-friendly but also works for shared-screen or group play.
How it works • You read a short prompt (missing key info by design). • You ask questions; the host answers Yes/No/Unknown only. • You stop when you can state the full backstory. • Example starter (no spoiler): “A man orders turtle soup. After one sip, he pays, goes home, and shoots himself. Why?”
Tech notes • Next.js 15 (App Router), TypeScript, Tailwind, shadcn/ui. • State via Zustand; data via Drizzle ORM. • Prompting + guardrails to enforce the Yes/No/Unknown boundary and avoid accidental reveals. • Hosted on Vercel; anonymous play works—optional sign-in only to save progress.
What I’d love feedback on • Cases where the host leaks too much or answers inconsistently. • Puzzle quality: which ones feel fair vs. too obscure; which categories you want more of. • Mobile UX/latency reports. • Ideas for a “race mode” (fewest questions/least time) without making it feel grindy.
Roadmap • Race mode and public leaderboards. • Creator submission/review pipeline for new puzzles. • Better reasoning trace (question path visualization).
Happy to answer questions and share details about the constraint setup and evaluation harness.
Doing due diligence involves reading a lot of documents (10Ks, 10Qs, 8Ks, earnings call transcripts, etc.) and analyzing years of financial statements. You also need to research key executives - their biographies, compensation, tenure, etc. Analyzing this amount of information leaves room for missing key information.
The AI Agents follow the best practices for a thorough and comprehensive due diligence on stocks. You can also set the agents to repeat this on a recurring basis like every week or month. Great for monitoring a stock you are looking for an opportunity to buy or a stock you already own.
See full details on the blog post here: https://decodeinvesting.com/automated_due_diligence_and_stoc...
You can try out the agents here: https://agents.decodeinvesting.com/
My sloppy testing on Linux shows a slightly better result compared to ripgrep, but it varies because of software and hardware jitters, on macOS (sonoma) simdgrep sometimes performs worse.
More benchmarks are welcomed.
After some deep introspection, and thinking about the explosion of Lovable, Replit, Cursor, Claude Code vibe-coded apps, I thought about what's the newest newest and most dreadful pain points in the dev arena right now. And I came up with the scenario of debugging some non-obvious errors, where your AI of choice will reply "You're absolutely right! Let me fix that", but never nailing what's wrong in the codebase.
So I built Ubon for the last week, listing thoroughly all the pain points I have experienced myself as a software engineer (mostly front-end) for 15 years. Ubon catches the stuff that slips past linters - hardcoded API keys, broken links, missing alt attributes, insecure cookies. The kind of issues that only blow up in production.
And now I can use Ubon by adding it to my codebase ("npx ubon scan .", or simply telling Claude Code "install Ubon before commiting"), and it will give outputs that either a developer or an AI agent can read to pinpoint real issues, pinpointing the line and suggested fix.
It's open-source, free to use, MIT licensed, and I won't abandon it after 7 days, haha. My hope is that it can become part of the workflow for AI agents or as a complement to linters like ESlint.
It makes me happy to share that after some deep testing, it works pretty well. I have tried with dozens of buggy codebases, and also simulated faulty repos generated by Cursor, Windsurf, Lovable, etc. to use Ubon on top of them, and the results are very good.
Would love feedback on what other checks would be useful. And if there's enough demand, I am happy to give online demos to get traction of users to enjoy Ubon.
Works smoothly on Linux/macOS. On Windows, there’s still a libstdc++ linking issue with the exe, contributions or tips are welcome.
GitHub: https://github.com/allocata/sip
Features:
- Current conditions (temp, wind, humidity, pressure, precipitation)
- ASCII icons + color output
- Input by city or lat/long
- Metric/imperial units, JSON output, multi-lang (EN, RU, ES)
- Live mode with auto-refresh
- Works out-of-the-box with Open-Meteo (no API key), also supports OpenWeatherMap
Install via `cargo install rustormy` or grab a prebuilt binary from releases page Repo: https://github.com/Tairesh/rustormy
Would love feedback, feature ideas, or bug reports -- especially from CLI/TUI fans.
I built DocsOrb to solve a simple but stressful problem (and my own problem too since many years!): keeping track of important documents like passports, rental contracts, and insurance papers. Too often they're scattered across folders, emails, or piles at home... and you only realize it when you urgently need them.
DocsOrb helps you: > Scan documents with auto-crop and enhancements (mobile camera or file upload) > Organize them around life's "moments" (travel, housing, insurance, etc.) > Search quickly using Key Information > AI extracts Key Information so the most important details are always at your fingertips > Export or share in one tap > AI Bulk organize: load up multiple images from your Photos to automatically organize them as documents, put them in the right folders, extract Key Information and also suggest a recommended name and description.
Everything stays on your device by default, with optional cloud backup if you want it. Privacy-first, so you're always in control.
Tech-wise: it's built with Nuxt + Capacitor, Supabase for structured storage, and a custom scanning flow (to avoid pricey SDK lock-ins).
I'd love your feedback: > Does this flow make sense to you? > What's missing in how you manage important documents? > Any suggestions before I go full blast on Marketing?
I'm Tony, CEO at Cosmic (YC W19). We're excited to announce the launch of the Cosmic AI Platform, a revolutionary system that transforms how you create content, build, and deploy applications.
The Cosmic AI Platform reinvents what a CMS can be - it's now a complete AI-enabled application development and deployment platform. Build and deploy anything you can imagine using natural language, in minutes.
Key features:
- AI Content Studio: Generate complete content models and content from natural language descriptions
- AI-Powered Development: Build applications with Next.js, React, Astro, and Vue.js using AI
- Full Deployment Pipeline: GitHub integration, Vercel deployments, environment management, and domain services
The platform is available now on all Cosmic accounts. We've increased our free tier to 300k tokens (up from 10k) so you can try these new capabilities.
Log in to the Cosmic dashboard to try it out: https://app.cosmicjs.com/login and read more about the release here: https://www.cosmicjs.com/blog/introducing-the-cosmic-ai-plat... We'd love to get your feedback!
Tony
I built Titan Breach (https://platform.titanbreach.com) to solve the pain of juggling multiple tools for cybersecurity intelligence. It’s an AI-powered toolkit that unifies URL phishing detection, IP infrastructure mapping, typo-domain/DNS monitoring, ransomware insights, and website availability checks.
Why I built this: As a security researcher, I was frustrated by switching between disjointed tools to track emerging threats. I wanted a single platform that delivers fast, cohesive intel using AI.
What it does: - Detects phishing via URL scoring and social-engineering pattern analysis. - Maps infrastructure with port scans, fingerprinting, and visualizations. - Tracks typo-squatted domains and suspicious DNS registrations. - Monitors DNS for anomalies and takeover risks. - Aggregates ransomware intel from public leaks and threat group trends. - Verifies website availability with multi-technique checks.
Tech stack: Built with Next.js for server-side rendering, Python for backend logic, Elasticsearch for indexing, MongoDB for storage, and LLMs for analyzing patterns (e.g., phishing intent from URL structures).
Caveats: - It’s an MVP—some features are limited, but the free tier requires no signup. - Stable for now, but heavy traffic might stress it. I’m ready to scale if needed.
I’d love help with: - UX: Is the dashboard intuitive? Any clunky workflows? - Features: What needs more polish—phishing detection, visualizations, or something else? - Bugs: Notice any glitches or inconsistencies? Just let me know.
Thanks for checking it out.
Happy to answer questions or hear what you'd like to see next!
I built a small tool: [https://image2pdf.vu](https://image2pdf.vu)
It’s a simple image-to-PDF converter, designed to be: - fast (runs directly in the browser), - private (no files are stored on the server), - free.
The idea came up when I needed to merge a few photos into a single PDF and most solutions I found were slow, full of ads, or required sign-ups.
I’d love some feedback: - How do you find the simplicity of the interface? - What features would you expect (e.g. batch conversion, more image formats, compression)?
Thanks!
*What it is:* A self-hosted, open-source alternative to Resend that uses Amazon SES for actual email delivery while maintaining 100% API compatibility.
*Why it matters:* - 85% cost reduction (Amazon SES rates: $0.10/1k emails vs premium pricing) - True drop-in replacement - just set RESEND_BASE_URL env var - Auto-creates DNS records if you use Digital Ocean (optional) - Complete control over your email infrastructure
*Technical stack:* - Next.js 15 + TypeScript - Direct PostgreSQL (no external DB services) - AWS SES SDK v3 - Digital Ocean API integration - Docker ready
*The magic moment:* Add domain → auto-create 7 DNS records → sending emails in <60 seconds vs 15+ minutes of manual DNS setup.
I've been running it in production across multiple projects for months. MIT licensed, well-documented, includes test scripts.
*GitHub:* https://github.com/eibrahim/freeresend
*Medium post:* https://medium.com/@eibrahim/introducing-freeresend-the-open...
I've been trying to requalify for United Premier Gold status this year. If you're also on this path, you know how difficult it is get the necessary Premier Qualifying Points and Premier Qualifying Flights before the end of the year. Partner and Star Alliance airlines are particularly tricky, because the calculation depends on flight distance, fare class, airline, and other ticket attributes.
I built a free, simple web app to show how many PQP/PQF you need to qualify for the next tier. Importantly, you can create different scenarios and itineraries that result in different PQP/PQF amounts, so you can pick the scenario that gets you there the fastest.
This was built completely for fun and for personal use, but happy to take feedback and requests for improvement! Right now it's only built for United, but if there's enough interest I could expand this to American/Delta/Alaska.