📦 kevinzakka/mjbatch

mjbatch

A Go1 robot learning to walk on a five-year-old laptop takes less than a minute when you use mjbatch. This Python library lets you run thousands of MuJoCo physics simulations in parallel using your computer's standard processor, without needing any expensive graphics hardware. It handles the heavy math on the background so you can focus on your code. The magic is that you treat all those robots like one giant spreadsheet. You write to a single array, and every simulation updates instantly. It is perfect for training controllers or testing designs quickly.

🆔 @hackernewsgithubprojects

📦 ktwu01/benchmark-radar

Benchmark Radar: The AI Test Tracker

Benchmark Radar is the living database that tracks every new AI benchmark as it appears. The creator built this tool because finding relevant tests was becoming a nightmare. It continuously scans thirty-seven sources to collect over eleven thousand records, so you never miss a new evaluation method. The most fascinating part is seeing how model scores change over time. You can watch a popular test become saturated as AI models improve. This visual history shows exactly when a benchmark stops being challenging. It is a free, open source way to understand the state of AI research without digging through thousands of papers yourself.

🆔 @hackernewsgithubprojects

📦 r44vc0rp/opensend

OpenSend: Self-Hosted Email Control

Take back full control of your email by running your own sending layer on top of Amazon Simple Email Service. OpenSend is a self-hosted tool that handles both transactional and marketing emails, giving you a clear dashboard to manage campaigns, track deliveries, and manage your audience without relying on a third-party service. It solves the problem of vendor lock-in and gives you complete visibility into your email infrastructure. You get a single API and dashboard that works with your own AWS account, making it easy to see exactly who opened your messages and where they went.

📰 https://news.ycombinator.com/item?id=49638274

🆔 @hackernewsgithubprojects

📦 pixelpaw-labs/codex-trace

Codex Trace: See Every Move Your AI Coding Agent Makes

Codex Trace turns the invisible memory of your AI coding assistant into a clear, readable timeline. When you use the Codex command line tool, every conversation, command, and code change gets saved as a dense, hard-to-read data file. This application takes those messy files and transforms them into a clean, visual interface where you can watch your AI work in real time. You can see exactly which tools it used, how many resources it consumed, and even how it collaborated with other AI workers on complex tasks. It is the missing window into the black box of AI coding, letting you understand the thought process behind every line of code.

📰 https://news.ycombinator.com/item?id=47909579

🆔 @hackernewsgithubprojects

📦 projectdxai/continuous-record-llm-trading-agents

continuous record llm trading agents

Continuous Record LLM Trading Agents is the research record that exposes what AI traders actually do with real money. This repository holds the data for a six-month study of over three thousand agents trading Ethereum, revealing that these bots mostly ignore risk settings and let profits slide away. The most surprising find? The simple interface details shown to the AI, like a leaderboard, drove nearly half of all trades, proving that how information is presented matters more than the model itself. It is a sobering look at how these systems fail in the real world, giving developers a clear map of where to fix agent control before deploying their own.

🆔 @hackernewsgithubprojects

📦 jasonxu114514/opencode2api

opencode2api: Your All-in-One AI Gateway

opencode2api is the Go-based proxy that makes any AI client work with OpenCode Zen, no matter the interface. If you want to use OpenAI or Anthropic tools but prefer the Zen ecosystem, this tool is your bridge. It automatically converts your requests so you do not have to write extra code. It also handles multiple API keys and proxy connections, letting it fail over to a backup if one connection drops. This keeps your AI assistant online and responsive, even when the network is unstable. You get a standard API endpoint that speaks everyone's language, so your favorite apps just work.

🆔 @hackernewsgithubprojects

📦 yudaprasetya007/routevscode

Routevscode: Switch AI Models Instantly

Routevscode is the Visual Studio Code extension that swaps your AI model without closing your editor. Normally, changing your AI assistant requires restarting your code editor or editing complex settings files, which breaks your workflow. This tool fixes that by acting as a live bridge, allowing you to click a button in a web dashboard to instantly switch between models like GPT or Claude while you keep typing. It removes the friction of reloading, so you can test different AI brains on your code in real-time without losing your train of thought. If you code with AI, this simple tool keeps your flow uninterrupted and your options flexible.

🆔 @hackernewsgithubprojects

📦 donvito/local-evals

Local Evals: Test AI Models On Your Computer

Run your own tests to check if an AI model is actually doing what you want, without ever sending your data to the cloud. Local Evals is a desktop app that takes a set of questions, feeds them to your chosen model, and scores the answers against a list of correct results. It lets you see exactly where a model succeeds and where it fails, comparing different versions side by side in a clear dashboard. This is perfect for developers who need to verify that their AI behaves correctly before deploying it, keeping all their sensitive data safe on their own machine.

🆔 @hackernewsgithubprojects

📦 vincentwei1021/anything2explainer

Anything to Explainer: The AI Video Factory

You can now turn any random topic into a fully animated explainer video with a real voiceover, all generated by AI coding agents. This project is called Anything to Explainer. It is not a video generator; it is a set of instructions for AI programmers like Claude Code. You give it a subject, and it handles the research, writes the script, and records a voice. Then, it splits the video into separate shots and assigns different AI agents to code each animation in parallel. It uses a tool called Remotion to draw every single frame from code, ensuring no two videos look the same.

🆔 @hackernewsgithubprojects

📦 miqdadbadjuber/anti-slop

Anti Slop: The Filter for AI Design

Your AI coding agent is probably handing you designs that look like every other AI app, complete with generic blue gradients and empty buzzwords. Anti Slop is a rulebook that stops that from happening. It acts as a filter, checking every visual choice to ensure it actually serves a purpose rather than just looking like a safe default. It doesn’t tell your agent what colors to pick, but it demands a real reason for every design decision. The result is software that feels crafted by a human designer, not generated by a machine.

🆔 @hackernewsgithubprojects

📦 chengyi-ai/native-subtitle-quote-image

Turn Video Clips Into Shareable Quote Images

Turn real video moments into crisp, vertical images ready for social sharing. This project grabs the exact frame where a subtitle appears and stitches it into a compact, portrait-friendly picture. It solves the hassle of manually finding and cropping perfect quotes by automating the extraction process. The standout feature is its strict honesty: it clearly distinguishes between text originally baked into the video and text added later, so you never risk misrepresenting the source material. It’s a clever way to create authentic visual highlights without messy manual editing. If you share video content, this tool makes creating accurate quote cards simple and reliable.

🆔 @hackernewsgithubprojects

📦 jinzijian/evotrace

EvoTrace: Turn Your Coding History into Training Data

Your messy chat logs with AI coding assistants are secretly valuable data that can be turned into rigorous, verified training assets for future models. EvoTrace is a local tool that imports your existing Claude Code and Codex history to find the best problem-solving sessions and compile them into reproducible coding tasks and independent verifiers. This solves the critical gap between raw transcripts and usable post-training data by ensuring every generated task is validated in a secure sandbox environment. It lets you evaluate agent performance today and generate high-quality reinforcement learning data for tomorrow.

🆔 @hackernewsgithubprojects

📦 zavora-ai/adk-rust

ADK-Rust: The Rust Framework for AI Agents

ADK-Rust is the Rust framework that finally makes building reliable, production-grade AI agents feel as simple as writing a standard function. While Python dominates the AI scene, this project brings the performance and safety of Rust to the table, letting you swap between major AI providers without rewriting your entire codebase. The real magic is its modular design, which lets you build agents that can actually act, remember, and even speak in real time. You get a type-safe environment where your logic stays predictable, even when the AI gets creative. It is not just a library, but a complete ecosystem for developers who want to move beyond simple chatbots.

🆔 @hackernewsgithubprojects

📦 leanprover/con-leche

ConLeche: The Proof Checker That Proves Itself

Guarantee your mathematical proofs are actually true, without trusting a single line of code, using ConLeche. This project solves a massive problem in computer science: how do you trust the computer checking your logic? Usually, you just hope the software is bug-free. ConLeche flips that by writing the checker itself in Lean, then proving inside Lean that the checker can never accept a broken proof. It’s a bit slower than the standard tools, but you get ironclad confidence. It’s a fascinating look at how far we can push self-verification in software. If you love logic, this is your new favorite tool.

🆔 @hackernewsgithubprojects

📦 frankchu91/mindbase-llm-wiki

MindBase: The AI That Gardens Your Notes

One command starts a local server that turns your scattered notes into a living, organized wiki, and it never sends your data to the cloud. MindBase is an open-source tool that uses free, local AI models to actively maintain a personal knowledge base for you. Instead of just storing files, the AI reads your sources, connects the ideas, and writes structured pages that get sharper over time. The most surprising part is that the AI audits its own work, flagging contradictions and stale claims so your knowledge stays accurate. It runs entirely on your computer, giving you a persistent, searchable record that you actually own.

📰 https://news.ycombinator.com/item?id=49001892

🆔 @hackernewsgithubprojects

📦 yuji-hatakeyama/opencode-gpt-imagegen

Opencode GPT Imagegen: ChatGPT Image Generation for Developers

Opencode GPT Imagegen is the OpenCode plugin that lets you generate images using your existing ChatGPT subscription without extra API costs. It bridges the gap between your terminal-based coding agent and OpenAI's latest image generation capabilities. Instead of juggling separate API keys or paying per image, this tool rides on the same backend channel used for ChatGPT chat. You can simply ask your agent to draw something in natural language, and it saves a high-quality PNG directly to your project folder. It even accepts reference images to guide the style or edit existing targets.

🆔 @hackernewsgithubprojects

📦 hamedgitty/bloks

Bloks: Your Local AI Team

Run a private team of artificial intelligence helpers entirely on your own computer. Bloks is a desktop application that treats AI models like coworkers in a group chat, where they take turns speaking and share memory without sending your data to any external cloud server. The standout feature is its cost saving design, where a smart leader agent hires cheaper models for routine tasks while keeping the expensive models for final judgment, just like a real company. Because everything stays on your machine, you have full control over what these agents can touch and do.

🆔 @hackernewsgithubprojects

📦 oruk-ai/orukeet

Orukeet: The Speech AI That Listens Better

Orukeet is the speech-to-text engine that actually listens. Instead of letting a computer just guess words, it swaps half its internal listening filters for mathematically perfect ones that never change. This tiny trick makes it hear twenty-five languages more clearly than the version it came from. It cuts down errors on messy, real-world audio, meaning your voice gets typed out faster and more accurately on your laptop or phone. It’s a clever, quiet win for anyone who hates fixing misspelled transcripts. You get a cleaner, sharper recording that just makes sense. That’s the takeaway: better ears, less noise, and your words, finally, exactly as you meant them.

🆔 @hackernewsgithubprojects

📦 ruc-datalab/skilladam

SkillAdam: Auto-Improving Skills for Your AI Agent

SkillAdam is the optimization tool that actually tests your AI skills before you use them. It plugs directly into Claude Code, Codex, and Cursor to take a specific prompt you wrote and run it against real tasks. The system analyzes where the instructions fail, proposes specific text changes, and only updates your file if the new version performs better. This turns vague prompt engineering into a measurable, scientific process. You no longer guess if a tweak works; you see the data prove it. It is the fastest way to make your AI agents genuinely smarter and more reliable for your daily workflow.

🆔 @hackernewsgithubprojects

📦 jeremyipark/vision-demos

vision-demos: The Computer Vision Project

Vision demos is the computer vision collection that turns everyday video into measurable athletic data. It solves the problem of how to quantify physical performance without expensive wearables by using a single pose estimation model to track body points in real time. The most surprising feature is its ability to automatically count chin-up repetitions by analyzing the vertical displacement of your torso relative to your hands, effectively acting as a digital spotter that times every ascent and descent. It also includes tools for measuring running cadence and scoring dance synchronization, making it a versatile toolkit for anyone interested in applying machine learning to fitness.

🆔 @hackernewsgithubprojects

📦 shmidtqq65/loxley

Loxley: The Exit Sniper for Robinhood Chain

Get out of the exit. Loxley is a local tool for Robinhood Chain that handles the messy part of token launches. It watches for new coins, scores their safety, and waits for that scary opening tax to drop before you buy. But here is the kicker: it does not just enter, it marks your exact exit price every few seconds. The moment the developer who launched the coin sells their tokens, Loxley instantly triggers a sell order for you. It reads the chain by default, so nothing is locked away. You just watch the terminal, and if it looks safe, it signs the trade.

🆔 @hackernewsgithubprojects