📦 haolpku/worksurface-bench

How to Tell if Your AI Agent Knows Where to Look

WorkSurface-Bench is the benchmark that finally checks if your AI agent knows which tools to grab. Most agents just guess, but this project tests if they can actually route to the right source. It gives them documents, tables, and maps to choose from. Then it scores if they picked the right one. It separates smart routing from lucky guessing. That is the real test of enterprise intelligence. So you know exactly where your agent stands.

🆔 @hackernewsgithubprojects

📦 xhluca/session-migrate

Session Migrate: Move Your Coding History

Session Migrate lets you lift a live coding conversation out of one AI tool and drop it straight into another, keeping the full context intact. Instead of pasting text or starting from scratch, this tool reads the native session file from an agent like Claude Code or Codex and writes a new, resumable session into a completely different harness, like Pi or Copilot. It handles the messy translation of messages and tool results so the new agent knows exactly where you left off. It is not just an export feature; it creates a fresh, native session that the target application can pick up and continue as if nothing happened.

📰 https://news.ycombinator.com/item?id=49425594

🆔 @hackernewsgithubprojects

📦 skillsynchq/txcript

Continue Claude Chats in Codex with txcript

Pick up a coding conversation where you left off, even if you switched tools. txcript lets you export a full session from one AI assistant, including every message, tool call, and reasoning step, and import it directly into another. This solves the annoying problem where your AI coding agent forgets everything when you change platforms. Instead of starting over, you carry your entire work history across different coding environments like Claude Code and Codex. It acts as a universal translator for your AI chats. If you are moving projects between tools, you do not have to re-explain your context.

📰 https://news.ycombinator.com/item?id=49719696

🆔 @hackernewsgithubprojects

📦 tencentcloud/octop

Octop: Your Private AI Team

Build a private AI assistant that actually runs on your own hardware. Octop is a self-hosted system that lets multiple people use separate, specialized AI agents on a single server. It solves the problem of generic chatbots that lack context or privacy, giving you full control over who can access what. The coolest part is that it supports multi-agent setups, meaning you can have different digital workers handle different tasks for different team members simultaneously without them getting confused. It is essentially a private, customizable office where your data never leaves your walls. If you want total ownership over your AI experience, this is your playground.

🆔 @hackernewsgithubprojects

📦 meituan-longcat/longcat-video

LongCat Video: The Open Source Video Generator

LongCat Video is the open-source video model that finally keeps long generations stable without the usual quality drop. It solves the headache of stitching together short clips by natively creating minutes-long footage where colors and details stay consistent from start to finish. Developed by Meituan, this thirteen billion parameter tool handles text-to-video, image-to-video, and even audio-driven character animation all in one place. You can run it on your own hardware to generate high-definition footage quickly using a smart generation strategy that balances speed and quality. It is a huge step for anyone wanting to build their own cinematic content without relying on closed, expensive commercial services.

🆔 @hackernewsgithubprojects

📦 vercel/eve

Eve: Code Agents Using Plain Files

Build AI agents using nothing but plain text files. Vercel’s Eve framework treats your folder structure as the source of truth, meaning your agent’s instructions, tools, and schedules all live right in front of you. This solves the black box problem because you can actually read, edit, and inspect exactly what your AI is doing. It’s a filesystem first approach that makes debugging and customization far more intuitive than hidden configuration files. By keeping everything visible and standard, you get a project you truly own. Next time you build an agent, try dropping a simple markdown file into a folder and see how much control you gain over the behavior.

🆔 @hackernewsgithubprojects

📦 fluxvla/fluxvla

FluxVLA: One Platform for Real Robot AI

Deploying robot AI usually means wrestling with mismatched codebases and slow inference pipelines. FluxVLA solves this by providing a single, unified platform that handles everything from data processing to running models on physical hardware. The standout feature is its ability to run heavy vision language models directly on compact edge devices like the Jetson Orin. This means you don't need expensive cloud servers to make a robot think and act in real-time. By standardizing the interface, it removes the technical friction that usually slows down research.

📰 https://news.ycombinator.com/item?id=49724566

🆔 @hackernewsgithubprojects

📦 nhtera/oximux

OxiMux: Parallel Coding Agents for Mac

OxiMux lets you run multiple AI coding assistants at the same time on a single Mac. The most surprising part is that it gives every agent its own private, isolated copy of your code project. This means Claude and Codex can work on different features simultaneously without ever breaking each other's work. The tool handles all the messy setup automatically, so you just focus on the code. It turns your computer into a multi-agent cockpit where you can review changes side by side. If you want to get more done without the chaos of manual branching, this is the tool to watch.

🆔 @hackernewsgithubprojects

📦 jedisct1/turbocrypt

TurboCrypt: Keep Private Files in Public Git Repos

TurboCrypt lets you commit private notes to a public Git repository without anyone seeing the actual content. This command line tool encrypts files so fast that it handles entire directory trees in parallel, keeping your secrets safe even when the code itself is public to the world. It solves the common problem of needing to share encrypted files with teammates while keeping them out of reach from everyone else. The software uses modern cryptography to secure your data and works on almost any operating system. You can also mount these encrypted folders to work with your normal apps.

📰 https://news.ycombinator.com/item?id=49725409

🆔 @hackernewsgithubprojects

📦 maelic/relateanything

RelateAnything: Any Relation, Any Image

RelateAnything identifies relationships between objects in an image using words you provide at the moment of analysis, rather than a fixed list trained into the system. This small, fast model processes visual data in real time to find connections like riding or standing next to, even for concepts it has never seen before. It solves the problem of rigid software that fails when you need a specific, unusual connection. Because it is tiny and lightweight, it runs instantly on standard hardware, making it easy to build applications that understand complex interactions in video streams. You can define exactly what relationships matter to your project without retraining the whole system.

📰 https://news.ycombinator.com/item?id=49713313

🆔 @hackernewsgithubprojects

📦 madarco/agentbox

AgentBox: Spin Up Instant Coding VMs

AgentBox lets you launch a private, isolated computer for your coding AI with a single command. The standout feature is its checkpoint system. It creates a warm snapshot of your environment that boots in under a second. This means you can spin up multiple parallel agents on your local machine or in the cloud without the usual startup lag. The tool automatically moves your project files into these secure boxes. It keeps your private credentials safe on your host while letting the agents work freely. It is a practical way to test code in a sandbox that never touches your real system.

📰 https://news.ycombinator.com/item?id=49728120

🆔 @hackernewsgithubprojects

📦 ordewell/ordewell

Ordewell: The Coding Agent Orchestrator

Ordewell is the task orchestrator that finally lets you control exactly which AI model writes each part of your code. You type a single goal, and a read-only planner studies your repository to build a detailed, step-by-step roadmap. The coolest part is that you get to assign different tools to different tasks. Maybe a cheap model handles a simple update while a powerful one tackles a complex refactor. You can tweak this plan in real time before anything runs, ensuring the right brain is applied to the right problem. It is a smart way to manage multiple coding agents without losing your mind.

📰 https://news.ycombinator.com/item?id=49712276

🆔 @hackernewsgithubprojects

📦 thebanri/limoni

Lemons: The Terminal UI That Renders 3D

Limoni is the terminal interface framework that renders real-time 3D graphics directly in your command line. While most terminal tools are stuck in black and white, this Go library draws rotating meshes with lighting and shading right in your text editor, running at sixty frames per second. It achieves this speed by using a zero-allocation design that prevents your computer from freezing during heavy animations. You get the familiar feel of modern desktop apps without leaving your terminal, making it perfect for building fast, responsive tools. If you write software in Go, this is the framework that brings high-fidelity visuals to the humble command prompt.

🆔 @hackernewsgithubprojects

📦 seaof0/dsh-redteam-model

dsh-redteam-model

A massive library of AI agent skills is now dedicated to authorized red team security research. The dsh-redteam-model project acts as a specialized brain for offensive security professionals, covering everything from penetration testing to code auditing. It solves the problem of scattered knowledge by organizing hundreds of detailed playbooks and reverse engineering techniques into a single, modular framework. You can plug these skills into your own AI agents to automate complex security assessments, effectively giving your tools the expertise of a seasoned pro. This is a powerful resource for anyone looking to bring structured, high-level methodology to their authorized security work.

🆔 @hackernewsgithubprojects

📦 spikecalls/flydrones

FlyDrones

FlyDrones is the open-source project that turns a real fruit fly’s brain into a drone pilot. It takes the complete wiring diagram of a male fruit fly’s nervous system, simulates it in your computer, and feeds it to a real drone. The coolest part? The brain doesn't get direct commands like 'fly up.' Instead, it sees optic flow illusions generated by your hand gestures, and its own reflexes decide how to steer. It’s a genuine closed loop of biology and code, proving that a tiny insect brain can genuinely guide a machine. If you love the intersection of neuroscience and robotics, this is the one to check out.

🆔 @hackernewsgithubprojects

📦 zhengkid/dream-rsi

How Dream-RSI Lets AI Learn From Its Own Dreams

Dream-RSI is the AI research project that turns past failures into future success. It solves a tricky problem where AI systems waste huge amounts of computer time guessing how to improve themselves. Instead of starting from scratch every time, this system records every step of its previous attempts. It then replays those recorded steps like a video game, allowing the AI to practice and test new strategies without actually running the heavy calculations again. This means the agent can learn from its own history, getting significantly faster and more efficient at finding solutions for complex math and coding tasks.

🆔 @hackernewsgithubprojects

📦 yifanzhang-pro/flashreinforce

How FlashREINFORCE Simplifies AI Training

Train language models to act independently without the messy overhead of traditional reinforcement learning. FlashREINFORCE lets you optimize AI agents using just one pass per task, which solves a huge headache for developers. Usually, training these systems requires generating multiple attempts for every problem and waiting for them all to finish, which is slow and wasteful. This project changes that by letting you use a single attempt, then cleverly adjusting the math to fix any outdated information. That means your AI learns from new data the instant it is ready, without getting stuck in a queue. It is a smarter, faster way to teach computers to solve real-world tasks.

🆔 @hackernewsgithubprojects

📦 eternity4719/howtolivebetter

How To Live Better: The Cost-Benefit Guide to Life

How To Live Better is the evidence-based life guide that treats your existence as a budget to be managed. It cuts through health noise and legal confusion by listing four hundred and ninety-eight practical actions, each ranked by a simple cost-benefit ratio. The project calculates exactly what you spend in money, time, or effort against what you gain in longevity, financial safety, and personal freedom. It refuses to cite blogs or opinions, relying instead on peer-reviewed journals and official government documents to back up every claim.

🆔 @hackernewsgithubprojects

📦 bigpengsays/bigpeng-hot-gzh

BigPeng Hot GZH: 7 Viral Formulas for AI Content

BigPeng Hot GZH is the content toolkit that distills viral AI articles into seven reusable title formulas. It solves the blank-page problem by analyzing over a hundred successful pieces to create a structured guide for WeChat creators. Instead of guessing what will pop, you use proven templates for digital lists or reversal conclusions to craft headlines that actually get clicked. It is a pure strategy pack, helping you pick the right angle before you write a single sentence. It gives you the specific language that drives traffic, so you spend less time brainstorming and more time writing.

🆔 @hackernewsgithubprojects

📦 arsumbris/arsumbris

Arsumbris: The IDE That Thinks

Turn scattered markdown notes into a living, typed knowledge graph that AI agents can actually use. Arsumbris is an experimental, malleable IDE designed specifically for structured information rather than just code. It reads your folders of documents and turns them into a live, queryable network where every field is checked for accuracy. This is the big idea: your agent does not just guess what your notes say. It sees exactly what is missing, broken, or connected, and gives you clear feedback. You and your agent can build the same tools and views using the same simple building blocks, creating a shared workspace where data is both read and written safely.

📰 https://news.ycombinator.com/item?id=49688958

🆔 @hackernewsgithubprojects