📦 gotempsh/temps

One Binary, Zero Subscriptions

Deploy your code, track your errors, and send your emails without juggling six different subscriptions. Temps is a single, self-hosted tool written in Rust that replaces the usual stack of hosting, monitoring, and analytics platforms. It gives you over four hundred command-line actions to manage everything from your server status to your user behavior, all from one place. The real kicker? It plugs directly into AI coding assistants, letting them handle your infrastructure for you. You get full control and complete transparency without the monthly bills. Stop renting your stack and start owning it.

🆔 @hackernewsgithubprojects

📦 sulabhdubey/rta-smriti-brain

Rta-Smriti Brain: The AI Memory That Never Forgets

Rta-Smriti Brain treats your AI coding agent like a new hire who needs a map, not just a prompt. It builds a private, local database of your project’s code, decisions, and history so the agent knows exactly what is true right now. The coolest part is how it labels evidence, distinguishing a verified test result from a wild guess so the AI never acts on stale data. It keeps everything on your machine, no cloud required, meaning your code stays yours. If you are tired of copy-pasting context to explain your project, this tool fixes that friction permanently.

📰 https://news.ycombinator.com/item?id=49511544

🆔 @hackernewsgithubprojects

📦 vermellcc/vermell

Vermell: The C++ Web Framework That Actually Exists

Vermell is the C++ web framework that refuses to drag in a single external dependency. It was built from scratch to prove you can serve fast web traffic using nothing but raw Linux sockets and a smart event loop. The coolest part? It runs a full non-blocking engine that keeps your server responsive even when slow clients try to drag it down, all without a garbage collector or heavy runtime. It is strictly typed, incredibly fast, and compiles with just one simple command. If you have ever wanted to write a high-performance server in pure C++ without fighting a massive ecosystem, this is your answer.

📰 https://news.ycombinator.com/item?id=49518121

🆔 @hackernewsgithubprojects

📦 zoharbabin/due-diligence-agents

How Due Diligence Agents Connects the Dots

Due Diligence Agents is the open source tool that connects the dots between legal and finance teams. It uses thirteen artificial intelligence specialists to read through hundreds of contracts at once. The standout feature is that it traces every risk it finds to a specific page and quote in your documents. This means you never have to guess where a problem comes from. It helps your team see the big picture without spending weeks manually linking separate reports. It is built to speed up your review process, not to replace your professional advisors.

📰 https://news.ycombinator.com/item?id=48170393

🆔 @hackernewsgithubprojects

📦 alayalab/worldsculpt

WorldSculpt

WorldSculpt can turn a single video of a cluttered room into a complete, interactive 3D model of every single object inside it. Most tools struggle to separate a messy scene into distinct items, but this project solves that by using a model trained on single objects to rebuild complex environments with hundreds of pieces. It takes grounded video data and sculpts it into a compositional mesh, meaning each chair, cup, and table exists as its own editable 3D entity rather than one giant blob. This is huge for game developers and creators who need precise, manageable assets from real-world footage.

🆔 @hackernewsgithubprojects

📦 slavkoprytula/qcell

QCell

QCell takes a single microscopic image containing two hundred and twenty-three distinct cells and correctly identifies every individual one. Standard image tools fail here because transparent cell membranes blend into one another, making boundaries nearly invisible. This research project fixes that problem by breaking each cell’s digital signal into parts representing the visible section and the hidden parts beneath neighbors, then reassembling them to see the complete shape. It also nudges the software to treat different cells as distinct entities rather than merging them. Accepted at a major 2026 computer vision conference, this work provides a clearer way for machines to count and map crowded biological samples without human guessing.

🆔 @hackernewsgithubprojects

📦 truespar/paddock

Paddock: Rust-Powered AI Inference

Paddock is a native Rust inference engine that runs large language models directly on your NVIDIA graphics card without relying on external Python wrappers. It processes raw model files by executing custom graphics card code to generate text, handle images, and transcribe audio at high speed. The project includes a built-in web studio that lets you run local models side by side with cloud services using a single interface. By handling everything from memory management to network requests internally, it provides a fast, self-contained way to host your own artificial intelligence. If you want full control over your data and hardware, this open-source tool is worth checking out.

📰 https://news.ycombinator.com/item?id=49561513

🆔 @hackernewsgithubprojects

📦 git-pkgs/forge

One Command, Every Code Host

Forge lets you run the exact same terminal commands no matter which code hosting platform your project lives on, automatically detecting the remote to handle the rest. Instead of memorizing unique syntax for GitHub, GitLab, or Bitbucket, you type one familiar instruction to list issues or manage pull requests, and the tool routes it to the correct backend. It solves the headache of switching between different tools or APIs for every new repository you join, giving developers a consistent experience across any environment. You stop worrying about which specific platform features are available and just get your work done.

📰 https://news.ycombinator.com/item?id=49595229

🆔 @hackernewsgithubprojects

📦 konohamaru04/comfyui-nvidia-dlss-frame-interpolation

ComfyUI NVIDIA DLSS Frame Interpolation: Real-World Results

A five-second video clip can be processed on a single graphics card to jump from twenty-four to one hundred and twenty frames per second, adding ninety-six completely new frames that did not exist in the original file. This is the core capability of the ComfyUI NVIDIA DLSS Frame Interpolation project, a tool that plugs into the ComfyUI creative suite to use neural processing for video and image enhancement. It solves the problem of choppy motion by generating smooth intermediate frames and simultaneously increasing resolution, which is useful for creators wanting higher quality outputs without manual editing.

🆔 @hackernewsgithubprojects

📦 stephen-gzk/videosearcher

VideoSearcher

An AI agent can now watch a long video and hunt for specific facts by zooming into single frames and running reverse image searches. This project, VideoSearcher, teaches a model to investigate videos like a curious detective. It does not just summarize what happens; it actively searches for evidence. The agent uses specialized tools to locate exact moments, magnify blurry details, and verify facts against the web. This solves a major problem where current AI misses crucial visual context in longer clips. By combining reinforcement learning with these specific tools, it creates a deeper understanding of video content.

🆔 @hackernewsgithubprojects

📦 howmp/dsh-pentest

D-S-H Pentest: Visual Hacking Logs

D-S-H Pentest is the hacker companion that turns messy AI coding sessions into a clear visual map. It plugs directly into DeepSeek Harness to record every goal, clue, and vulnerability you uncover during an authorized security test. The coolest part is the live web dashboard that draws your investigation as an interactive graph, linking your specific actions to the assets they impact. It even generates a clean Markdown report with reproducible steps, so you never lose track of how you found that critical flaw. If you are using AI to assist with security work, this tool keeps your logic organized and your proof documented in one place.

🆔 @hackernewsgithubprojects

📦 aloshdenny/reverse-synthid-text

Reverse Synthid Text

Reverse Synthid Text is the reverse engineering toolkit that exposes the fragile nature of Google's AI watermark. While Synthid was designed to tag artificial writing, this project shows that you can erase that tag by simply paraphrasing the text with a different model. Because the watermark hides inside the exact sequence of words, changing the wording destroys the hidden fingerprint entirely. This means the label meant to identify machine-generated content can be wiped away by rewriting, which is a critical insight for anyone tracking the origin of digital text.

🆔 @hackernewsgithubprojects

📦 tencent/teamai-cli

TeamAI: The Tool That Syncs Your Whole Team's AI

TeamAI is the command-line tool that finally makes sure every developer on your team is using the exact same AI setup. It solves the messy problem of everyone tweaking their own coding assistants in isolation. By storing your team's preferred rules, skills, and knowledge in one shared repository, it automatically syncs them to any AI agent you use, from Claude to Cursor. The coolest part is that it turns your team’s hard-won debugging lessons into shared memory that every AI can recall. It basically makes your whole team smarter at the same time.

📰 https://news.ycombinator.com/item?id=49596714

🆔 @hackernewsgithubprojects

📦 nvidia/personaplex

Talk to Any Character Instantly with PersonaPlex

Make an AI voice that actually listens and reacts like a real person. PersonaPlex is a real-time speech model from Nvidia that lets you talk with any character you invent. Instead of just typing text, you can pick a specific voice and give the AI a detailed role, like a friendly teacher or a busy customer service agent. It is built on the Moshi architecture, which means it processes your words and speech at the same time, creating a natural back-and-forth flow. The coolest part is that it can handle interruptions and pauses, so it feels like a genuine conversation rather than a robotic exchange.

🆔 @hackernewsgithubprojects

📦 advaitpaliwal/feynman

Feynman: The AI Research Agent

Feynman is an open source AI agent that reads scientific papers, checks their code, and writes summaries you can trust. It solves the problem of information overload by acting as a tireless lab assistant. You ask it to review a topic, and it searches databases, compares claims against actual code repositories, and cites its sources. It even runs four distinct internal agents to research, review, write, and verify facts. This means you get a verified research brief instead of a generic summary. If you are a developer or scientist, this tool turns hours of manual reading into a few minutes of automated, accurate analysis.

🆔 @hackernewsgithubprojects

📦 uvwt/agentdock

AgentDock: ChatGPT Runs Your Computers

AgentDock lets you take full control of your physical computers and remote servers directly from a standard web browser chat. Instead of keeping your AI locked inside a secure sandbox, this tool hands the model safe, permission-bound access to your actual machine so it can write code, run commands, and deploy projects in the real environment. It’s like giving your favorite chat assistant a secure remote control for your digital workspace, letting you manage multiple devices from one conversation without juggling endless terminal tabs or burning through premium coding credits.

🆔 @hackernewsgithubprojects

📦 huggingface/funes

Funes: Give Your AI Agents a Long-Term Memory

Stop letting your coding assistants forget everything the moment you close the terminal. Funes solves this by building a durable, searchable memory of every past session you’ve had with agents like Claude Code or Codex. It’s like giving your AI tools a persistent brain that remembers the decisions, dead ends, and rationale from weeks ago. The coolest part? Your entire work history becomes a portable dataset you can publish to the Hugging Face Hub. Suddenly, your teammate or even another machine can recall exactly how you solved a problem before. It’s not just about saving text; it’s about sharing the actual thinking process behind your code.

🆔 @hackernewsgithubprojects

📦 vllm-project/agentic-api

Run OpenAI Codex on Your Own GPUs

Point OpenAI's coding agent at your own hardware and let it run without sending your code to the cloud. The project called Agentic API sits in front of your local model server, taking over the messy work of remembering conversation history and executing tools. Instead of your app tracking every message, you simply send one request, and the server handles the rest. It even speaks the exact language Codex uses, so you can drive that powerful agent using open-source models you control. It is a clean way to build self-hosted AI applications without losing the convenience of stateful conversations.

🆔 @hackernewsgithubprojects

📦 topoteretes/cognee-integrations

Cognee Integrations

Give your AI agents a real long-term memory. The Cognee Integrations repo gives popular agent frameworks, like CrewAI and LangGraph, a persistent knowledge layer. Instead of forgetting everything after a session ends, these agents remember past conversations and facts. The cool part is it works across different tools, so your agent doesn't start from zero every time. It builds a lasting brain for your apps, making them feel much smarter and more personal. If you build agents, this is the missing piece for true continuity. Grab it to make your AI actually remember who it is talking to.

🆔 @hackernewsgithubprojects

📦 agenticoding/pi-agenticoding

pi-agenticoding: Your AI Coding Crew Manager

pi-agenticoding is the workflow layer that lets you define AI coding crews once and reuse them forever. It solves the problem of constantly re-explaining which models should handle which tasks. You save a specific workflow, like a critical code review, and the system automatically assigns the right models to the right roles. It uses a shared notebook to keep important decisions safe while discarding the messy working notes. This means you get consistent, high-quality code checks without manually steering every single step. It turns repetitive AI management into a simple, repeatable command that actually works. Stop wasting tokens on instructions and start focusing on the code that matters.

🆔 @hackernewsgithubprojects