📦 truefoundry/trueforge

TrueForge: Stop Building Agent Plumbing

Building an AI agent usually means drowning in code just to manage memory and tool calls. TrueForge changes that by acting as a complete runtime layer that handles the heavy lifting for you. It manages model calls, sandboxes, and session states, so your agent actually works instead of just chatting. The coolest part is its sandbox-as-a-tool feature, which only spins up isolated execution environments when the agent truly needs to run code. This keeps your setup fast and your secrets safe, while letting you focus on the logic, not the plumbing. If you want to build real agents without building a house every time, this is your shortcut.

🆔 @hackernewsgithubprojects

📦 yetone/cumora

Cumora: The Chat App Where AI Agents Are Coworkers

Cumora is a team chat app where AI agents are actual employees, not just chatbots. You invite them to a group, and they start working alongside humans, grabbing tasks from a shared board, sending real emails, and chatting in the same threads. The coolest part is that these agents have persistent memory and can coordinate with each other without stepping on toes, whether they run on cloud servers or your own local machine. It is basically a digital office where the robots handle the busywork so you can focus on the big picture. If you want a team that never sleeps and never argues, this is how you build it.

📰 https://news.ycombinator.com/item?id=49338707

🆔 @hackernewsgithubprojects

📦 coreyhaines31/makerskills

Makerskills: Your AI Operator Brain

Give your AI agent the exact playbooks to handle your business decisions, research, and creative workflows. Makerskills is a collection of twenty structured skills that turns tools like Claude Code into a serious personal operator. It stops the agent from guessing and forces it to follow proven frameworks for everything from picking domain names to modeling financial scenarios. The coolest part is the Maker Council skill, which simulates a board of advisors like Paul Graham and Jeff Bezos to pressure-test your ideas. Instead of getting generic advice, you get specific, conflicting perspectives that force you to make a better call.

🆔 @hackernewsgithubprojects

📦 electricitysheep/dsh-handbook

The Missing Manual for DeepSeek's New AI Engine

DeepSeek just open-sourced a new engine that turns any AI model into a fully customizable, code-running assistant. The dsh handbook is the missing guide that teaches you how to actually use it. It’s not just a list of commands; it’s a deep dive into the framework’s inner workings, showing you how to build your own plugins, tweak performance, and run complex real-world tasks. The coolest part? You can see exactly how to cut costs by nearly half just by adjusting a few settings, and learn how to build your own tools from scratch.

🆔 @hackernewsgithubprojects

📦 dfin/neural-network-visualisation

Neural Network Visualisation: Draw a Digit, Watch the AI Think

Neural Network Visualisation is the browser tool that makes you actually see how an AI reads your handwriting. You draw a number on a simple grid, and instantly, a 3D model of a neural network lights up. It’s not just a black box guessing; you watch the raw pixels travel through the network, with colors showing how strongly each part is firing. It’s like watching a thought happen in real time. The coolest part? You can scrub a timeline to see the network learning. It’s a fun, tactile way to understand the magic behind digit recognition without reading a single line of code. Just draw, watch, and learn.

🆔 @hackernewsgithubprojects

📦 robonuggets/gauntlet-loop

The AI Trick That Refuses Good Enough

Gauntlet Loop is the prompt generator that forces your AI agent to stop settling for good enough output. The core problem with most AI tools is that they grade themselves against vague ideas, which leads to drift and weak results. This project fixes that by turning any goal into a strict competition. You pick a real, existing reference, like a specific website or article, and the system creates a prompt where a builder works on the task while a separate, harsh critic blind compares the new work against that reference. The loop keeps running until your version beats the reference, not just until the AI feels satisfied.

🆔 @hackernewsgithubprojects

📦 awesome-dsh-plugin/dsh-find-plugin

Dsh Find Plugin: The Agent Tool That Finds Plugins For You

Dsh find plugin is the discovery tool that lets your coding agent find and install new capabilities for you. You simply tell the agent what you need, such as a way to get notified when a task finishes, and it automatically searches the public GitHub ecosystem. The results are ranked by stars, so you see the most popular options first. Each result includes a short description and a ready-to-run install command that the agent can execute for you. If a plugin is on the curated awesome list, you even get a hand-written, bilingual description. It is a simple, live search that turns your agent into a plugin finder.

🆔 @hackernewsgithubprojects

📦 aiscientists-dev/academic-humanizer

Academic Humanizer: The Edit That Fixes AI Slop Without Killing the Science

Academic Humanizer acts as a surgical filter for research papers that strips out generic AI phrasing while keeping every data point and citation exactly where it belongs. It solves the problem of AI drafts sounding inflated, vague, and disconnected from the author’s actual voice. The tool works by auditing claims against the evidence, ensuring that verbs match the strength of the data without flattening the necessary precision of scholarship. It even has a specific mode for grant proposals that preserves the ambition reviewers look for on the first page. This is the editing pass that keeps your science sharp and your voice distinct.

🆔 @hackernewsgithubprojects

📦 opensparx/masteragent

MasterAgent: On-Device AI

Run a full AI assistant directly on your car or phone without ever touching the internet. This open source project, called MasterAgent, lets you build smart agents that process voice commands locally, meaning your personal data never leaves your device. The standout feature is its speed; it handles eighty percent of requests using simple pattern matching instead of heavy calculations, so it responds in under one hundred milliseconds. This makes it ideal for situations where cloud latency is too slow or privacy is a concern. By keeping everything local, it offers a fast and private alternative to traditional cloud-based AI tools.

🆔 @hackernewsgithubprojects

📦 viettran-edgeai/codex_workflow

Cut Token Costs with Codex Workflow

Slash your AI coding costs without sacrificing quality. Codex Workflow installs a smart layer onto your coding agent that drastically reduces overall token usage. It does this by intelligently routing tasks through three distinct levels of effort, ensuring simple questions don't trigger expensive, complex workflows. Instead of dumping every detail into a massive context window, the system actively manages your project's progress and documentation across sessions. It acts like a diligent project manager, preserving operational context so you never have to repeat yourself.

🆔 @hackernewsgithubprojects

📦 cwida/superkmeans

Super K-Means: 100x Faster Clustering

Super K-Means can process ten million high-dimensional embeddings in under a minute on a single CPU. This library solves the bottleneck where vector clustering usually slows down search systems. It groups similar data points together, but it does so much faster than existing tools without sacrificing quality. The trick is smart math that skips unnecessary calculations while keeping results accurate. You can use it in Python or C++ on standard hardware. It’s a powerful tool for anyone building search features who needs speed. The takeaway is simple: if your clustering is slow, this is the upgrade you need.

🆔 @hackernewsgithubprojects

📦 aloshdenny/claude-awm

How Invisible Unicode Characters Beat AI Watermarks

You can hide text changes from AI detectors by using invisible unicode characters that look like standard formatting but actually break the math. This project, called Claude AWM, proves that while most simple edits get caught by basic cleanup, specific unicode marks slip right through because they are technically valid parts of other languages. It shows that while you can scrub away zero-width spaces, you cannot simply delete variation selectors without breaking real emoji and Asian text. This exposes a critical blind spot in how we trust automated text verification, proving that the weakest link is often the assumption that invisible equals useless.

🆔 @hackernewsgithubprojects

📦 junbolian/admitor

Admitor: Learning Without Labels

Admitor is the research framework that proves AI can learn from experience without needing labeled answers. This project tackles the tough problem of getting large language models to solve complex optimization puzzles like routing or scheduling without a teacher holding the keys. It works by letting the AI build a personal library of skills through trial and error, effectively learning what works by simply trying different approaches until it finds a solution. This is a huge deal because it removes the expensive need for expert-labeled data, allowing the model to figure out the right path on its own.

🆔 @hackernewsgithubprojects

📦 joelhooks/pdf-brain

PDF Brain: Your Local AI Research Assistant

pdf-brain turns your scattered research files into a smart, searchable library that lives entirely on your machine. It processes your PDFs and markdown notes, uses local AI to read and understand them, and tags everything with meaningful concepts. This means you can ask natural questions and get precise answers drawn from your own documents. It runs locally, so there are no cloud costs or data privacy concerns, making it a powerful tool for anyone who needs to organize and query their personal knowledge base without sending sensitive files to external servers.

🆔 @hackernewsgithubprojects

📦 integral2u/sharpmind

Run AI Models Entirely in C Sharp

Run large language models entirely in C Sharp with SharpMind. This project handles everything from loading models to fine-tuning them, all without needing Python or native code. The standout feature is its disk-streaming mode, which lets you run models bigger than your computer’s memory by paging layers in and out on the fly. It feels like building a custom engine from scratch, complete with advanced speed boosts and training tools built right into the same simple application. If you are tired of switching between languages just to train and deploy models, this is your all-in-one solution that keeps your entire workflow in one familiar, fast language.

🆔 @hackernewsgithubprojects

📦 antinomie-lab/pi-book

Inside the Pi Agent Loop

Pi Book is the architecture guide that actually shows you the code behind the agent loop. It is a workspace for a book that breaks down a specific library, explaining exactly how it handles tool execution and data streaming without becoming a bloated framework. The coolest part is that every single claim in the text is backed by a direct link to the exact line of source code, so you can verify every detail yourself without guessing. It is organized to build your mental model step by step, from the big picture down to the intricate parts.

🆔 @hackernewsgithubprojects

📦 inclusionai/areno

AReno: Train AI Models on Your Laptop

Train a large language model on your single local machine using AReno. This toolkit eliminates the need for massive server clusters or complex infrastructure, letting you run reinforcement learning, supervised fine-tuning, and even agentic workflows directly on your own hardware. It works by handling everything from data loading to the final training step in one self-contained package, so you are not stuck wiring together separate tools. You can even watch a small model learn to play a browser game from scratch, proving the system works end to end. The real value is accessibility. You no longer need a giant cluster to experiment with advanced AI techniques.

🆔 @hackernewsgithubprojects

📦 bytedtsinghua-sia/cuda-agent

This AI Beats Top Models at GPU Coding

Watch this AI agent outcode the best general purpose models at writing fast GPU code. Researchers trained a specialized agent to generate high performance CUDA kernels, beating advanced systems on difficult benchmarks. The project releases the training data and the exact workflow rules used to build it. It turns complex hardware optimization into a repeatable, automated process. This gives developers a proven blueprint for accelerating their own models. You get a clear path to faster code without years of manual trial and error. This is how the next generation of efficient AI training is being built. Check it out to see how agentic reinforcement learning is changing GPU programming.

🆔 @hackernewsgithubprojects

📦 perseus-computing-llc/mimir

Give Your AI Agent a Permanent Brain

Stop letting your AI agent forget everything the moment a session ends. Mimir gives your AI persistent, encrypted memory that lives locally on one file, completely offline. It uses military-grade encryption to keep your data safe, so your agent can remember past lessons and facts across different projects without sending anything to the cloud. It works with any tool, and because it is local-first, you keep total control. No servers, no subscriptions, just smart, private memory. If you want your AI to actually learn and stop repeating mistakes, this is the tool you need.

🆔 @hackernewsgithubprojects

📦 thanhhff/selectanytree

SelectAnyTree: Click to Segment Forests

One click on a 3D forest scan isolates a single tree, and it nails the accuracy target almost instantly. SelectAnyTree solves the nightmare of manually labeling millions of LiDAR points in dense woods, which is usually tedious and error-prone. It works like a smart paint tool where you simply tap a spot to select a specific tree, and the system instantly draws the outline. The clever part is that it processes the entire forest scene just once, so switching between different trees is fast because it doesn't have to re-analyze the raw data every time. It beats previous tools by a wide margin using far fewer computing resources.

🆔 @hackernewsgithubprojects

📦 michaelbf16/bedrock-rl

Train AI Agents in Minecraft with Bedrock RL

Bedrock RL is the modular framework that finally lets you train AI agents inside Minecraft with reproducible, verifiable results. It solves the headache of messy AI training by giving you a single command to generate synthetic data, check your setup, and run reinforcement learning experiments. The coolest part is its synthetic data generator, which creates realistic gameplay trajectories that you can replay in three different visual styles to prove your agent actually learned the task. You define your goal in simple config files, and the system handles the heavy lifting of data creation and model training.

🆔 @hackernewsgithubprojects