📦 facok/comfyui-krea2-controlnet

comfyui-krea2-controlnet

Control the composition of your AI generations by bringing advanced spatial guidance directly into your local image workflows. This project introduces custom nodes for ComfyUI that let you apply Krea2 Control LoRA models to direct your image layouts. It works by taking reference images, like depth maps or pose sketches, encoding them through a specialized visual encoder, and injecting those details as control signals directly into the model during generation. It even plays nicely with quantized models to save memory. Setting this up gives you precise, professional-grade steering over your Krea2 outputs without losing the speed of your standard setup.

🆔 @hackernewsgithubprojects

📦 jmliu206/sefi-image

sefi-image

Generate highly detailed images from text prompts by letting the AI plan the layout before painting the fine details. Instead of trying to figure out shapes and textures all at once, this approach runs a semantic-first diffusion process that maps out the overall structure of your scene slightly ahead of the texture details. This gives the generation process a much cleaner layout to build on, leading to beautifully organized compositions. You can run this locally using the simple python API, which offers multiple model sizes and even a fast turbo option that creates images in just four steps.

🆔 @hackernewsgithubprojects

📦 thinking0things/albertmicro

AlbertMicro

AlbertMicro is a miniature 3D-printed quadruped robot powered by an ESP32 microchip that packs a surprisingly big personality. Walking on four servo-powered legs, this tiny companion uses an onboard screen to display expressive, animated eyes with moods ranging from happy to curious. With a simple Bluetooth command from your phone, the robot sits down and flips its face into an information hub showing the local weather, the current time, and your live YouTube subscriber count. It bridges the gap between a desk toy and a functional smart display, making custom robotics incredibly accessible and fun to build.

🆔 @hackernewsgithubprojects

📦 adolfousier/agentverse

agentverse

Your AI agents can now live, walk, and work together inside a virtual 3D office space. agentverse is an open-source platform written in Rust that provides a visual, real-time environment for artificial intelligence agents written in any programming language. By connecting over a simple local network API, your custom bots get assigned physical coordinates in a simulated office world where they can walk to desks, navigate obstacles using smart pathfinding, and send messages directly to each other's inboxes. You can monitor their active tasks and manage their coordination in real time using either a beautiful three-dimensional visual interface or a lightweight terminal dashboard.

🆔 @hackernewsgithubprojects

📦 vyralcontent/content-skills

content-skills

The content-skills repository is the AI toolkit that finally takes the guesswork out of writing scripts that people actually watch. Built on data from analyzing over two hundred thousand viral videos, this project gives you specialized prompts and templates designed to plug right into tools like Claude and Cursor. Instead of relying on generic advice, it focuses on the math of viewer retention, teaching your AI assistant how to craft strong three-second hooks, write platform-specific scripts, and avoid silent-viewing traps. It is the ultimate shortcut to drafting videos that get saved and shared. Give it a try for your next post.

🆔 @hackernewsgithubprojects

📦 source-robotics/msg-compliant-ai-stepper-gripper

msg-compliant-ai-stepper-gripper

You can now build a fully open-source robotic gripper that delivers precise force feedback using cheap stepper motors. The msg-compliant-ai-stepper-gripper uses advanced field-oriented control drivers to turn standard stepper motors into highly sensitive, force-aware grabbers. This means your robot can handle delicate assembly tasks, safely work alongside humans, or collect high-quality data for training advanced AI models. It features a camera-friendly design and a completely modular setup, letting you swap rail lengths and motor sizes to adjust your grip span and squeezing power. It is an incredibly clever, budget-friendly way to give your robot a gentle touch.

🆔 @hackernewsgithubprojects

📦 sakanaai/sheaf-admm

Sheaf-ADMM

Researchers have built a training framework called sheaf-admm that solves complex reasoning tasks by letting neural-network agents negotiate with their neighbors. Instead of using one giant network, this JAX-based system breaks problems like mazes and sudoku puzzles into small, overlapping local views. Each agent solves its own tiny convex subproblem, and they coordinate using a classical optimization method guided by mathematical cellular sheaves to make sure their solutions agree where they overlap. Because the whole negotiation process is unrolled and fully differentiable, you can train the entire system end-to-end to learn complex, multi-agent coordination from scratch.

🆔 @hackernewsgithubprojects

📦 majidmanzarpour/threejs-procedural-dungeon

threejs-procedural-dungeon

Watch a beautiful, fully connected dungeon carve itself out of thin air, room by room, right in your browser. This project uses clever math to scatter rooms, push them apart, and wire them up with perfect shortcuts using minimum spanning trees. The coolest part is that the entire generation pipeline is completely deterministic. You feed it a single seed number, and it builds the exact same map down to the very last torch, complete with custom lighting, particles, and five different visual themes. It is a brilliant way to see procedural level design come to life in real time.

🆔 @hackernewsgithubprojects

📦 realiti4/claude-swap

Claude Swap

Rotate between multiple Claude Code accounts instantly without logging out by letting this clever tool handle your credentials in the background. If you run into rate limits during intense coding sessions, it monitors your usage in real time and automatically swaps in a fresh account with the most remaining quota before your active session gets blocked. It works safely alongside your running processes on macOS, Linux, and Windows, featuring a full keyboard-driven terminal dashboard to view all your account limits at a glance. It is the perfect way to keep your developer workflow moving smoothly without manual login hassles.

🆔 @hackernewsgithubprojects

📦 mathews-tom/archex

Archex

Stop burning your AI agent's context window on endless backtracking. archex lets you gather, rank, and bundle the exact code, type definitions, and structural context your agent needs before it even starts reasoning. Running entirely locally with zero hosted inference or API keys, it analyzes your repository and returns a token-budgeted context bundle alongside an explanatory receipt that proves what was included or skipped. By handling the heavy lifting of dependency mapping and code retrieval up front, it saves precious tokens and ensures your AI has everything it needs to act. Simply run it through your command line or plug it directly into Claude Code as an MCP server for...

🆔 @hackernewsgithubprojects

📦 professorpalmer/puppetmaster

Puppetmaster AI

Puppetmaster turns the AI agents you already use into a coordinated swarm of background workers that cost up to ninety-eight percent less to run. Instead of wasting expensive tokens on long chat histories, this tool runs independent worker processes in the background and writes their discoveries directly into local SQLite databases. It automatically routes every task to the cheapest model that can handle it and lets you do follow-up reads instantly for zero tokens. It is like having a smart project manager that organizes messy, multi-file engineering tasks while keeping your API spend locked down. You should give it a try.

📰 https://news.ycombinator.com/item?id=48338669

🆔 @hackernewsgithubprojects

📦 huggingface/pi-llama

pi-llama

Run and manage your local AI models right inside your terminal assistant by connecting your coding agent directly to a running server. The pi-llama repository bridges the gap between the Pi coding agent and llama.cpp. Instead of manually configuring API endpoints, this extension automatically discovers which models are running on your machine, dynamically detects their context windows, and registers them instantly. It even shows you real-time progress bars as your computer loads model weights into memory. Grab this extension to turn your local machine into a private, offline development powerhouse.

🆔 @hackernewsgithubprojects

📦 elara-labs/code-context-engine

Code Context Engine

Feeding an entire codebase into an AI assistant is a fantastic way to burn through your budget, but code-context-engine stops the bleeding by cutting AI input token costs by up to ninety-four percent. Instead of forcing your developer tools to re-read giant files every single session, this clever local tool parses your code into smart semantic chunks and runs a hybrid search right on your machine. It hooks into editors like Cursor and Claude Code, allowing your AI assistant to search for specific code snippets rather than swallowing entire files. It even tracks your real-time dollar savings on a local dashboard.

📰 https://news.ycombinator.com/item?id=48803896

🆔 @hackernewsgithubprojects

📦 funador/claude-code-merge-queue

Claude Code Merge Queue

Your laptop doesn't have to melt just because four parallel Claude Code agents are trying to run heavy builds and push their code at the exact same time. The open-source tool claude-code-merge-queue is a local, zero-cost merge queue that acts like traffic control for your AI coding sessions. Instead of letting concurrent agents trigger flaky database tests or race each other’s git pushes, this tool serializes builds and lands changes one at a time. It runs entirely on your own machine using smart, crash-safe locks, saving you from expensive cloud-runner bills. You get clean history and zero-cost safety without any coordination headaches.

🆔 @hackernewsgithubprojects

📦 pozzettiandrea/comfyui-unirig

comfyui-unirig

Comfyui-unirig is the ComfyUI extension that lets you automatically generate functional skeletal rigs for 3D character meshes directly in your node editor. Instead of spending hours manually placing joints and painting weight maps, this tool uses advanced computer vision models to analyze your 3D models and instantly build an animatable skeleton. It bundles everything you need, including a built-in instance of Blender, so you can easily change poses, skin your models, and export the finished animations. It makes the notoriously difficult process of 3D character rigging accessible to anyone building creative pipelines.

🆔 @hackernewsgithubprojects

📦 tranvuongquocdat/sidescreen

sidescreen

You can now turn any spare Android tablet into a fully functional second monitor for your Mac, completely bypassing Apple's iPad-only restrictions. An open-source project called sidescreen makes this happen using either a USB-C cable or wireless connection. It is not just mirroring your desktop; it creates a true virtual display with full touch support and sharp Retina scaling. By utilizing hardware-accelerated video streaming, it keeps lag under sixteen milliseconds so dragging windows and tapping buttons feels incredibly responsive. It is the perfect way to give an old tablet a second life as a highly portable productivity companion.

🆔 @hackernewsgithubprojects

📦 aiming-lab/simplemem

Simplemem

Simplemem is the unified memory stack that finally gives AI agents a highly efficient, self-evolving long-term memory across text, images, audio, and video. Most memory systems either overwhelm models with massive chat histories or burn through tokens with constant, slow reasoning loops. Instead, this project uses a clever three-stage pipeline to compress interactions into compact, lossy-free facts right when they happen. Even better, it features an autonomous self-evolution loop that constantly diagnoses its own retrieval mistakes and adjusts its search settings over time. It is a brilliant way to make AI assistants remember past context perfectly while slashing token costs.

🆔 @hackernewsgithubprojects

📦 nikhilkain/vyxel-apps

Vyxel Apps

Install and update your favorite open-source Android apps directly from their GitHub releases. This clever platform bypasses traditional app stores by turning GitHub into a completely decentralized, ad-free marketplace. It automatically detects your device's architecture to fetch the correct installation file, and even monitors the background to notify you when new updates are available. You get built-in safety tools like a trust score based on repository activity, and a comparison mode to inspect different apps side by side. It is the perfect way to safely explore community-driven mobile software without any tracking. Give it a spin to customize your mobile experience today.

🆔 @hackernewsgithubprojects

📦 shadcn/improve

Improve

Improve is the codebase advisor that acts like a senior architect, mapping your repository to write flawless implementation specs for cheaper AI models to execute. Instead of burning expensive tokens on basic coding tasks, it analyzes your code, identifies security flaws or performance bottlenecks, and generates self-contained markdown plans. These specs are so detailed, complete with testing boundaries and strict stop conditions, that any lightweight model can run them in an isolated workspace without losing context. It is the perfect way to let your smartest AI do the heavy thinking while leaving the tedious execution to cheap, fast assistants.

🆔 @hackernewsgithubprojects

📦 michalparkola/tapestry-skills

Tapestry Skills

tapestry-skills is the productivity toolkit that finally turns passive learning into immediate action for your AI agent. Instead of just reading articles or watching videos, you just type one command with a link, and your AI automatically strips away the clutter, grabs the core transcript or text, and formats a concrete action plan. It solves the trap of endless consumption by mapping out five progressive milestones you can actually ship. Whether you are stuck on a vague task or drowning in open tabs, it forces you to focus on building rather than planning. Give it a spin and start shipping.

🆔 @hackernewsgithubprojects