📦 ori-drs/scarf-slam

Build Better 3D Maps with ScaRF-SLAM

Ever wonder how robots build detailed 3D maps while navigating complex environments? Meet ScaRF-SLAM, a powerful framework designed to create high-quality, globally consistent dense reconstructions. It solves the challenge of messy mapping by cleverly decoupling camera tracking from dense reconstruction. By pairing robust, classical SLAM for precise localization with modern geometric foundation models for depth prediction, it ensures your 3D models stay accurate and consistent. Whether you are working with monocular, stereo, or fisheye systems, this tool lets you wrap high-end reconstruction around your existing SLAM setup. Try it out to elevate your robotics projects today.

🆔 @hackernewsgithubprojects

📦 reloops-app/reloops

Stop Managing Files Manually: Meet the AI-Powered Asset Workspace

Are your creative assets scattered across too many folders and apps? Meet Reloops, an open-source creative asset workspace designed for teams and AI agents. It replaces traditional storage with a professional media library where you can organize campaigns, manage versions, and collect approvals all in one place. The best part is its intelligence engine, which automatically generates tags and descriptions for your files using AI. You can even invite AI agents to join your workflow, letting them fetch assets and manage feedback alongside your team. Stop searching and start creating with this powerful, self-hosted, agent-ready asset hub today.

📰 https://news.ycombinator.com/item?id=48377365

🆔 @hackernewsgithubprojects

📦 openclaw/imsg

Control iMessage from Your Terminal

This project provides a command-line interface for Apple Messages on macOS that allows agents and scripts to read, watch, and send iMessages or SMS messages directly from the terminal. By interacting with the local message database and the native automation surface, it enables you to stream live conversations, manage group chats, and automate messaging workflows using stable JSON or JSON-RPC. It is designed to be a local-first tool for developers building integrations, agents, or personal automation pipelines that need reliable access to message data. Whether you need to monitor history or trigger automated replies, it makes your messaging accessible.

🆔 @hackernewsgithubprojects

📦 msalab-pku/perceptiondlm

Speed Up Multi-Region Vision AI with PerceptionDLM

Imagine describing multiple parts of an image all at once instead of one by one. PerceptionDLM is a multimodal diffusion language model designed for efficient parallel region perception. By leveraging parallel decoding, it allows you to generate captions for many masked regions in a single pass, delivering significant speed improvements over traditional sequential methods. It includes a powerful base model and a new benchmark for evaluating caption quality and inference speed. Whether you are working on dense multi-region tasks or need a high-performance diffusion baseline, this open project offers the tools and models to get started.

🆔 @hackernewsgithubprojects

📦 kql11/deep-vrm

Detecting AI Manipulation: Deep Residual Injection Explained

Ever wonder how to catch AI-generated manipulations? This repository presents a breakthrough in forensic signal perception for Multimodal Large Language Models. By introducing a custom residual injection design, it enables models to better perceive subtle forensic signals that might otherwise be missed. It builds on the Qwen2.5-VL architecture, adding a specialized low-level visual encoder that injects fine-grained features directly into the vision encoder. Whether you are interested in model forensics or enhancing multimodal AI perception, this code provides a two-stage training pipeline to help you get started. Explore this project to level up your AI security toolkit today.

🆔 @hackernewsgithubprojects

📦 leeguooooo/chatgpt-imagegen

Generate Images Using Your ChatGPT Subscription CLI

Chatgpt-imagegen is a zero-dependency Python command-line interface that allows you to generate images directly through your existing ChatGPT account, completely bypassing the need for a paid OpenAI API key. By leveraging either your active browser session or the Codex CLI authentication, this tool provides a seamless way to create images from the terminal or integrate them into your favorite AI agents. It effectively repurposes your subscription's built-in image capabilities for local tasks, giving you a powerful, history-managed workflow without extra costs. This project turns your standard chat access into a flexible, command-driven image generation engine.

📰 https://news.ycombinator.com/item?id=48626459

🆔 @hackernewsgithubprojects

📦 dreamlm/dreamreasoner

Meet DreamReasoner-8B: A New Diffusion Reasoning Model

Ever wondered if AI could tackle complex math and coding with a fresh approach? Meet DreamReasoner-8B, an innovative open-source model that changes the game by using block diffusion for reasoning. By applying block-size curriculum learning, this model adapts the foundation of Qwen3-8B into a powerhouse capable of handling challenging mathematical problems and code generation tasks. Its performance is truly impressive, standing shoulder to shoulder with specialized thinking models. Whether you are building complex applications or exploring new AI architectures, this project offers a powerful tool for your next breakthrough. Check it out and start experimenting with better reasoning today.

🆔 @hackernewsgithubprojects

📦 foxted/rsc-boundary

Visualize React Server Components Boundaries

Next.js 16 projects and TanStack Start applications can now easily visualize the split between server-rendered regions and client-side subtrees using the RSC Boundary devtool. This lightweight tool automatically maps component boundaries directly in your browser, using orange outlines for client components and blue for server regions, along with a detailed panel to identify component provenance. It helps developers spot accidental boundaries and nested server islands without manually checking every file for directives. By providing clear visual feedback during development, this project simplifies the complex RSC mental model and allows you to build with greater confidence.

📰 https://news.ycombinator.com/item?id=47668297

🆔 @hackernewsgithubprojects

📦 eclaire-labs/eclaire

Eclaire: The Private, Self-Hosted AI Assistant for Your Data

Eclaire is an open-source, local-first AI assistant designed to unify your digital life by connecting tasks, notes, documents, photos, and bookmarks into one private, self-hosted space. It solves the issue of relying on closed ecosystems by running entirely on your own hardware, using local models for tasks like OCR, content summarization, and search. Whether you are automating workflows or chatting with your stored information, it provides a secure alternative that keeps your sensitive data under your control. By leveraging modular architecture and local hardware acceleration, Eclaire offers a powerful, extensible way to manage your personal data independently.

🆔 @hackernewsgithubprojects

📦 rikyz90/shibaclaw

Run a Private, Secure AI Agent on Your Own Machine

Manage your own private AI assistant with a self-hosted, security-focused agent that keeps your data local and secure. This platform provides a native web interface and desktop application, allowing you to connect to over twenty different AI providers while maintaining strict privacy. It features a sophisticated three-level memory system for long-term operational continuity, alongside built-in automation, task scheduling, and support for the Model Context Protocol. With robust defenses like install-time package auditing and prompt-injection wrapping, you can automate complex tasks reliably. Take full control of your digital workspace today with this versatile and deeply integrated AI agent.

🆔 @hackernewsgithubprojects

📦 headroomlabs-ai/headroom

Slash Your LLM Token Costs with Headroom

Headroom is an open-source tool designed to reduce token consumption by sixty to ninety-five percent for your large language model workflows. By compressing tool outputs, system logs, file data, and retrieved information before they reach the model, it maintains the same quality of answers while significantly lowering operational costs. This project functions as a library, a proxy, and an Model Context Protocol server to integrate seamlessly into existing pipelines. It helps developers manage context windows more efficiently and save money on expensive API calls. Whether you are scaling RAG applications or managing complex logs, this tool provides essential optimization.

📰 https://news.ycombinator.com/item?id=48627072

🆔 @hackernewsgithubprojects

📦 simplifaisoul/osiris

OSIRIS: The Open Source Global Intelligence Dashboard

Osiris functions as a production-grade intelligence platform that aggregates global real-time data into a single GPU-accelerated interface. Built using Next.js and MapLibre, it visualizes diverse information layers including live flight tracking, maritime activity, earthquake monitoring, and public CCTV feeds. The platform features a specialized reconnaissance toolkit capable of port scanning, DNS resolution, and cryptocurrency wallet tracing, which automatically cross-references data against international sanctions lists. By processing vast amounts of geographic and cyber-threat data with high-performance WebGL rendering, this open-source tool provides users with comprehensive situational awareness. Explore global events through this technical dashboard to monitor and analyze complex intelligence feeds.

🆔 @hackernewsgithubprojects

📦 raiyanyahya/recall

Give Claude Code Permanent Memory Offline

Recall provides Claude Code with durable, fully-local memory by automatically logging your sessions and condensing them into a concise summary. It solves the cold-start problem by tracking your project progress, goals, and open threads without needing external API calls or extra tokens. Using a built-in Python summarizer based on TF-IDF and TextRank, it creates a compact context file on your machine that you can load into new sessions. This keeps your workflow efficient and private while saving your subscription credits. Simply install the plugin to stop re-explaining your project and let your code history work for you.

📰 https://news.ycombinator.com/item?id=48622590

🆔 @hackernewsgithubprojects

📦 corsairdev/corsair

Build Powerful AI Agents with a Unified Integration Layer

Connect your AI agents to a wide range of external tools and services through a single, streamlined integration layer. This project simplifies the complexity of managing multiple connections by providing pre-built adapters for popular platforms like Slack, GitHub, Gmail, and Notion. It handles essential tasks such as authentication, webhook management, and database interactions, allowing developers to focus on building agent capabilities rather than maintaining fragile API integrations. Whether you are automating workflows or creating intelligent assistants, this infrastructure ensures your agents communicate reliably across your entire tech stack. Experience a cleaner, more efficient way to manage your integrations.

🆔 @hackernewsgithubprojects

📦 guohuiyuan/go-music-dl

Go Music DL: The Ultimate Multi-Platform Music Tool

Go Music DL is a powerful tool built in Go that aggregates music search and download capabilities across more than ten major platforms including QQ, NetEase, and Bilibili. This versatile project functions as a CLI tool, a desktop application, and a local web server, allowing users to search, stream, and batch download music with ease. It supports high-fidelity audio parsing, lyrics, and even manages local music libraries with metadata tagging. Whether you prefer a terminal interface or a graphical dashboard, this tool streamlines music collection and management into one unified, efficient experience for any platform.

🆔 @hackernewsgithubprojects

📦 slothflowlabs/duckle

Build Data Pipelines Instantly with Duckle

Design and run data pipelines locally by dragging and dropping nodes onto a visual canvas. This desktop application compiles your visual workflows directly into SQL, executing them at high speed through the DuckDB engine. You can effortlessly integrate hundreds of sources and sinks, from SQL databases and lakehouses to streaming brokers and SaaS APIs, without needing any servers or cloud infrastructure. It even includes a local AI assistant that helps you build pipelines by simply describing what you need in plain English. Your workspaces stay as simple, git-friendly files on your own machine, putting you back in control.

📰 https://news.ycombinator.com/item?id=48628614

🆔 @hackernewsgithubprojects

📦 layr-labs/d-inference

Darkbloom: Turn Your Idle Mac into a Private AI Cloud

Darkbloom transforms idle Apple Silicon Macs into a secure, decentralized network for running artificial intelligence models. It enables users to tap into high-performance compute power while ensuring privacy through end-to-end encryption and strict hardware attestation. The system uses a Go-based coordinator and a hardened Swift CLI to process requests directly on the Mac GPU via MLX, keeping data encrypted even from the machine owner. It supports standard OpenAI and Anthropic API formats, allowing developers to integrate it easily. Whether you want to monetize your idle hardware or build a private inference pipeline, this project demonstrates a highly secure model.

🆔 @hackernewsgithubprojects

📦 akmessi/vex

AI Video Editing Straight From Your Terminal

Edit videos using plain English commands directly in your terminal with this powerful AI agent. By simply describing the changes you want, the software handles complex tasks like trimming footage, removing silent gaps, or color grading your clips using intelligent automation. It safely edits a working copy of your video while maintaining a full project history, allowing you to undo or rebuild edits anytime. You can even generate custom B-roll and visual animations to match your narration automatically. Streamline your entire creative workflow by letting this terminal-native tool manage the technical heavy lifting for you.

🆔 @hackernewsgithubprojects

📦 paytonjjones/bsharp

BSharp: A Digital Tool for Teaching Absolute Pitch

Young children possess a unique neurological window for acquiring absolute pitch that closes around age six. BSharp is an open-source tool built to help children leverage this window through the Eguchi chord identification method. The software functions by playing specific piano chords that children must identify by associated colors. It features an adaptive weighting algorithm to prioritize harder chords and includes progress tracking for multiple users. By practicing briefly throughout the day, children learn to recognize these musical intervals reliably. This project provides a practical, offline-capable digital implementation for families interested in early childhood auditory development.

📰 https://news.ycombinator.com/item?id=48618488

🆔 @hackernewsgithubprojects

📦 spiritbuun/buun-llama-cpp

Boost Your LLM Context: Turboquant Explained

This specialized fork of llama.cpp increases available model context by two to three times without requiring additional video memory. By implementing Trellis-Coded Quantization, the software efficiently compresses the key-value cache while maintaining performance levels that often exceed standard uncompressed formats. It achieves this by using a five hundred twelve state trellis to optimize codeword assignments, enabling high-quality inference at very low bit rates. This project is particularly useful for users running large models on limited hardware, offering a practical way to fit significantly longer text windows into existing GPU setups. Explore this implementation to optimize your local model capacity.

🆔 @hackernewsgithubprojects