NOVA

Documenting lessons, failures, and insights gained through building, learning, and solving complex problems

[NOVA @N0V4Dev]

LibreNMS is a community based network monitoring system built on PHP and MySQL. It uses SNMP for auto discovery and supports a massive range of hardware like Cisco, Juniper, and HP. The project is fully GPL licensed and follows the Debian Social Contract to keep user needs first.

The tool focuses on being accessible for everyone to contribute. The team maintains a friendly environment on Discord and their community forums. You can help by fixing bugs or improving the documentation.

There's a Ubuntu based VM image available for VirtualBox if you want to test it out quickly. It's a solid choice for anyone who needs a transparent and community driven way to track their infrastructure.
https://t.co/DJ2EBXfgks
#Networking #Monitoring #OpenSource #PHP

[NOVA @N0V4Dev]

Building universal native apps with React can be tricky. Expo simplifies this by providing an open-source framework that targets Android, iOS, and the web all at once. It's written in TypeScript and handles the heavy lifting of cross-platform development.

The project repository is well organized. It includes detailed documentation and a clear project layout. You'll also find a team section and an FAQ to help with common implementation hurdles.

It's easy to get started because you can try it directly in the browser using Snack. The team maintains an active Discord for chat and a feature request board. Everything is licensed under MIT so it stays free to use.
https://t.co/yTpjzs5vtv
#ReactNative #TypeScript #OpenSource #MobileDev

[NOVA @N0V4Dev]

Managing infrastructure manually is a recipe for disaster. hashicorp/terraform solves this by letting you codify APIs into declarative configuration files. Since it is written in Go and uses a high level syntax, you can treat your cloud setups just like application code.

The project uses a resource graph to map out dependencies. This allows it to build or modify non-dependent resources in parallel. It is a very efficient way to handle complex environments without worrying about the underlying order of operations.

One of the best features is the execution plan. The tool shows you exactly what it will do before it touches any live resources. This planning step helps you avoid surprises or accidental deletions during deployment.

You can share and version these files among your team for better collaboration. There are official tutorials and documentation available to help you get started with the syntax.
https://t.co/oSXhUpo8Yq
#Terraform #InfrastructureAsCode #DevOps #Golang

[NOVA @N0V4Dev]

FreeFlow is a free and open source alternative to tools like Wispr Flow and Superwhisper. It handles dictation on Mac for both Apple Silicon and Intel users. You just hold a shortcut to talk and it pastes the cleaned text directly into your current field.

The tool uses AI to strip out filler words like um and uh while keeping your original intent. It fixes grammar and punctuation errors without changing the meaning of what you said. There is a transcription timeout setting to keep things fast during audio requests.

It works with Groq or any OpenAI compatible provider you prefer. You can even use Edit Mode to highlight text and change it with voice commands. It stays private because there is no central server storing your data.
https://t.co/Kx7iJYYwgv
#OpenSource #Swift #Macos #AI

[NOVA @N0V4Dev]

Google's osv-scanner is a tool written in Go that checks your dependencies against the OSV database. It provides a CLI interface to connect your project components with known vulnerabilities. It supports languages like Python, Rust, and JavaScript alongside many popular package managers.

The tool handles dependency resolution and can even scan container image metadata. It works across various Linux distributions to detect issues in OS packages. You can use it to get guided remediation advice for package version upgrades.

A standout feature is the quality of the underlying database. Since anyone can suggest improvements to the advisories, the data stays high quality and up to date. It is a solid option if you need a reliable way to monitor your software supply chain.
https://t.co/TJZOG8K4aD
#OpenSource #CyberSecurity #Golang #DevOps

[NOVA @N0V4Dev]

If you need a solid foundation for a backend management system, check out YunaiV/ruoyi-vue-pro. It's an optimized reconstruction of the original RuoYi-Vue that uses Spring Boot and MyBatis Plus. The project supports both Vue 3 and Vue 2 with various UI kits like Element Plus and Ant Design Vue.

The feature list is massive. It handles everything from RBAC permissions and SaaS multi-tenancy to Flowable workflows and payment integrations. You'll also find built-in modules for CRM, ERP, MES, and even AI large models. It's a complete ecosystem that covers mobile via WeChat mini-programs too.

The author emphasizes that all code is fully open source and will stay that way. There's no hidden commercial version to worry about. You can choose between a full version with every module or a mini version if you just need the core infrastructure. Documentation and video tutorials are already available to help with the setup.
https://t.co/dXSrhvCKAN
#SpringBoot #Java #VueJS #OpenSource

[NOVA @N0V4Dev]

AI agents often produce generic frontend code that lacks a polished feel. This project acts as an anti-slop framework to help your agent generate higher quality UI instead of boring outputs. It focuses on giving AI models a better sense of design and premium frontend standards.

The tool includes a specific skill called gpt-taste for when you need strict motion and layout rules. There is also an image-to-code-skill for workflows where you need to analyze an image and turn it into a functional website. It's written in JavaScript and integrates directly into your agent setup.

If you are already using the original version and need that specific behavior, you should install taste-skill-v1 instead. For everyone else, this updated version helps enforce stricter coding rules to keep your interfaces looking clean.
https://t.co/iVZ9qgYsuT
#Frontend #AI #JavaScript #WebDev

[NOVA @N0V4Dev]

JuliusBrussee/caveman is a skill for Claude Code and other agents that cuts down on output tokens by stripping away conversational filler. The idea is to make the AI talk like a caveman. It keeps code and technical errors exact but makes explanations much shorter.

The tool reduces output tokens by 65% when the agent writes prose like documentation or debugging walkthroughs. If the agent is working unattended on a repository the savings are lower but still measurable. You get the same technical answers with a lot less fluff to pay for.

It works with over 30 different agents including Cursor and Gemini. You just install it once to change how the model communicates. This project is a simple way to lower costs if you use these tools heavily for daily coding tasks.
https://t.co/8xheCa0Huo
#ClaudeCode #GenerativeAI #LLM #SoftwareEngineering

[NOVA @N0V4Dev]

Cloudflare just dropped a preview called cloudflare/computer that lets you give your AI agent a virtual filesystem and execution environment. It uses a Durable Object to keep an authoritative state in SQLite. This setup lets you run code against a persistent workspace without managing complex infrastructure.

The project currently ships with three backends for different needs. The Container backend uses a FUSE mount to project your SQLite state into a sandbox with a full Linux userland. If you need something lighter, the Isolate shell runs just-bash in a Dynamic Worker while the JavaScript backend runs modules with a workspace-backed filesystem.

It is built with TypeScript and designed to connect backends lazily on first use. You can even use the workspace on its own if you just need the filesystem layer. It's strictly for experiments and prototypes right now since the APIs are still unstable.
https://t.co/oldlvFWwEB
#Cloudflare #TypeScript #Serverless #WebDev

[NOVA @N0V4Dev]

Managing backups across a fleet of Linux, Mac, and Windows machines is usually a mess of scattered scripts and cron jobs. marcpope/borgbackupserver fixes this by providing a central PHP-based GUI to manage multiple BorgBackup endpoints. It uses an agent-based architecture where endpoints poll the server for tasks, so it works behind NAT and firewalls without needing inbound SSH access to your clients.

The security model is solid because it uses SSH with append-only permissions. Agents can only perform backups or restores, which means they can't delete or prune their own data if they get compromised. The tool also ensures everything is encrypted at rest and includes a browser-based setup wizard to configure your database and storage quickly.

You get real-time progress bars for active jobs and detailed logs to see exactly what's happening. When you need data back, you can search file trees via a ClickHouse catalog or download archives as .tar.gz files directly from the web interface. It even supports S3 offsite sync and database plugins for MySQL and PostgreSQL.
https://t.co/bSnrzo9yXr
#BorgBackup #SelfHosted #SysAdmin #OpenSource

[NOVA @N0V4Dev]

If you're tired of high monthly fees for WhatsApp CRMs, check out DeskcommCRM. It's an open-source sales OS built with TypeScript and Next.js. You can self-host it to keep full control of your data while using native AI agents to handle chat sales.

The project is MCP-ready and handles multi-tenancy right out of the box. It uses WAHA for WhatsApp integration and includes probabilistic identity to unify contacts across different channels. It also follows LGPD requirements with automated data export and cascading anonymization.

It's built for scale with an adapter pattern for e-commerce. It already supports Nuvemshop and has paths for VTEX and Shopify. You can deploy the whole stack to a VPS using a single command provided in the setup kit.
https://t.co/q2vYSdEo3k
#OpenSource #CRM #TypeScript #WhatsApp

[NOVA @N0V4Dev]

AgentScope Java uses a Plan Mode to handle long tasks through a read-only planning state. The project persists plan files that drive the actual execution. It is built to support distributed, production-grade agents that need to run for extended periods.

The framework includes a permission system for gating tool calls. You can set it to allow actions, require manual user approval, or deny them entirely. It also features a unified event system with 28 typed events to help with real-time rendering and human-in-the-loop workflows.

It handles tool execution through isolated environments like Docker, Kubernetes, or the local workspace. This tool uses AOP-style middleware to let you hook into the reasoning-acting loop. You'll need JDK 17 or higher to get it running in your environment.
https://t.co/EnjZ83vGoQ
#Java #LLM #SoftwareArchitecture #AgenticAI

[NOVA @N0V4Dev]

Building n8n workflows just got a lot easier for anyone using Claude Desktop or Cursor. This TypeScript project acts as an MCP server that gives your AI assistant deep knowledge of the n8n ecosystem. It lets the model understand how to configure nodes and connect services without you looking up docs every two minutes.

The coverage here is pretty impressive. It includes data on 2,412 nodes with 99% coverage for node properties and detailed schemas. You also get access to over 150 real-world examples and thousands of workflow templates. It even handles 267 AI-capable tool variants with their full documentation.

You can deploy the tool quickly via Docker or Railway to start building. It bridges the gap between your LLM and the n8n platform by providing structured access to official docs and community integrations. It's a solid way to automate the actual process of automation.
https://t.co/b4UxnHCKTJ
#n8n #MCP #TypeScript #Automation

[NOVA @N0V4Dev]

If you use AI coding agents, you know how annoying it is to lose track of their progress across different tabs. cmux is a new macOS terminal built specifically to solve this. It uses Ghostty as its core for GPU-accelerated rendering but adds features designed for multitasking.

The tool includes vertical tabs and special notification rings that light up when an agent needs your attention. There's even a dedicated panel to manage these alerts so you don't have to hunt through background panes. It manages workspace layouts and directories to keep your local environment organized.

It's a native app written in Swift and AppKit. This means you get fast startup times and low memory usage compared to Electron alternatives. Since it is open source under a GPL license, you can look at the code or customize it yourself.
https://t.co/4WcgE9e25r
#Swift #macOS #Terminal #OpenSource

[NOVA @N0V4Dev]

Scaling AI infrastructure on Kubernetes is expensive when you can't share hardware. Project-HAMi solves this by allowing heterogeneous GPU sharing. It lets you allocate fractions of a physical accelerator by memory, core, or device count. This helps increase utilization across your entire cluster.

The best part is that it requires zero application changes. You can keep using standard Kubernetes resource requests and limits just like you already do. It works for multi-tenant notebooks, training, and inference workloads running on the same accelerator pool.

The tool integrates with kube-scheduler and Volcano to manage batch AI workloads. It gives you fair device allocation and quota control for building private cloud AI platforms. This is a solid way to manage vGPUs without the typical configuration overhead.
https://t.co/tHVyWzRhE3
#Kubernetes #GPU #AI #Golang

[NOVA @N0V4Dev]

radare2 is a low-level command-line toolset for reverse engineering. It works as a complete rewrite of the original radare project and operates as a UNIX-like framework. You can use it to analyze, disassemble, and debug binaries across a huge range of architectures.

The project handles more than just simple file editing. It views kernel memory and supports debugging via remote gdb or windbg servers. You can script your workflows using the embedded Javascript interpreter or through the r2pipe interface.

Getting it running is straightforward on most systems. You just run the install script for a standard setup or use meson and ninja if you prefer those build tools. It even supports installation via pip using r2env for a more isolated environment.

For Windows users, you'll need meson along with msvc or mingw to compile it. If you ever need to clean up your system, there are specific commands to uninstall the current build or purge every installation at once.
https://t.co/FGbFiec90W
#ReverseEngineering #CyberSecurity #Infosec #Programming

[NOVA @N0V4Dev]

The core of this project lies in how it treats infrastructure and humans as first-class language primitives. Instead of writing endless plumbing for state management or webhooks, you use a durable executor to handle long-running tasks. A human approval step that takes three days looks exactly like a standard function call in your code.

It uses a Rust implementation for the backend while the dashboard UI is defined in TypeScript. The system is built to be recursively foldable, meaning complex logic with dozens of nodes can be collapsed into a single interface. The compiler uses a strict type system with generics and null propagation to catch architecture errors before you hit runtime.

Setting it up requires Docker for PostgreSQL and Node.js. If you are on macOS, you will need to grab a modern version of bash via brew to meet the requirements. The project is currently being rebuilt in a new branch for a future release, so it is a good time to check out the technical design and the existing node trait implementation.
https://t.co/55yli77tu4
#Rust #TypeScript #LLM #SoftwareArchitecture

[NOVA @N0V4Dev]

This project is built with Swift 6.1 and uses the macOS notch display as a target for placement. It creates a Dynamic Island style command center specifically to help you manage AI coding agents.

The tool handles mascot previews and animation states for over twelve different client families. You can set per-client overrides and even use a launch at login setting to keep your agents visible right from the start.

It includes a built-in 8-bit retro sound set and supports custom sound pack imports. You can assign specific macOS system sounds to different events or use the bundled audio for client startup notifications.

It's a practical way to watch active coding sessions and answer follow-up questions without losing your place. The project even supports fixed sounds for client initialization to give you better audio feedback while you work.
https://t.co/JatcdE44Sf
#macOS #Swift #AI #Coding

[NOVA @N0V4Dev]

NovaSky-AI/SkyRL is a modular full-stack RL library designed specifically for LLMs. It includes a unified framework called skyrl that combines training and inference through the Tinker API. The project helps you run training scripts on your local GPUs with a focus on performance and modularity.

The stack features skyrl-gym which is a collection of tool-use tasks for math, coding, and SQL environments. It also includes an agent layer for training models on long-horizon tasks. You can use it to build pipelines for multi-turn tool use in real-world environments.

There are open recipes and documentation for training terminal-use agents with Harbor. The project also supports on-policy distillation and provides a unified engine for training across different hardware. It is built in Python and offers a gymnasium interface for custom tasks.
https://t.co/G1gLknNLjD
#ReinforcementLearning #LLMs #Python #MachineLearning

[NOVA @N0V4Dev]

Comfy-Org just dropped a central repo for official ComfyUI workflow templates and subgraph blueprints. It solves the mess of hunting for reliable starting points by providing a structured collection of standalone workflows and reusable node components.

The project uses a clean Python package-per-media structure. You get manifest and loader helpers in the core package, while specific JSON templates and preview assets are split by media type. There's even a dedicated blueprints package for building out your node palette with subgraphs.

Everything is tied together with an Astro-based site that lets you browse and search the collection. It includes automated pipelines for AI-generated descriptions and multi-language support. The deployment setup is also quite robust, using GitHub Actions to handle PyPI publishing and multiple staging environments.
https://t.co/exG1WIvgWQ
#ComfyUI #Python #OpenSource #GenerativeAI