Preset
Background
Text
Font
Size
Width
Account Tuesday, April 21, 2026

The Git Times

“Technology is not a neutral tool. It is a system that carries its own values.” — Ursula Franklin

AI Models
Claude Sonnet 4.6 $15/M —GPT-5.4 $15/M —Gemini 3.1 Pro $12/M —Grok 4.20 $6/M —DeepSeek V3.2 $0.89/M —Llama 4 Maverick $0.60/M —
Full Markets →

Kami Imposes Editorial Discipline on AI Documents 🔗

Design skill for Claude and Codex agents enforces warm parchment aesthetic across professional formats

tw93/Kami · HTML · 416 stars 0d old · Latest: V1.0.0

Kami equips AI coding assistants with a strict design system that turns natural-language requests into print-ready documents. Instead of generic corporate templates or flashy gradients, the skill applies eight invariant rules: a warm parchment canvas, a single ink-blue accent, serif-led hierarchy, and editorial whitespace tuned for paper.

The system supports six core formats: one-pagers, long reports, formal letters, portfolios, resumes and slide decks.

Kami equips AI coding assistants with a strict design system that turns natural-language requests into print-ready documents. Instead of generic corporate templates or flashy gradients, the skill applies eight invariant rules: a warm parchment canvas, a single ink-blue accent, serif-led hierarchy, and editorial whitespace tuned for paper.

The system supports six core formats: one-pagers, long reports, formal letters, portfolios, resumes and slide decks. Each follows the same visual logic so outputs feel like they came from the same careful hand. Three inline SVG primitives—architecture diagrams, flowcharts and quadrant charts—embed cleanly without breaking the aesthetic.

Installation is a one-line command. npx skills add tw93/kami registers the skill with Claude Code, Codex or generic agents that read from the skills directory. Once loaded, users simply describe the deliverable and the model generates matching HTML that respects every constraint: Newsreader serif at weight 500, no bold or italic, precise spacing, no shadows.

The project matters because AI excels at content yet routinely fails at presentation. Kami closes that gap by codifying taste. Documents emerge ready for immediate professional use, whether printed or shared as PDF, without requiring designers or design arguments.

Kami is not a UI framework or theme. It is an aesthetic constraint system that lets the model focus on substance while the skill guarantees form.

Use Cases
  • Founder requests polished startup one-pager from AI
  • Engineer builds technical resume under fixed design rules
  • Speaker generates keynote slides with editorial spacing
Similar Projects
  • pandoc - converts markup but lacks curated aesthetic rules
  • Typst - offers programmatic typesetting without AI integration
  • Quarto - supports publishing yet requires manual configuration

More Stories

QwenPaw Deploys Self-Hosted AI Agents Across Platforms 🔗

Python framework gives users full control of memory, skills and multi-agent collaboration

agentscope-ai/QwenPaw · Python · 15.7k stars 1mo old

QwenPaw is a Python-based system for running personal AI assistants on local machines or private cloud servers. It connects to messaging platforms including Discord, Telegram, DingTalk and Feishu without sending data to third-party hosts.

The framework maintains complete user control over memory and personalization.

QwenPaw is a Python-based system for running personal AI assistants on local machines or private cloud servers. It connects to messaging platforms including Discord, Telegram, DingTalk and Feishu without sending data to third-party hosts.

The framework maintains complete user control over memory and personalization. Skills determine capabilities: built-in modules handle PDF and Office document processing, news digestion, scheduling, and local file organization. Custom skills load automatically with no vendor lock-in.

Multi-agent collaboration is central. Users create independent agents, assign roles, and enable communication skills so agents can divide complex tasks. Version 1.1.2 adds Mission Mode (/mission), which lets agents autonomously plan, execute, iterate and self-correct across multiple phases. The release also introduces ACP protocol support for delegating work to external coding agents with permission guards and streamed output, plus the qwenpaw doctor command for environment diagnostics and remediation.

Security runs at multiple layers: tool guards, file access controls and skill security scanning. The agent runtime and LLM tooling make it straightforward to combine scheduled tasks with custom skills into persistent automation.

The project matters because it moves agentic workflows out of remote services and onto infrastructure users control, while keeping extensibility and operational safety intact.

Use Cases
  • Researchers track AI news and query personal knowledge bases
  • Professionals receive automated email digests in enterprise chat
  • Developers generate prototypes from overnight natural-language goals
Similar Projects
  • CrewAI - offers agent orchestration but requires more external API calls
  • AutoGen - supports multi-agent dialogue yet lacks native multi-channel connectors
  • LangGraph - builds agent workflows with weaker emphasis on local security controls

Browser Extension Turns Web Pages Into Clean Markdown 🔗

One-click tool uses Readability library to strip noise and deliver LLM-ready structured text

Ademking/MD-This-Page · TypeScript · 367 stars 4d old

The MD This Page extension converts any webpage into clean Markdown using a right-click context menu or Alt+M keyboard shortcut. It opens a dedicated preview tab where users view, refine, and export the result.

The extension relies on Mozilla's Readability library to detect and extract the primary article content while discarding navigation, advertisements, scripts, and boilerplate HTML.

The MD This Page extension converts any webpage into clean Markdown using a right-click context menu or Alt+M keyboard shortcut. It opens a dedicated preview tab where users view, refine, and export the result.

The extension relies on Mozilla's Readability library to detect and extract the primary article content while discarding navigation, advertisements, scripts, and boilerplate HTML. Output preserves headings, lists, paragraphs, and logical hierarchy without deeply nested DOM artifacts.

This approach addresses a practical limitation in AI workflows. Large language models demonstrate measurably better performance on clean Markdown than on raw HTML or cluttered web extracts. Markdown reduces token usage, maintains semantic structure, and minimizes parsing errors that occur with inconsistent DOM trees.

Core capabilities include:

  • Toggle controls for images, hyperlinks, metadata (title, author, date), source URL, and generated document maps
  • Three export methods: copy to clipboard, download as .md file, or copy as pre-formatted AI prompt

Version 1.0.0 is available for both Chrome and Firefox. By turning noisy web pages into compact, structured documents, the tool improves context efficiency and reasoning quality for developers and researchers who regularly feed web-sourced material to LLMs.

Use Cases
  • AI engineers clean web articles for prompt engineering pipelines
  • Researchers extract structured content for model training datasets
  • Technical writers convert documentation pages into repository Markdown
Similar Projects
  • MarkDownload - offers basic one-click conversion but lacks preview tab and AI-specific export options
  • Turndown - JavaScript library requiring manual integration instead of a ready browser extension
  • SingleFile - saves complete HTML snapshots rather than producing token-efficient Markdown

VoxCPM2 Refines Tokenizer-Free TTS Voice Controls 🔗

Version 2.0.2 update sharpens cloning precision and vLLM serving across 30 languages

OpenBMB/VoxCPM · Python · 15.2k stars 7mo old

The OpenBMB team shipped VoxCPM 2.0.2 this spring, tightening controls and deployment efficiency for its diffusion autoregressive TTS model.

The OpenBMB team shipped VoxCPM 2.0.2 this spring, tightening controls and deployment efficiency for its diffusion autoregressive TTS model. Unlike token-based systems, VoxCPM2 directly generates continuous speech representations from text, eliminating discretization artifacts and producing more natural prosody.

The 2B parameter model, built on a MiniCPM-4 backbone and trained on over 2 million hours of speech, now delivers several production-focused improvements. It supports 30 languages without requiring explicit tags, automatically inferring appropriate rhythm and emphasis from content alone.

Key technical advances in the current release include Voice Design, which creates entirely new speakers from plain-language prompts specifying gender, age, tone, pace and emotion. Controllable Cloning lets users supply a short reference clip plus optional style guidance to steer expression while preserving timbre. Ultimate Cloning accepts both reference audio and transcript, enabling seamless continuation that replicates every nuance at 48kHz.

The integrated AudioVAE V2 performs asymmetric encode/decode with built-in super-resolution, removing the need for separate upsamplers. Real-time factor sits at ~0.3 on an RTX 4090 and drops to ~0.13 with vLLM-Omni or Nano-vLLM acceleration. Official omni-modal serving support simplifies integration for streaming applications.

These changes matter now because they reduce external dependencies and give developers precise, text-driven voice manipulation at studio quality. The architecture’s avoidance of discrete tokens yields measurably smoother transitions and fewer artifacts in long-form synthesis.

**

Use Cases
  • Game studios designing unique NPC voices via text prompts
  • Accessibility teams building personalized speech interfaces for users
  • Podcast producers generating expressive multilingual episodes from scripts
Similar Projects
  • CosyVoice - uses discrete tokens where VoxCPM2 generates continuous representations
  • VALL-E - offers cloning but lacks native text-based voice design
  • Tortoise-TTS - provides zero-shot cloning yet without 48kHz output or vLLM serving

KillerPDF Brings Portable PDF Editing to Windows 🔗

Self-contained tool enables local annotation, merging and text editing without subscriptions or telemetry

SteveTheKiller/KillerPDF · C# · 528 stars 5d old

KillerPDF is a portable PDF editor that runs as a single executable roughly 6 MB when zipped. Built in C# with WPF, it targets .NET Framework 4.

KillerPDF is a portable PDF editor that runs as a single executable roughly 6 MB when zipped. Built in C# with WPF, it targets .NET Framework 4.8 — already present on every supported Windows 10 and 11 system — and requires no installer, no runtime download, no account and no configuration files.

The application uses PDFium through the Docnet.Core library for high-quality rendering. Users can merge multiple PDFs, split out selected pages, and reorder them with drag-and-drop. It supports inline text editing that attempts to match the original document’s fonts, plus text boxes, freehand drawing, highlight overlays with adjustable color, size and opacity, and reusable signatures that can be drawn once and placed anywhere.

Additional functions include full-text search with result highlighting, drag-to-copy selection, and printing that flattens annotations into the output. The project is licensed GPLv3, with both the prebuilt binary and corresponding source provided for every release.

Its developer created KillerPDF after repeated frustration with Adobe Acrobat’s size, subscription model and telemetry. The goal was a minimal, local-only tool — “the PDF equivalent of Notepad.” Version 1.1.1, released this week, fixed maximize behavior across multiple monitors via a WM_GETMINMAXINFO hook and cleared two nullability warnings.

The entire application can be rebuilt with a single dotnet publish command using the .NET 8 SDK; Costura bundles everything into one EXE.

Use Cases
  • Field technicians annotate reports on-site without internet
  • Administrators merge documents and reorder pages locally
  • Professionals add reusable signatures to forms offline
Similar Projects
  • PDFsam Basic - open-source merge/split tool but lacks inline text editing
  • LibreOffice Draw - handles PDF export within larger suite requiring full install
  • SumatraPDF - fast lightweight viewer but offers no annotation or editing features

Twenty v2.0 Refines CRM Layout and Deployment Tools 🔗

Enhanced page controls and EKS support strengthen open-source alternative for technical teams

twentyhq/twenty · TypeScript · 44.8k stars Est. 2022

Twenty has shipped v2.0.0, introducing concrete upgrades to its code-first CRM that let engineering teams treat customer data systems like the rest of their stack.

Twenty has shipped v2.0.0, introducing concrete upgrades to its code-first CRM that let engineering teams treat customer data systems like the rest of their stack. The platform supplies modular primitives—objects, fields, views, workflows, and agents—that developers define directly in TypeScript and ship through a CLI.

Recent changes focus on practical editing friction. Page layout tools now support reset-to-default actions, tab icon pickers, and visible deactivated tabs. Studio improvements add SVG export and refined halftone controls. A new Docker target bundles AWS CLI for EKS deployments, easing operations for teams running the system on Kubernetes.

The workflow is deliberate: scaffold an app with npx create-twenty-app, declare a deal object with typed fields for name, amount, and close date, then run npx twenty deploy. Everything lives in version control. Backend services built on NestJS, BullMQ, PostgreSQL, and Redis pair with a React frontend in an Nx monorepo.

These updates matter as organizations outgrow rigid SaaS CRMs. Twenty gives engineering-led companies the ability to version, test, and iterate their sales infrastructure at the same cadence as product code, while keeping AI agents and logic functions inside the same extensible environment.

(178 words)

Use Cases
  • Engineering teams defining custom objects and fields in TypeScript
  • DevOps engineers self-hosting CRM on AWS EKS with Docker
  • Sales teams building version-controlled workflows and AI agents
Similar Projects
  • Odoo - broader ERP with traditional modules instead of code-first schemas
  • SuiteCRM - conventional open-source CRM lacking native TypeScript SDK
  • EspoCRM - self-hosted CRM focused on UI extensions over infrastructure-as-code

Open Source Codifies Reusable Skills to Supercharge AI Agents 🔗

Community distills expertise into modular components that let coding agents compose sophisticated workflows across domains.

Trendai-agents
alchaincyf/huashu-designkangarooking/cangjie-skillagentscope-ai/QwenPawdust-tt/dustVoltAgent/awesome-design-md+32 more

The open source ecosystem is coalescing around a powerful new primitive: agent skills. Rather than treating large language models as blank slates that require fresh prompting for every task, developers are packaging domain knowledge, workflows, tools, and evaluation criteria into standardized, executable modules that agents can discover, invoke, and compound.

This pattern is visible across dozens of recent projects.

The open source ecosystem is coalescing around a powerful new primitive: agent skills. Rather than treating large language models as blank slates that require fresh prompting for every task, developers are packaging domain knowledge, workflows, tools, and evaluation criteria into standardized, executable modules that agents can discover, invoke, and compound.

This pattern is visible across dozens of recent projects. alirezarezvani/claude-skills ships more than 232 plugins spanning engineering, marketing, compliance, and C-level advisory. addyosmani/agent-skills focuses on production-grade engineering practices, while coreyhaines31/marketingskills encodes CRO, copywriting, SEO, and growth loops. On the design front, alchaincyf/huashu-design delivers HTML-native high-fidelity prototyping, slide decks, animations, and even MP4 export directly inside Claude Code. VoltAgent/awesome-design-md takes the idea further by collecting real-world DESIGN.md files so agents can replicate existing visual systems without manual specification.

The technical shift is significant. Skills are typically expressed as structured YAML, tool schemas, example trajectories, and lightweight scripts that agents can interpret at runtime. kangarooking/cangjie-skill demonstrates how to distill an entire book into such a runnable skill set. Memory and context layers are also being formalized: thedotmack/claude-mem automatically captures, compresses, and re-injects session history, while mksglu/context-mode sandboxes tool output to cut context-window usage by up to 98 %.

Frameworks that consume these skills are proliferating in parallel. openai/openai-agents-python supplies a lightweight backbone for multi-agent orchestration. multica-ai/multica adds task assignment, progress tracking, and skill compounding so agents behave like genuine teammates. dust-tt/dust and vercel-labs/open-agents provide full platforms for building and deploying custom agent teams. Domain-specific expressions include Donchitos/Claude-Code-Game-Studios (49 agents, 72 workflow skills, studio-style hierarchy), SWE-agent/SWE-agent for autonomous GitHub issue resolution, and HKUDS/CLI-Anything which seeks to make every command-line tool natively agent-friendly.

Collectively these repositories reveal where open source is heading: from prompt hacking toward skill engineering. Just as npm standardized code reuse, a emerging skill layer standardizes behavior reuse. Agents are no longer one-off scripts but composable ensembles that inherit best practices from a growing public registry. The long-term implication is an order-of-magnitude increase in agent reliability and specialization, moving AI-assisted development from novelty to repeatable engineering discipline.

Use Cases
  • Engineers equipping coding agents with domain-specific skills
  • Teams orchestrating multi-agent platforms for complex workflows
  • Educators distilling curricula into personalized learning agents
Similar Projects
  • obra/superpowers - Delivers a broader agentic methodology that complements the modular skill libraries
  • HKUDS/DeepTutor - Focuses on agent-native tutoring, extending the skills pattern into education
  • microsoft/ai-agents-for-beginners - Provides structured lessons that teach the same skill-first agent composition

AI Agents Supercharge the Evolution of Web Frameworks 🔗

Open source projects fuse traditional web tooling with intelligent agents that understand design, generate code, and unify model access across domains.

Trendweb-frameworks
Ademking/MD-This-Pagemedusajs/medusaairbytehq/airbytefreeCodeCamp/freeCodeCampAzure/azure-sdk-for-net+22 more

An emerging pattern is reshaping open source web development: the fusion of conventional web frameworks with AI coding agents that treat design systems, APIs, and UI patterns as first-class, machine-readable inputs. Rather than单纯 libraries for routing or rendering, these projects form agent-native ecosystems where large language models actively participate in the development loop.

Evidence appears across the cluster.

An emerging pattern is reshaping open source web development: the fusion of conventional web frameworks with AI coding agents that treat design systems, APIs, and UI patterns as first-class, machine-readable inputs. Rather than单纯 libraries for routing or rendering, these projects form agent-native ecosystems where large language models actively participate in the development loop.

Evidence appears across the cluster. VoltAgent/awesome-design-md collects DESIGN.md files from popular sites so coding agents can drop production-grade design systems into new projects and generate matching UI without human pixel wrangling. Leonxlnx/taste-skill explicitly trains agents to reject "boring, generic slop," raising the aesthetic baseline of AI-generated frontends. On the tooling side, Gitlawb/openclaude and badlogic/pi-mono ship unified LLM APIs, TUI interfaces, and CLI agents that speak natively to OpenAI, Gemini, Ollama, and 200+ compatible endpoints.

Domain-specific platforms illustrate the breadth. medusajs/medusa delivers a headless commerce engine while twentyhq/twenty offers an AI-ready Salesforce alternative, both built in TypeScript and designed for extensibility by both humans and agents. gohugoio/hugo remains the benchmark for lightning-fast static sites, and phaserjs/phaser shows the same performance-first philosophy applied to Canvas and WebGL game frameworks. Lower-level components such as karlseguin/http.zig and projectdiscovery/httpx provide the fast, retry-capable HTTP primitives these agent-driven stacks require.

Interoperability layers complete the picture. OpenAPITools/openapi-generator auto-produces clients and server stubs from any spec, while QuantumNous/new-api, mnfst/awesome-free-llm-apis, and Wei-Shaw/sub2api create unified gateways that abstract multiple LLM providers. airbytehq/airbyte ensures data flows cleanly into these AI-augmented applications, and Ademking/MD-This-Page turns arbitrary web content into clean Markdown training fuel.

Collectively, the cluster reveals where open source is heading: toward agent-first web stacks. Technically this means standardized function-calling interfaces, parseable design artifacts, cross-model routing layers, and a shift from framework-as-library to framework-as-environment that an AI can explore, modify, and extend autonomously. The human role moves from writing boilerplate to curating taste, verifying outputs, and defining high-level constraints.

This pattern lowers the cost of sophisticated web applications while raising their quality floor, pointing to a future in which autonomous agents and human developers collaborate inside richly instrumented open frameworks.

Use Cases
  • Engineers training AI agents to generate production-grade web UIs
  • Teams building headless commerce platforms with unified LLM backends
  • Developers auto-generating API clients from OpenAPI specifications
Similar Projects
  • Next.js - Delivers React-based full-stack primitives that AI coding agents can easily introspect and extend
  • Strapi - Provides headless CMS capabilities that complement the agent-native design systems in this cluster
  • FastAPI - Offers Python web APIs with automatic OpenAPI generation that integrates seamlessly with the LLM unification tools shown here

Open Source Forges Tools for the Age of AI Coding Agents 🔗

From token-efficient proxies to autonomous bug fixers and knowledge graphs, developers are building infrastructure that lets LLMs act as independent software engineers

Trenddev-tools
Ademking/MD-This-Pagemarkqvist/Sidebandremix-project-org/remix-projectOpenAPITools/openapi-generatorFincept-Corporation/FinceptTerminal+21 more

An emerging pattern is reshaping open-source dev-tools: the rapid construction of agent-native infrastructure that treats large language models not as assistants but as primary users. Rather than simply wrapping existing AI APIs, these projects create specialized interfaces, data formats, and orchestration layers optimized for autonomous coding agents.

The evidence spans multiple technical fronts.

An emerging pattern is reshaping open-source dev-tools: the rapid construction of agent-native infrastructure that treats large language models not as assistants but as primary users. Rather than simply wrapping existing AI APIs, these projects create specialized interfaces, data formats, and orchestration layers optimized for autonomous coding agents.

The evidence spans multiple technical fronts. SWE-agent takes a GitHub issue and uses an LM to explore a repository, edit code, and submit fixes without human intervention. rtk-ai/rtk attacks the economics of agentic development head-on, acting as a CLI proxy that strips redundant context and reduces token consumption by 60-90% on common dev commands. safishamsi/graphify converts entire folders of code, documentation, papers, and images into queryable knowledge graphs, giving agents structured semantic memory instead of brittle context windows.

A parallel ecosystem of agent interfaces has appeared. Gitlawb/openclaude, farion1231/cc-switch, and router-for-me/CLIProxyAPI wrap Claude Code, Gemini CLI, Codex, and other models behind unified, OpenAI-compatible endpoints or desktop switches. HKUDS/CLI-Anything pushes the boundary further with its stated goal of making “ALL Software Agent-Native,” while alirezarezvani/claude-skills and kepano/obsidian-skills ship hundreds of reusable skills that teach agents to manipulate Markdown, JSON Canvas, Bases, and CLI tools reliably.

Supporting projects reinforce the pattern. ChromeDevTools/chrome-devtools-mcp exposes browser debugging capabilities directly to coding agents. paperclipai/paperclip supplies orchestration primitives for “zero-human companies,” and badlogic/pi-mono packages coding-agent CLIs, unified LLM APIs, TUIs, and vLLM infrastructure into a single toolkit. Even established tools like OpenAPITools/openapi-generator and remix-project-org/remix-project gain new relevance: they produce machine-readable SDKs and browser-based IDEs that agents can more easily consume and extend.

Collectively, this cluster signals where open source is heading. The next generation of dev tools will be judged by how legible they are to LLMs, how efficiently they manage context and cost, and how cleanly they expose capabilities through CLIs, graphs, and skill plugins rather than human-first GUIs. The boundary between “developer environment” and “agent operating system” is dissolving.

Technically, the pattern emphasizes deterministic outputs, structured memory, cost-aware routing, and composable skills—foundations required for agents to move from toy demonstrations to production engineering workflows.

Use Cases
  • Engineers equipping agents to autonomously resolve GitHub issues
  • Developers optimizing token usage across LLM-powered CLI workflows
  • Teams converting codebases into queryable knowledge graphs for AI
Similar Projects
  • LangChain - Provides general agent orchestration but lacks the specialized coding CLI, token proxies, and dev-specific skills seen here
  • Auto-GPT - Early autonomous agent framework that these projects surpass with tighter integration into existing dev tools and IDEs
  • Aider - CLI for LLM pair programming that this cluster extends into full agent-native infrastructure and multi-model routing layers

Deep Cuts

Turn Books Into Executable AI Agent Skills 🔗

Cangjie Skill revolutionizes how developers transform written knowledge into reusable AI agent capabilities

kangarooking/cangjie-skill · Unknown · 431 stars

Hidden among GitHub’s lesser-known corners, cangjie-skill offers a genuinely new primitive for the agent era. The project distills any book into a library of executable agent skills — modular, prompt-driven components that capture not just facts but decision frameworks, procedures, and heuristics.

Using clever prompt-engineering templates and a custom skill-generator, it parses long-form content, extracts core mental models, then packages them as reusable building blocks.

Hidden among GitHub’s lesser-known corners, cangjie-skill offers a genuinely new primitive for the agent era. The project distills any book into a library of executable agent skills — modular, prompt-driven components that capture not just facts but decision frameworks, procedures, and heuristics.

Using clever prompt-engineering templates and a custom skill-generator, it parses long-form content, extracts core mental models, then packages them as reusable building blocks. These skills can be chained, triggered, or composed inside larger agent-workflows, turning static knowledge into living automation.

The real power lies in moving beyond retrieval. Instead of an agent searching a vector database and hoping for relevant chunks, cangjie-skill creates native capabilities the model can invoke with confidence. The resulting skills remain editable, versionable, and composable, giving builders fine-grained control over how book wisdom is applied.

For anyone shipping specialized agents, this changes the economics of knowledge injection. Legal texts become compliance agents, technical manuals become autonomous debuggers, and strategy books become decision engines. What once required weeks of prompt tuning now emerges from an automated knowledge-distillation pipeline.

In a landscape crowded with wrappers and chat interfaces, cangjie-skill stands out by asking a deeper question: what if every great book could be refactored into code-like intelligence?

Use Cases
  • AI engineers converting technical manuals into autonomous debugging agents
  • Educators transforming textbooks into adaptive tutoring skill libraries
  • Consultants distilling strategy books into executive decision agents
Similar Projects
  • LlamaIndex - indexes books for retrieval but stops short of executable skill generation
  • LangChain - excels at chaining agents yet requires manual prompt crafting from source material
  • AutoGen - focuses on multi-agent conversation but lacks automated book-to-skill distillation

Quick Hits

advisor-ledger Build robust advisor systems with this Python ledger that tracks client interactions, recommendations, and compliance with full auditability and customization. 404
freeCodeCamp Extend freeCodeCamp's open-source codebase and interactive curriculum to deliver free hands-on learning in math, programming, and computer science. 443.3k
medusa Create fully custom e-commerce experiences with Medusa's modular commerce platform that gives developers complete control over every layer. 32.7k
odoo Integrate modular open-source apps for ERP, CRM, inventory, and accounting to build scalable business management systems that adapt to your needs. 50.2k
hop 399

Microsoft Updates Generative AI for Beginners to Version 3 🔗

The 21-lesson curriculum now incorporates latest techniques in prompt engineering, semantic search, and responsible model deployment

microsoft/generative-ai-for-beginners · Jupyter Notebook · 109.6k stars Est. 2023

Microsoft has refreshed its generative-ai-for-beginners repository with Version 3, aligning the curriculum with current production realities of large language models and multimodal systems. The project delivers 21 self-contained lessons that teach developers how to move from foundational concepts to building functional generative AI applications.

The course is structured so each lesson stands alone.

Microsoft has refreshed its generative-ai-for-beginners repository with Version 3, aligning the curriculum with current production realities of large language models and multimodal systems. The project delivers 21 self-contained lessons that teach developers how to move from foundational concepts to building functional generative AI applications.

The course is structured so each lesson stands alone. Builders can start with prompt engineering, dive into transformer mechanics, or explore semantic search without sequential prerequisites. All content lives in Jupyter Notebooks that interweave explanation, runnable code, and exercises using Azure OpenAI services, the OpenAI Python SDK, and supporting Azure infrastructure.

Core technical material covers how transformer architectures process tokens through attention mechanisms, how embeddings power semantic search, and how to construct retrieval-augmented generation pipelines that combine vector stores with GPT models. Lessons demonstrate practical implementations: creating conversational agents that maintain context across long sessions, generating and iterating on images with DALL-E, and building search systems that replace keyword matching with cosine similarity on high-dimensional vectors.

Version 3 updates emphasize production concerns that have gained urgency in the past year. Notebooks now address token budget optimization, implementing content filters and guardrails, evaluating output quality with structured metrics, and orchestrating multi-step reasoning chains. Azure-specific guidance shows how to deploy these capabilities securely within enterprise environments, including private endpoints, managed identity authentication, and cost controls for high-volume usage.

The course solves a precise problem for builders: the gap between consuming AI APIs and engineering robust systems. While documentation from providers explains individual calls, few resources connect the dots between model behavior, prompt design, data retrieval, and responsible deployment. By combining theory and concrete Jupyter examples, the material equips software engineers to make informed architectural decisions rather than following templates.

This refresh matters now as organizations shift generative AI from pilots to production workloads. Teams need developers who understand both the underlying mechanics and the operational realities of running these systems at scale. The Microsoft Cloud Advocates’ curriculum provides that bridge without requiring advanced machine learning degrees.

For mid-level engineers and technical leads working in Azure environments, the repository functions as both learning path and living reference. Its pragmatic focus on real-world patterns accelerates the transition from experimentation to deliverable applications.

**

Use Cases
  • Software engineers implementing semantic search with Azure embeddings
  • Backend developers integrating GPT models into enterprise applications
  • Teams building multimodal prototypes using DALL-E and Azure OpenAI
Similar Projects
  • huggingface/course - Offers broader transformer education but targets the open-source Hugging Face stack instead of Azure production patterns
  • openai/openai-cookbook - Supplies advanced API recipes that assume foundational knowledge this curriculum explicitly teaches
  • karpathy/nanoGPT - Focuses on training small models from scratch while this project prioritizes application integration and prompt techniques

More Stories

Microsoft Updates AI Agents Beginner Curriculum 🔗

Revised lessons incorporate current patterns in agentic RAG and multi-agent orchestration

microsoft/ai-agents-for-beginners · Jupyter Notebook · 57.1k stars Est. 2024

Microsoft has refreshed the ai-agents-for-beginners repository with updated Jupyter Notebook lessons that reflect 2026 realities of production agent development. The 12-module course provides concrete instruction on building AI agents that reason, use tools, maintain memory, and collaborate.

Lessons progress from basic agent design to advanced implementations using Semantic Kernel for plugin-based tool integration and AutoGen for multi-agent conversation management.

Microsoft has refreshed the ai-agents-for-beginners repository with updated Jupyter Notebook lessons that reflect 2026 realities of production agent development. The 12-module course provides concrete instruction on building AI agents that reason, use tools, maintain memory, and collaborate.

Lessons progress from basic agent design to advanced implementations using Semantic Kernel for plugin-based tool integration and AutoGen for multi-agent conversation management. Developers learn to construct agents that dynamically plan execution paths, call external functions, and recover from errors without constant human intervention.

A core module details agentic RAG techniques, showing how agents evaluate knowledge gaps then retrieve and synthesize information from vector stores and structured databases. Other notebooks demonstrate state persistence across long-running tasks, role specialization in multi-agent teams, and safe tool-calling patterns that limit model hallucinations.

The updates arrive as enterprises deploy agentic systems for workflow automation and customer support. Engineers familiar with Python can complete each self-contained lesson in under two hours, gaining working code samples that map directly to Azure and local deployment targets.

By focusing on battle-tested patterns from Microsoft's own AI platforms, the curriculum helps developers avoid common pitfalls when moving from prototypes to reliable production agents.

**

Use Cases
  • Software engineers implementing agentic RAG retrieval pipelines
  • Teams building multi-agent collaboration systems with AutoGen
  • Developers integrating Semantic Kernel plugins into enterprise tools
Similar Projects
  • langchain-ai/langchain - provides production agent chains with wider integrations
  • microsoft/semantic-kernel - supplies the core SDK taught across multiple lessons
  • crewAI/crewAI - focuses on role-based multi-agent task decomposition

Hermes Agent Adds Seamless Tool Gateway Integration 🔗

Nous Portal subscribers gain web search and automation without extra API keys

NousResearch/hermes-agent · Python · 106.7k stars 9mo old

Hermes Agent v2026.4.16 introduces the Nous Tool Gateway, allowing paid Nous Portal subscribers to access production-grade tools through their existing subscription.

Hermes Agent v2026.4.16 introduces the Nous Tool Gateway, allowing paid Nous Portal subscribers to access production-grade tools through their existing subscription. Web search via Firecrawl, image generation with FAL and FLUX 2 Pro, OpenAI text-to-speech, and Browser Use automation are now available with zero additional credentials.

The integration is deliberately frictionless. After running hermes model and selecting Nous Portal, users toggle individual tools with the use_gateway setting. The runtime automatically prefers gateway implementations even when direct API keys exist, replacing earlier environment-variable hacks with clean subscription-based detection.

This release ships more than 180 commits focused on reliability across the CLI, tool system, and runtime. The gateway works alongside Hermes's existing closed learning loop: the agent can now search the live web, generate visuals, or drive browsers as part of its autonomous skill creation and self-improvement cycles. Memory nudges, Honcho user modeling, and FTS5 conversation search remain unchanged but gain richer data sources.

Deployment flexibility is untouched. The agent runs on a $5 VPS, Modal serverless instances that hibernate when idle, or Docker/SSH environments. Interfaces span a full TUI, Telegram, Discord, Slack, and WhatsApp from one gateway process.

By collapsing tool authentication and billing into a single subscription, the update removes a major operational burden for developers who previously managed separate keys across providers.

**

Use Cases
  • Engineers deploying self-improving agents on Modal serverless backends
  • Researchers searching conversation history while generating FLUX images
  • Teams scheduling natural-language cron jobs across Telegram and email
Similar Projects
  • Auto-GPT - lacks native subscription tool gateway and persistent skill evolution
  • CrewAI - requires separate API key management and offers weaker cross-session memory
  • OpenAI Swarm - focuses on orchestration but provides no built-in TUI or Telegram continuity

Supabase Open Sources Multigres Kubernetes Operator 🔗

Zero-downtime upgrades and PITR backups now available for Postgres orchestration

supabase/supabase · TypeScript · 101.2k stars Est. 2019

Supabase has released its Multigres Kubernetes operator under an open source license, giving teams direct pod management, zero-downtime rolling upgrades, pgBackRest point-in-time recovery backups, and OpenTelemetry tracing.

The operator extends the platform's longstanding Postgres foundation, which combines a dedicated database with realtime subscriptions over websockets, auto-generated REST and GraphQL APIs through PostgREST, edge functions, file storage, and vector capabilities via pgvector. Developers can now run these components in self-managed Kubernetes environments with production-grade controls previously limited to Supabase's hosted service.

Supabase has released its Multigres Kubernetes operator under an open source license, giving teams direct pod management, zero-downtime rolling upgrades, pgBackRest point-in-time recovery backups, and OpenTelemetry tracing.

The operator extends the platform's longstanding Postgres foundation, which combines a dedicated database with realtime subscriptions over websockets, auto-generated REST and GraphQL APIs through PostgREST, edge functions, file storage, and vector capabilities via pgvector. Developers can now run these components in self-managed Kubernetes environments with production-grade controls previously limited to Supabase's hosted service.

Simultaneous platform changes broaden access. GitHub integration is now available on every tier, letting free-plan users connect repositories and deploy schema migrations from the main branch through existing CI/CD pipelines. Supabase has also joined the Stripe Projects developer preview as a co-design partner. The CLI tool provisions linked services including Supabase, Vercel, and Clerk, then writes credentials directly to .env files.

These moves reinforce Supabase's method of assembling enterprise open source tools rather than rebuilding proprietary equivalents. The result is a Postgres-centric development platform suitable for web, mobile, and AI workloads that prioritizes operational transparency and deployment flexibility.

Use Cases
  • Teams deploying scalable realtime apps on self-hosted Postgres
  • Engineers automating migrations via GitHub CI/CD pipelines
  • AI developers managing vector embeddings in Kubernetes clusters
Similar Projects
  • Firebase - proprietary closed-source alternative with less self-hosting
  • Neon - serverless Postgres focus without bundled auth or realtime
  • Appwrite - open source backend platform using different database core

Quick Hits

ML-For-Beginners Master classic machine learning through this 12-week hands-on curriculum of lessons, quizzes, and Jupyter notebooks for any builder. 85.3k
AI-For-Beginners Dive into AI fundamentals with this 12-week interactive curriculum of 24 lessons that makes complex concepts accessible to all. 46.7k
pytorch Build dynamic neural networks in Python with PyTorch’s tensors and GPU acceleration for flexible high-performance deep learning. 99.3k
Deep-Live-Cam Create real-time face swaps and one-click video deepfakes from a single image with this powerful vision toolkit. 91.4k
spec-kit Kickstart Spec-Driven Development with this practical toolkit that streamlines writing, managing, and implementing specifications. 89.8k

PythonRobotics Sharpens Control Algorithms for Legged and Aerial Systems 🔗

Recent updates expand nonlinear MPC and bipedal planning modules as developers tackle increasingly dynamic real-world autonomy problems

AtsushiSakai/PythonRobotics · Python · 29.2k stars Est. 2016

A decade after its creation, Atsushi Sakai’s PythonRobotics remains a practical workbench for engineers who need transparent implementations of core robotics algorithms. The most recent commits have strengthened nonlinear model predictive control using C-GMRES and added an inverted-pendulum bipedal planner, reflecting the shift toward legged robots and precision aerial maneuvers that dominate current research agendas.

The library’s value lies in its deliberate simplicity.

A decade after its creation, Atsushi Sakai’s PythonRobotics remains a practical workbench for engineers who need transparent implementations of core robotics algorithms. The most recent commits have strengthened nonlinear model predictive control using C-GMRES and added an inverted-pendulum bipedal planner, reflecting the shift toward legged robots and precision aerial maneuvers that dominate current research agendas.

The library’s value lies in its deliberate simplicity. Each module is written to expose the fundamental idea rather than hide it behind abstractions. Extended Kalman Filter localization, Particle filter localization, and histogram filtering sit alongside Gaussian grid mapping, ray-casting, and Lidar-to-grid conversion. These are not toy examples; they use the same covariance propagation and resampling steps found in production systems yet require only NumPy, SciPy, and Matplotlib.

SLAM coverage follows the same pattern. Iterative Closest Point (ICP) matching and FastSLAM 1.0 give developers working code to study data association and particle depletion problems without pulling in an entire middleware stack. Path-planning chapters span grid-based Dijkstra and A*, D* Lite, potential fields, and sampling methods including RRT, LQR-RRT, state-lattice planning, and quintic-polynomial trajectory generation. Reeds-Shepp curves handle non-holonomic constraints cleanly.

Recent aerial and bipedal additions address timely constraints. The drone 3D trajectory follower and rocket-powered landing sequence demonstrate how to manage underactuated dynamics and terminal velocity constraints. The bipedal planner treats the robot as an inverted pendulum, letting teams prototype footstep timing and center-of-mass trajectories in under a hundred lines.

Path-tracking implementations complete the picture. Stanley control, rear-wheel feedback, Linear–quadratic regulator (LQR) speed-and-steering, and nonlinear MPC with C-GMRES let engineers compare trade-offs between computational cost and tracking accuracy on the same vehicle model. All scripts produce animated visualizations that reveal transient behavior—such as how the Dynamic Window Approach respects acceleration limits or how potential fields escape local minima.

For builders, the project solves a persistent friction point: moving from equations on paper to working prototypes without licensing heavy toolboxes or learning proprietary APIs. Because dependencies are minimal, the same code runs on laptops, edge computers, and within larger Python-based autonomy stacks. Documentation maps directly to the source files, and the referenced academic paper supplies the theoretical grounding.

As autonomous-driving fleets, warehouse robots, and legged delivery platforms move from laboratories to streets, the ability to inspect, modify, and recombine these building blocks has become a competitive advantage. PythonRobotics keeps the signal high and the boilerplate low, exactly what engineering teams need when iteration speed determines which concepts reach deployment first.

Use Cases
  • Autonomous vehicle engineers validating nonlinear MPC controllers
  • Robotics researchers prototyping FastSLAM and ICP mapping pipelines
  • Aerospace teams simulating drone trajectory and rocket landing dynamics
Similar Projects
  • robotics-toolbox-python - Delivers broader kinematic and dynamic modeling for manipulators but provides fewer animated planning examples.
  • curobo - NVIDIA's CUDA library accelerates motion generation at production speeds yet omits the educational step-by-step algorithm breakdowns.
  • nav2 - Supplies a complete ROS2 navigation stack for deployment while PythonRobotics focuses on isolated, dependency-light reference implementations.

More Stories

Crocoddyl 3.2 Adds Actuation Limits for Thrusters 🔗

Update strengthens general action models, state safety and ROS integration for contact-rich control

loco-3d/crocoddyl · C++ · 1.2k stars Est. 2019

Crocoddyl has released version 3.2.0, introducing actuation limits that properly handle thrusters and general action models.

Crocoddyl has released version 3.2.0, introducing actuation limits that properly handle thrusters and general action models. The December 2025 update expands the library’s reach beyond traditional legged platforms to systems requiring constrained propulsion or non-standard actuators.

New safe difference and integration functions for states improve numerical stability during trajectory optimization. These additions reduce floating-point errors in differential-geometry calculations and long-horizon planning, outcomes that matter when feedback gains must be computed reliably on real hardware.

ROS users gain immediate benefits from the jrl_cmakemodules dependency and fresh Kilted CI support. The changes simplify builds and continuous testing inside established robotics workflows.

Under the hood the library continues to rely on efficient DDP-style solvers that compute optimal trajectories and feedback gains. It uses Pinocchio for fast rigid-body dynamics and analytical derivatives, while supporting analytical and sparse derivatives, multiple integrators, costs, constraints, and automatic differentiation through CppAD. Cache-friendly design, OpenMP multi-threading, and code generation in both C++ and Python keep runtime short enough for model-predictive control on physical robots.

Six years after its initial release, Crocoddyl remains a focused tool for contact-rich optimal control. The latest refinements address concrete limitations developers encounter when moving from simulation to machines with thrusters, aerial manipulators or hybrid locomotion.

Use Cases
  • Legged robotics teams optimizing dynamic gaits under contact constraints
  • Aerospace engineers modeling thruster limits in free-floating robot control
  • ROS developers deploying model predictive control on manipulation platforms
Similar Projects
  • acados - emphasizes real-time embedded nonlinear MPC with different solver backends
  • CasADi - supplies general optimal control tooling without native robotics models
  • Drake - offers broad simulation and control but uses alternative optimization methods

Newton 1.1 Boosts Deformable GPU Robotics Simulation 🔗

Latest release adds implicit MPM upgrades, TetMesh USD support and enhanced VBD joint handling

newton-physics/newton · Python · 4.4k stars 12mo old

Newton 1.1.0 delivers targeted upgrades to its GPU-accelerated physics engine, expanding capabilities that roboticists and simulation researchers rely on for scalable, differentiable modeling.

Newton 1.1.0 delivers targeted upgrades to its GPU-accelerated physics engine, expanding capabilities that roboticists and simulation researchers rely on for scalable, differentiable modeling.

The release significantly advances the implicit material point method solver with new material models, discretization choices, and wider example coverage. These additions enable more stable and accurate simulation of soft tissues and complex materials at scale. A new TetMesh asset class now supports direct USD and file loading for volumetric deformable meshes, streamlining workflows that previously required manual preprocessing.

Kinematic bodies gain full integration with variational brittle dynamics, which itself adds support for prismatic, revolute and D6 joints. Rendering improvements include Gaussian splats, tiled-camera textures, selectable Gaussian sorting modes, and production-grade PBR lighting with tone mapping inside the GL viewer. Contact sensing now aggregates friction data and performs better at high world counts.

Collision pipeline speed has increased while SDF memory footprint shrank. Validation suites for MJCF, USD, Newton-native and MJWarp conversion paths have been strengthened. Bug fixes address asset import inconsistencies, contact handling edge cases, viewer stability and multi-GPU execution.

Built on NVIDIA Warp and integrating MuJoCo Warp as primary backend, the engine remains focused on extensibility and rapid iteration. As a Linux Foundation project, Newton continues to evolve through community contributions first seeded by Disney Research, Google DeepMind and NVIDIA. The v1.1.0 changes give developers concrete performance and feature gains needed for next-generation differentiable control and soft-robotics research.

Use Cases
  • Roboticists modeling soft-body locomotion on GPU clusters
  • Researchers running differentiable MPM simulations for RL policies
  • Engineers importing USD volumetric meshes for joint validation
Similar Projects
  • Brax - JAX-based differentiable engine narrower in USD and rendering scope
  • Isaac Sim - Broader ecosystem but less emphasis on community extensibility
  • MuJoCo - Original CPU simulator whose Warp port powers Newton's backend

URDF Studio 1.0 Stabilizes Professional Robot Workflows 🔗

Browser-based editor solidifies structured modes, AI assistance, and multi-format export pipelines

OpenLegged/URDF-Studio · TypeScript · 358 stars 4mo old

URDF Studio reached stable status with its 1.0.0 release, introducing automated versioning scripts and consistent app version display while locking in production-ready capabilities for robot topology, geometry, and hardware authoring.

URDF Studio reached stable status with its 1.0.0 release, introducing automated versioning scripts and consistent app version display while locking in production-ready capabilities for robot topology, geometry, and hardware authoring.

The application organizes work into three dedicated modes—Skeleton for kinematic trees, Detail for visual and collision geometry, and Hardware for motor libraries and metadata. Users construct link-joint hierarchies, assign meshes, and configure actuators without writing raw XML at every step. Workspace management supports simultaneous multi-robot assemblies connected by bridge joints, synchronized file trees, edit history, and prepared resolution caches for URDF, MJCF, USD, Xacro, SDF, and .usp archives.

Visualization runs on a shared React Three Fiber canvas that combines runtime URDF/MJCF viewing with a vendored USD stage renderer. AI assistance handles model generation, automated inspection, and report export. The project publishes @urdf-studio/react-robot-canvas@0.1.0 independently, enabling developers to embed the 3D workspace in other web tools. MuJoCo export and roundtrip archive flows reduce friction when moving designs into simulation.

These refinements matter now because robotics teams increasingly need browser-based environments that maintain fidelity across design, assembly, and simulation formats without desktop CAD dependencies.

Use Cases
  • Robotics engineers editing kinematic trees in structured modes
  • Teams assembling multi-robot systems with bridge joint tools
  • Developers exporting models to MuJoCo for simulation testing
Similar Projects
  • urdf-editor - desktop XML-focused tool lacking web AI assistance
  • Isaac Sim - high-fidelity simulator without browser-based workflows
  • Blender URDF add-on - mesh authoring tool missing hardware metadata pipelines

Quick Hits

IsaacLab IsaacLab unifies robot learning on NVIDIA Isaac Sim, letting builders train diverse policies for legged, wheeled, and arm robots in high-fidelity sims. 7k
navigation2 Navigation2 delivers production-grade ROS 2 autonomy with modular mapping, localization, and dynamic path planning for reliable robot navigation. 4.2k
emanual ROBOTIS e-Manual gives builders complete hardware, software, and programming guides for every robot platform they produce. 189
mujoco MuJoCo simulates complex multi-joint physics and rich contact dynamics at high speed, powering robotics and reinforcement learning breakthroughs. 13k
autoware Autoware supplies a full open-source stack for autonomous vehicles, integrating perception, prediction, planning, and control in one modular platform. 11.4k

SWE-Agent 1.1 Delivers Training Trajectories for Open-Weights SOTA 🔗

New release and SWE-Smith generator push autonomous repository repair while maintainers recommend simpler mini successor

SWE-agent/SWE-agent · Python · 19k stars Est. 2024 · Latest: v1.1.0

SWE-agent has established itself as a framework that lets language models act as autonomous software engineers. Feed it a GitHub issue and it uses tools to navigate codebases, edit files, run tests, and iterate until the issue is resolved. Version 1.

SWE-agent has established itself as a framework that lets language models act as autonomous software engineers. Feed it a GitHub issue and it uses tools to navigate codebases, edit files, run tests, and iterate until the issue is resolved. Version 1.1.0, released this week, centers on SWE-smith, a new system that has generated tens of thousands of training trajectories. Those trajectories enabled the team’s SWE-agent-LM-32b model to claim state-of-the-art results among open-weights systems on SWE-bench Verified.

The update is not simply additive. The release notes document several breaking changes. The trajectory data format replaced the messages field with query. Multiple tool bundles that relied on the “windowed” file viewer have been renamed, and the review_on_submit bundle has been replaced by review_on_submit_m. The windowed tools themselves no longer append a trailing newline when creating new files. These adjustments reflect months of hardening the agent’s interaction loops for more reliable real-world use.

Maintainers have been explicit that most current development effort has moved to mini-swe-agent. The newer codebase matches the original’s performance on SWE-bench while fitting in roughly 100 lines of Python. The recommendation is to adopt the miniature version for new projects, reserving the full SWE-agent repository for research that needs its extensive configurability and hackability.

At the technical core lies a single YAML file that defines the agent’s entire behavior: available tools, formatting prompts, and termination conditions. This design leaves maximal agency with the underlying language model—whether GPT-4o, Claude 3.7 Sonnet, or an open-weights alternative—rather than forcing it through brittle, hand-crafted workflows. Default tool bundles provide structured methods for viewing, searching, and editing code without overwhelming the model’s context window.

The same architecture extends beyond bug fixing. The EnIGMA mode adapts the agent for offensive cybersecurity capture-the-flag challenges, achieving competitive placement on multiple security leaderboards. Researchers have also used it to explore competitive programming, where the model must discover algorithmic solutions without human scaffolding.

For builders the significance is practical. SWE-agent demonstrated that an LLM paired with a carefully chosen tool interface can close real issues in untouched repositories. The newly released trajectories lower the barrier to training specialized coding agents, while the pivot to a minimal implementation lowers the barrier to adoption. Teams that once experimented with the original can now choose the version that matches their tolerance for complexity without sacrificing capability.

The project remains deliberately research-friendly. Princeton and Stanford contributors have kept the codebase transparent and extensible so others can modify the agent loop, test new tool designs, or integrate different foundation models. With v1.1.0 the community gains both richer training data and clearer guidance on where the project is heading next.

Use Cases
  • Developers resolving GitHub issues with LLMs
  • Researchers training specialized coding agent models
  • Security teams solving offensive CTF challenges
Similar Projects
  • OpenDevin - supplies a full sandboxed desktop environment for multi-agent software engineering
  • Aider - focuses on conversational git-based code editing inside developer terminals and editors
  • LangGraph - provides graph-based orchestration for stateful LLM agents with broader workflow emphasis

More Stories

Awesome-Hacking Refreshes Lists for Emerging Threats 🔗

Curated security repositories gain new sections on CI/CD attacks and detection engineering

Hack-with-Github/Awesome-Hacking · Unknown · 110.7k stars Est. 2016

Ten years after its creation, Hack-with-Github/Awesome-Hacking continues to function as the central index for security professionals seeking specialized collections. The repository aggregates more than 30 focused awesome lists, each dedicated to a distinct domain, and has recently incorporated expanded coverage of CI/CD Attacks, Detection Engineering, and Embedded and IoT Security.

These additions arrive as supply-chain compromises and operational-technology breaches increase.

Ten years after its creation, Hack-with-Github/Awesome-Hacking continues to function as the central index for security professionals seeking specialized collections. The repository aggregates more than 30 focused awesome lists, each dedicated to a distinct domain, and has recently incorporated expanded coverage of CI/CD Attacks, Detection Engineering, and Embedded and IoT Security.

These additions arrive as supply-chain compromises and operational-technology breaches increase. The CI/CD section compiles offensive research on build pipeline weaknesses and deployment process exploits. Detection Engineering resources help teams design and operate controls that identify sophisticated intrusions rather than simply blocking known malware.

The project's structure delivers immediate utility through a clean table format. Security practitioners jump directly to:

  • Bug Bounty write-ups and active program directories
  • Fuzzing materials covering root-cause analysis and initial exploit development
  • Industrial Control System Security references for critical infrastructure
  • Android Security and IoT and Hardware Security toolkits
  • Incident Response and Honeypots operational guides

Maintenance relies on community contributions that keep the index current. Reverse engineers consult the dedicated reverse-engineering list, while red teams reference pentesting-windows resources. In an environment of proliferating tools and techniques, the project's role as a single, well-organized entry point saves hours of scattered searching and surfaces training materials that match precise operational requirements.

Word count: 178

Use Cases
  • Red team operators building Windows exploitation toolkits
  • Bug bounty hunters researching active programs and write-ups
  • Detection engineers designing controls for cloud environments
Similar Projects
  • awesome-pentest - provides direct tool inventories rather than meta-lists
  • SecLists - supplies concrete wordlists and payloads for immediate use
  • PayloadsAllTheThings - delivers ready-to-use attack patterns instead of curated references

Web-Check 1.0 Prioritizes Self-Hosted OSINT Analysis 🔗

Stable release optimizes Docker deployments and shifts v1 maintenance to dedicated organization

Lissy93/web-check · TypeScript · 32.9k stars Est. 2023

Web-Check has shipped version 1.0.0, focusing engineering effort on reliable self-hosting for security and infrastructure teams.

Web-Check has shipped version 1.0.0, focusing engineering effort on reliable self-hosting for security and infrastructure teams. The stable release packages a React frontend with a suite of backend functions inside a single Docker image that includes an integrated Node server, allowing complete local operation without external dependencies.

Deployment has been simplified. Operators can run the full stack on any Docker-compatible host, or continue using one-click Netlify and Vercel options when cloud hosting is preferred. Configuration flags let administrators tune analysis depth, timeout values, and data retention to match internal policy requirements.

Maintenance of the 1.x branch has moved to the new xray-web GitHub organization. This separation lets the current stable codebase receive focused updates while the team prepares a next major version whose API changes will not guarantee backwards compatibility.

The dashboard consolidates 15 analysis modules: IP geolocation, SSL certificate chains, DNS records with DNSSEC validation, security headers, cookie inspection, tracker enumeration, open port detection, traceroute mapping, redirect ledger, crawl directives, technology fingerprinting, and page-performance metrics including carbon-footprint estimates. All results remain under user control when self-hosted.

For organizations conducting repeated or sensitive reconnaissance, the local deployment removes third-party logging and rate-limit friction that constrained earlier usage.

**

Use Cases
  • Security engineers auditing vendor sites for misconfigurations
  • System administrators validating DNS and SSL settings at scale
  • Privacy researchers mapping tracker networks on target domains
Similar Projects
  • `theHarvester` - aggregates public OSINT sources but lacks unified dashboard
  • `Amass` - emphasizes attack-surface mapping over Web-Check's deep per-site metrics
  • `testssl.sh` - specializes in TLS configuration while Web-Check covers broader infrastructure

PhoneSploit Pro v2.0 Adopts Modular Architecture 🔗

Refactored release adds Rich console, Nmap scanning and 62 expanded ADB actions

AzeemIdrisi/PhoneSploit-Pro · Python · 5.8k stars Est. 2022

PhoneSploit Pro has received its most significant update since launch with the arrival of v2.0. The Python tool, long used by security teams to exploit Android devices through exposed ADB ports, has been restructured from a single monolithic file into a maintainable modular package.

PhoneSploit Pro has received its most significant update since launch with the arrival of v2.0. The Python tool, long used by security teams to exploit Android devices through exposed ADB ports, has been restructured from a single monolithic file into a maintainable modular package.

The overhaul introduces configurable path resolution for ADB, Metasploit-Framework, scrcpy and Nmap, eliminating brittle startup failures. Console output now uses the Rich library to present tables, status spinners, themed panels and clear progress indicators. Network discovery leverages python-nmap to scan LANs, delivering concise summaries of devices advertising ADB on TCP 5555 along with direct connection hints.

Menu navigation has been expanded to five paged screens offering 62 discrete actions. New capabilities include port forwarding, WiFi toggle utilities, root detection heuristics, app lifecycle commands, permission inspection, split-APK installation and live logcat streaming. When a target presents an open ADB port, the integrated workflow automatically creates, installs and launches a Metasploit payload to obtain a Meterpreter session in one command.

Beyond exploitation, the tool functions as a complete remote ADB suite supporting USB and Wi‑Fi sessions for shell access, screen unlocking, keycode injection, reboots and device locking. The refactoring improves long-term extensibility while preserving the project’s original goal of removing command-line complexity for penetration testers.

Word count: 178

Use Cases
  • Penetration testers automating Meterpreter sessions on Android devices
  • Security auditors scanning LANs for exposed ADB debug ports
  • Developers managing remote Android devices via enhanced ADB toolkit
Similar Projects
  • AhMyth - supplies Android RAT features but lacks Metasploit integration
  • msfvenom - generates payloads without automated ADB deployment chain
  • Drozer - performs app-level security testing instead of full device control

Quick Hits

opennhp Enforce Zero Trust security across AI infrastructure, apps, and data with OpenNHP's lightweight cryptography toolkit. 13.8k
httpx Probe HTTP targets at lightning speed with httpx, a versatile toolkit that combines multiple reconnaissance features in one tool. 9.8k
airgeddon Audit wireless networks with airgeddon, a Swiss-army bash script packing every WiFi testing tool Linux users need. 7.7k
nuclei Hunt vulnerabilities in apps, APIs, networks, DNS, and cloud configs using Nuclei's blazing-fast YAML-based scanner. 28k
openzeppelin-contracts Build secure smart contracts with OpenZeppelin's audited Solidity library of battle-tested, reusable components. 27.1k

Kubernetes v1.35.4 Refines Orchestration for Large-Scale Container Management 🔗

Latest patch release strengthens stability and scheduling while continuing the Borg-inspired approach to production workloads.

kubernetes/kubernetes · Go · 121.8k stars Est. 2014 · Latest: v1.35.4

Kubernetes v1.35.4 arrived this week with targeted stability improvements and bug fixes that matter to teams running containerized workloads at scale.

Kubernetes v1.35.4 arrived this week with targeted stability improvements and bug fixes that matter to teams running containerized workloads at scale. The changes are catalogued in the CHANGELOG-1.35.md, with additional binary downloads available through the same document. Announcements continue to flow through the kubernetes-announce mailing list.

Eleven years after its initial commit, the project remains the standard engine for managing containerized applications across fleets of hosts. It supplies the core mechanisms for deployment, maintenance, and automatic scaling. The design draws directly from Google’s Borg, blending fifteen years of production experience with community innovations in scheduling, failure recovery, and declarative state management. Hosted by the Cloud Native Computing Foundation, Kubernetes serves as the reference implementation for container-packaged, dynamically scheduled, microservices-oriented systems.

For platform builders the development workflow has stayed pragmatic. With a working Go environment the steps are simple: clone the repository, enter the directory, and run make. Teams preferring isolation can use a Docker environment and run make quick-release. Both paths produce the binaries needed to test changes against real clusters. The community repository centralizes contribution guidelines, architecture decision records, and escalation paths. Use of the k8s.io/kubernetes module as a library in other applications is explicitly unsupported; the project is meant to be consumed as a complete system.

The v1.35.4 release focuses on hardening behavior under heavy load rather than introducing flashy features. Fixes address edge cases in the scheduler, improve reliability of state reconciliation, and tighten security boundaries for multi-tenant clusters. These refinements matter now because production clusters continue to grow in both size and complexity. As organizations adopt hybrid-cloud and edge topologies, the cost of scheduling mistakes or undetected race conditions rises sharply.

Kubernetes solves a fundamental problem: turning a collection of independent containers into a coherent, observable, self-healing distributed application. It maintains declarative objects that describe desired state, continuously reconciles reality to that state, and provides consistent APIs for operators and controllers. Builders who own platform layers or run large microservices fleets therefore keep a close eye on each minor release. The project’s sustained pace of maintenance, visible in the April 2026 push activity, signals that the foundation for cloud-native infrastructure remains actively guarded.

Teams should review the full changelog before upgrading. For most production environments the move from earlier 1.35 releases will be straightforward, yet the accumulated fixes reduce risk in the clusters that can least afford outages.

Use Cases
  • Platform engineers automating deployment and scaling of containerized applications
  • DevOps teams managing production workloads across distributed container hosts
  • SREs building self-healing microservices with declarative orchestration tools
Similar Projects
  • Nomad - HashiCorp's simpler orchestrator that trades Kubernetes complexity for easier operational overhead
  • Docker Swarm - Docker's lightweight clustering solution that avoids the control-plane weight of Kubernetes
  • Apache Mesos - Earlier resource manager that influenced Kubernetes design but offers lower-level primitives

More Stories

Public Rust Release of Claw Agent Harness 🔗

Source repository details CLI sessions, parity tracking and container workflows

ultraworkers/claw-code · Rust · 186.9k stars 3w old

ultraworkers/claw-code supplies the canonical Rust implementation of the claw CLI agent harness. The repository, now public, contains the primary workspace inside rust/ together with task-oriented guides for real-world use.

Setup starts with `USAGE.

ultraworkers/claw-code supplies the canonical Rust implementation of the claw CLI agent harness. The repository, now public, contains the primary workspace inside rust/ together with task-oriented guides for real-world use.

Setup starts with USAGE.md, which covers compilation, authentication flows, session management and initial verification. After building, the first recommended step is claw doctor, a dedicated health-check command that validates the local environment and dependencies. PARITY.md records the current Rust-port checkpoint against the reference implementation, while ROADMAP.md lists remaining work including proper ACP daemon support. The command claw acp (or claw --acp) currently serves as the authoritative status tool rather than source layout inspection.

The project warns that cargo install claw-code fetches a deprecated crates.io stub that only prints a rename notice and installs an unusable binary. Users must instead build directly from the repository or run cargo install agent-code to obtain the upstream agent executable.

Container-first deployment receives its own guide in docs/container.md, enabling reproducible agent environments across machines. Companion Python scripts in src/ and tests/ support auditing and reference validation during the transition. The Rust workspace delivers native performance for CLI-driven agent sessions without relying on interpreted layers.

PHILOSOPHY.md outlines the design priorities behind the harness, giving contributors clear context for future extensions.

Use Cases
  • Engineers building and verifying Rust CLI agent binaries
  • Teams managing authenticated agent sessions and workflows
  • Developers tracking parity status across implementation checkpoints
Similar Projects
  • LangChain - Python agent orchestration with heavier dependencies
  • Aider - terminal AI coding tool built on Python scripts
  • CrewAI - multi-agent coordination framework lacking native Rust core

Bitcoin Core 31.0 Refines Network Validation Rules 🔗

Latest release sharpens peer-to-peer handling and block verification for full nodes

bitcoin/bitcoin · C++ · 88.9k stars Est. 2010

Bitcoin Core maintainers released version 31.0, the newest stable tag from the project's integration and staging tree. The update focuses on tightening peer-to-peer protocol behavior, improving transaction relay efficiency, and hardening validation logic that protects the network from malformed data.

Bitcoin Core maintainers released version 31.0, the newest stable tag from the project's integration and staging tree. The update focuses on tightening peer-to-peer protocol behavior, improving transaction relay efficiency, and hardening validation logic that protects the network from malformed data.

As the reference implementation since 2010, Bitcoin Core connects directly to the Bitcoin network to download, fully validate, and propagate every block and transaction. It enforces consensus rules without intermediaries, maintaining the blockchain's integrity. The software optionally includes a wallet and graphical interface, though most operators run it in headless mode on servers.

Development continues on the master branch with rigorous review. The project explicitly states that testing remains the main bottleneck; contributors are urged to test others' pull requests before changes reach release branches. Unit tests run via ctest, and every merge undergoes extended CI validation. The MIT-licensed codebase, written predominantly in C++, receives patches through the bitcoin/bitcoin repository, while GUI work occurs in a dedicated monotree mirror.

Version 31.0 binaries are available only from bitcoincore.org, cryptographically signed to prevent supply-chain tampering. In an era of rising transaction demand and persistent security threats, the release reinforces the software's role as critical infrastructure for independent verification.

**

Use Cases
  • Node operators validating every Bitcoin block without third-party trust
  • Developers integrating consensus rules into custom cryptocurrency tools
  • Enterprises running hardened nodes for secure payment settlement systems
Similar Projects
  • ethereum/go-ethereum - Ethereum's primary client with different consensus model
  • litecoin-project/litecoin - Direct Bitcoin Core fork tuned for faster blocks
  • monero-project/monero - Privacy-focused cryptocurrency with distinct ring-signature code

Quick Hits

hugo Hugo is the world's fastest framework for building websites, generating static sites in milliseconds with flexible templating. 87.7k
awesome-go Curated list showcasing the best Go frameworks, libraries, and software to accelerate your development workflow. 170.6k
scrcpy Display and control your Android device from your PC with low-latency video streaming and full input support. 138.8k
curl Command-line powerhouse for transferring data across 20+ protocols with libcurl's extensive features for robust application integration. 41.4k
mpv Powerful command-line media player delivering high-fidelity video playback with advanced features and extensive customization options. 34.9k
llama.cpp LLM inference in C/C++ 105.4k

LibreHardwareMonitor Expands Motherboard Support in v0.9.6 Release 🔗

Latest update adds Arctic fan control, new AMD platform compatibility and queryable metrics while refreshing .NET dependencies for modern hardware monitoring.

LibreHardwareMonitor/LibreHardwareMonitor · C# · 8.2k stars Est. 2017 · Latest: v0.9.6

LibreHardwareMonitor v0.9.6 demonstrates the project's ongoing relevance for developers and system builders who require precise, vendor-neutral visibility into PC hardware behavior.

LibreHardwareMonitor v0.9.6 demonstrates the project's ongoing relevance for developers and system builders who require precise, vendor-neutral visibility into PC hardware behavior. Released amid rapid chipset turnover, the update focuses on practical compatibility rather than flashy redesigns.

Recent contributions have plugged gaps in motherboard coverage. Support for the Gigabyte GA-A320M now includes previously missing controls, while the MSI B850 GAMING PLUS WIFI6E (MS-7E80) receives full sensor integration. These changes address the fragmented ways manufacturers expose temperature, voltage and fan data through SMBus and embedded controllers. A new Arctic fan controller implementation further extends reach into specialized cooling hardware popular in quiet and high-performance builds.

The release also improves the software's Prometheus-compatible /metrics endpoint by adding query parameter support. This seemingly small enhancement allows developers to filter sensor streams more efficiently, reducing overhead when feeding data into dashboards or alerting systems. Multiple dependency updates—including System.IO.Ports, System.Management, System.Text.Json and Microsoft.Windows.CsWin32—keep the codebase aligned with .NET 8.0 through 10.0 runtimes and address security patches.

At its foundation, LibreHardwareMonitor tackles a persistent challenge: hardware vendors ship monitoring utilities that are often closed-source, telemetry-heavy or narrowly scoped. The project counters this with both a polished Windows Forms application and a reusable LibreHardwareMonitorLib. The latter ships as a NuGet package and follows a straightforward visitor pattern:

Computer computer = new Computer {
    IsCpuEnabled = true,
    IsGpuEnabled = true,
    IsMotherboardEnabled = true,
    IsStorageEnabled = true,
    IsControllerEnabled = true
};
computer.Open();
computer.Accept(new UpdateVisitor());

This design lets builders read temperatures, fan speeds, voltages, clock rates and load levels from motherboards, Intel and AMD CPUs, NVIDIA/AMD/Intel GPUs, NVMe drives and network adapters without proprietary lock-in.

For PC builders pushing thermal limits, the library's transparency matters. Overclockers can validate cooling efficacy, infrastructure teams can log long-term stability metrics, and application developers can embed real-time sensor feeds into custom tools. The project explicitly welcomes motherboard-specific pull requests because sensor access methods still vary significantly across vendors.

Eight years since its creation as a maintained fork, LibreHardwareMonitor continues to matter precisely because hardware complexity keeps increasing—higher TDPs, denser power stages and AI accelerators all generate sensor data that needs reliable collection. Version 0.9.6 keeps the tool accurate where it counts most: at the metal.

**

Use Cases
  • Overclockers validating CPU thermal performance under load
  • Developers integrating sensors into custom diagnostic applications
  • System builders logging GPU and NVMe temperatures during stress tests
Similar Projects
  • OpenHardwareMonitor - Original codebase that LibreHardwareMonitor forked and now maintains with far broader modern hardware support
  • lm-sensors - Linux command-line equivalent offering raw sensor access but lacking Windows GUI and .NET library integration
  • CoreTemp - Narrowly focused CPU-only monitor that omits comprehensive motherboard, storage and fan controller coverage

More Stories

OWL v3.0 Adds YOLO AI and Wireless Dashboards 🔗

Major update brings in-crop detection, GPS tracking and over-the-air models to Raspberry Pi sprayer

geezacoleman/OpenWeedLocator · Python · 458 stars Est. 2021

Four years after its debut, OpenWeedLocator has landed version 3.0.0 with substantial upgrades to its Raspberry Pi-based weed detection platform.

Four years after its debut, OpenWeedLocator has landed version 3.0.0 with substantial upgrades to its Raspberry Pi-based weed detection platform. The system uses camera imagery and relay-controlled solenoids to deliver precise spot spraying in both fallow fields and growing crops.

The rewritten core pipeline now supports green-on-green detection through Ultralytics YOLO models. Both PyTorch .pt and NCNN .zip formats are accepted, with NCNN delivering three to five times faster inference on ARM hardware. GPS integration adds real-time weed mapping and tracking, while over-the-air model deployment lets users update detection logic without physical access.

A new web dashboard system offers two modes. Standalone runs directly on the Pi as its own MQTT broker for single-unit control. Networked mode lets a laptop or tablet in the tractor cab manage multiple OWL units across a local network. The touch-friendly interface features large buttons, threshold sliders and an on-screen numpad designed for field use.

Installation follows the same compact workflow: clone the repository, run the setup script, choose detection and dashboard options, then reboot. The project remains built entirely from off-the-shelf components and 3D-printable parts. With herbicide resistance rising and regulatory pressure increasing, the updates make targeted weed control more practical and accessible for farmers and researchers.

**

Use Cases
  • Broadacre farmers spot-spraying weeds in fallow fields
  • Vegetable growers running YOLO models for in-crop detection
  • Researchers mapping weed populations with GPS-equipped units
Similar Projects
  • ultralytics/ultralytics - supplies YOLO models but omits hardware integration and spraying control
  • FarmBot/farmbot-arduino-firmware - automates full beds yet lacks OWL's mobile sprayer focus
  • opendronemap/OpenDroneMap - processes aerial imagery without real-time ground-based solenoid triggering

Silhouette Card Maker v2.1 Refines Card Cutting 🔗

Update fixes template geometry and adds language controls to MTG plugin for precise proxy production

Alan-Cha/silhouette-card-maker · Python · 136 stars Est. 2025

Alan-Cha/silhouette-card-maker has shipped v2.1.0, delivering targeted fixes and workflow improvements for users turning digital images into physical cards with Silhouette machines.

Alan-Cha/silhouette-card-maker has shipped v2.1.0, delivering targeted fixes and workflow improvements for users turning digital images into physical cards with Silhouette machines.

The release corrects a kink in the corners of the cutting templates, producing cleaner edges on both letter and tabloid stock. It also expands the Magic: The Gathering plugin with a --prefer_lang flag that selects Spanish, French, Japanese, Korean, Simplified Chinese, Phyrexian or other supported languages while falling back to English. An --ignore_set option lets users exclude specific expansions, and the scryfall_json handler now respects chosen card versions.

Core scripts remain unchanged: create_pdf.py arranges images into layouts such as 4×2 on letter stock for standard TCG size or 6×3 on A3 for larger runs. offset_pdf.py adds registration offsets, while calibration sheets help dial in machine accuracy. The repository supplies ready templates for standard, poker, bridge, and euro-business sizes plus plugins that pull artwork from Scryfall and other databases.

Documentation stresses that proxies are for casual playtesting only and must be clearly identifiable. The updated templates and language tools reduce manual post-processing, letting builders move from PDF to finished stack in minutes rather than hours.

**

Use Cases
  • MTG players printing language-specific proxies for deck testing
  • Board game designers cutting prototypes on desktop Silhouette machines
  • Pokemon collectors producing identifiable playtest cards at home
Similar Projects
  • nanDECK - generates card PDFs but requires manual cutting setup
  • MPC Autofill - automates image collection without Silhouette templates
  • Tabletop Simulator mod tools - focuses on digital rather than physical output

DoujinSoft 4.0 Upgrades WarioWare DIY Archive API 🔗

New Yonderu endpoints and sample-based audio improve Playdate and Pebble integration

Difegue/DoujinSoft · Java · 73 stars Est. 2017

DoujinSoft version 4.0.0 adds native support for the API used by its companion comic reader Yonderu!

DoujinSoft version 4.0.0 adds native support for the API used by its companion comic reader Yonderu! DoujinSoft, extending the Java web application’s usefulness to Playdate and Pebble watch users nearly nine years after its first commit.

The application automatically ingests .mio files dropped into its data directory, parsing them with DIYEdit and Mio-Micro on every restart. An SQLite database then powers searchable, paginated lists of games, comics and records. Comics render in-browser; records stream through a custom soundfont driven by mio-midi. Any archive item can be injected directly into a user-supplied game save.

The 4.0 release centers on the /yonderu endpoint. GET /yonderu?random and GET /yonderu?daily return JSON for a random comic or the day’s scheduled entry, the latter drawn from a 366-line yonderu.txt file. A POST route accepts star ratings (1–5) and comments (1–8). Existing search pages accept &format=json, and iframe support lets developers embed playable content on external sites.

Contributors also rewrote microgame and record playback to use raw samples instead of MIDI, bumping the wahdio dependency in the process. These changes tighten audio fidelity and simplify maintenance for a project that remains the de-facto public archive for WarioWare DIY fan creations.

Use Cases
  • Archivists ingesting and cataloging new .mio fan files
  • Playdate owners pulling daily comics through the Yonderu API
  • Web developers embedding playable DIY records via iframe
Similar Projects
  • DIYEdit - supplies the core parsing engine DoujinSoft builds upon
  • Mio-Micro - provides the file-format library at the heart of the archive
  • itch.io Doujin Pages - alternative storefront for companion comic apps

Quick Hits

streamdeck Build custom Stream Deck apps and integrations with this TypeScript SDK that unlocks powerful hardware macro control. 229
librealsense Add precise depth sensing and 3D vision to robots or AR projects with the librealsense C++ SDK. 8.7k
litex Construct complex FPGA-based SoCs and custom hardware in minutes using LiteX's streamlined open-source framework. 3.8k
moduleur Design and build your own fully analog modular synth with this beginner-friendly DIY circuit ecosystem. 190
iiab Turn a Raspberry Pi into an offline Library of Alexandria packed with educational content using Internet-in-a-Box. 1.8k

Phaser 4 Rebuilds Core Renderer Bringing Major Performance Gains to HTML5 Games 🔗

Ground-up architectural changes deliver faster rendering, new filter system, and specialized GPU layers while preserving the API developers know

phaserjs/phaser · JavaScript · 39.4k stars Est. 2013 · Latest: v4.0.0

Phaser 4.0.0 marks the most significant evolution in the framework's 13-year history.

Phaser 4.0.0 marks the most significant evolution in the framework's 13-year history. The HTML5 game framework, actively maintained by Phaser Studio Inc and its open-source community, has received a complete overhaul of its WebGL renderer while deliberately retaining the API that thousands of developers already understand.

The centerpiece is a new Render Node Architecture. The v3 pipeline system has been replaced by a clean, modular node-based renderer in which each node performs a single, well-defined task. WebGL state is now fully managed internally, context restoration is automatic, and extensibility is substantially improved. These changes translate directly into more reliable performance across desktop and mobile browsers.

A Unified Filter System merges the previous FX and mask capabilities into one consistent interface. Developers can now apply filters to any game object or camera without the earlier restrictions. The release ships with an extensive library including Blur, Glow, Shadow, Pixelate, ColorMatrix, Bloom, Vignette, Wipe, GradientMap, and Quantize filters.

Performance gains are dramatic in two new specialized layers. SpriteGPULayer can render a million sprites in a single draw call—up to 100 times faster than standard sprite rendering—while driving GPU animations for position, rotation, scale, alpha, tint, and frame selection. TilemapGPULayer renders entire tilemap layers as a single quad, enabling 4096×4096 maps with per-pixel shader costs and perfect texture filtering that eliminates seams.

The update also introduces an overhauled tint system supporting six distinct modes (MULTIPLY, FILL, ADD, SCREEN, OVERLAY, HARD_LIGHT), new game objects such as Gradient and multi-dimensional Noise generators, improved lighting with one-line activation, and a CaptureFrame object for render-to-texture workflows.

Installation and project setup remain deliberately straightforward. Developers can add Phaser through npm (npm install phaser), reference it from jsDelivr or cdnjs, or use the create-phaser-game CLI tool. The latter offers official templates for React, Vue, Svelte, SolidJS, Angular, Next.js, and Remix across Vite, Rollup, Webpack, and Bun, in both JavaScript and TypeScript flavors.

Games built with Phaser 4 target the web directly or export as YouTube Playables, Discord Activities, and Twitch overlays. Third-party tools allow compilation to iOS, Android, Steam, and native desktop applications. For teams that need high-performance 2D graphics without descending into raw WebGL, the framework continues to solve the core problem of delivering smooth, cross-platform gameplay from a single JavaScript or TypeScript codebase.

The release demonstrates that mature open-source projects can still execute radical internal improvements without disrupting their user base. Builders who have relied on Phaser for years now gain modern rendering capabilities while continuing to ship the same declarative scene, physics, and animation code they already maintain.

Use Cases
  • Indie studios shipping browser-based 2D platformers
  • Developers creating Discord Activities and YouTube Playables
  • Teams building interactive educational web games
Similar Projects
  • PixiJS - delivers lower-level WebGL rendering but lacks Phaser's integrated scene management, physics, and game object system.
  • melonJS - offers another 2D HTML5 engine focused on platformers yet provides less GPU specialization and fewer official framework integrations than Phaser 4.
  • Babylon.js - targets 3D WebGL experiences where Phaser maintains a streamlined 2D-first architecture with superior sprite and tilemap performance.

More Stories

Dear ImGui 1.92.7 Refines Immediate-Mode Tooling 🔗

Spring release improves stability and directs users to overlooked changelog features

ocornut/imgui · C++ · 72.7k stars Est. 2014

Dear ImGui v1.92.7 focuses on maintenance and feature completeness for teams already using the library in production tools.

Dear ImGui v1.92.7 focuses on maintenance and feature completeness for teams already using the library in production tools. The update delivers stability fixes across backends while encouraging developers to scan the full changelog, where numerous extensions for debugging and visualization have accumulated.

The library's immediate-mode design continues to differentiate it. Instead of synchronizing widget trees with application state, programmers issue direct function calls each frame. This eliminates entire classes of desynchronization bugs and accelerates iteration when building internal editors or debug overlays.

Vertex buffers remain optimized for direct submission to existing 3D pipelines. With no external dependencies, integration into game engines or custom renderers typically requires fewer than 30 lines of setup code. The latest release refines multi-platform consistency and improves several community-maintained backends.

Documentation updates to the wiki now surface additional widget extensions and engine-specific bindings more prominently. These changes matter for studios shipping tools alongside titles, where rapid customization and low overhead directly affect developer productivity.

Corporate sponsorships sustain the project. The maintainers note that paid support contracts and Test Engine licenses fund both bug fixes and the gradual addition of advanced capabilities still absent from the core library.

The release reinforces Dear ImGui's role in performance-critical environments where full internationalization and accessibility features are secondary to simplicity and speed.

Use Cases
  • Game studios building in-engine debug and profiling overlays
  • Engineers integrating UI into real-time 3D visualization pipelines
  • Teams prototyping content creation tools inside custom applications
Similar Projects
  • Nuklear - lighter C-only immediate mode with fewer C++ abstractions
  • egui - Rust immediate-mode library following similar per-frame design
  • Qt - retained-mode framework offering richer widgets at higher cost

SpacetimeDB 2.1 Adds Rust Wasm and Unreal Support 🔗

Release removes barriers for browser clients and C++ game engines while fixing subscription logic

clockworklabs/SpacetimeDB · Rust · 24.6k stars Est. 2023

Clockwork Labs has released SpacetimeDB v2.1.0, extending the database-server's reach into browser environments and professional game engines.

Clockwork Labs has released SpacetimeDB v2.1.0, extending the database-server's reach into browser environments and professional game engines. The Rust client SDK now compiles to WebAssembly, letting developers build reactive front-ends that connect directly to the database without intermediary application servers.

C++ module bindings and the Unreal Engine SDK have been updated to the 2.0 API surface and code-generation system. Teams using Unreal can now work with current SpacetimeDB primitives and automatic client code without earlier compatibility layers.

The release also corrects two subscription-related bugs: useTable no longer flips its isReady state after the first row event, and v2 clients no longer drop subscriptions when other v2 clients disconnect.

SpacetimeDB continues to merge relational storage with application execution. Developers write schema, queries, and business logic inside a module; the database compiles and runs it, then pushes incremental state changes to every connected client. All data lives in memory for low-latency access, while a commit log on disk supplies ACID durability and crash recovery. No separate web tier, container orchestration, or cache layer is required.

These changes matter for teams that want to ship real-time systems with minimal operational overhead. The entire backend of BitCraft Online—chat, inventory, terrain, and player positions—already runs as one module serving thousands of concurrent users.

Rust Wasm support and the refreshed Unreal SDK lower the cost of experimenting with direct-to-database architectures in both web and AAA game contexts.

Use Cases
  • Game studios synchronizing MMORPG state without dedicated servers
  • Web developers deploying full-stack apps as single database modules
  • Unreal Engine teams building real-time multiplayer titles in C++
Similar Projects
  • Supabase - supplies real-time PostgreSQL but still needs separate application servers
  • Convex - offers reactive queries and functions yet uses a non-relational model
  • Firebase - delivers client-synced data through NoSQL instead of ACID transactions

Quick Hits

gecs GECS delivers a high-performance Entity Component System for Godot, letting builders craft scalable, modular game architectures. 500
Hengo Hengo brings intuitive visual scripting to Godot, empowering rapid prototyping and logic design without deep coding. 333
SPIRV-Cross SPIRV-Cross reflects on and converts SPIR-V shaders back to GLSL, HLSL, and MSL for effortless cross-platform graphics work. 2.4k
engine PlayCanvas powers high-performance 3D web games and XR experiences with WebGL, WebGPU, WebXR, and glTF support. 14.8k
pyxel Pyxel gives Python developers a lightweight retro game engine for crafting authentic 8-bit pixel art games instantly. 17.4k
godot Godot Engine – Multi-platform 2D and 3D game engine 109.8k