Design
Preset
Background
Text
Font
Size
Width
Account Pricing

The Git Times

AI Models
Claude Fable 5.1 $50/M GPT-5.6 Luna $1.20/M Gemini 3.1 Pro Preview $12/M Grok 4.6 $6/M DeepSeek V4 Pro 0813 $3.35/M Qwen3.8 2.4T A95B $6/M Kimi K3 $15/M
Full Markets →

Curl 8.22.0 Adds WebSocket Support for Real-Time Data Transfers 🔗

New WS/WSS protocol handling expands curl’s utility beyond traditional HTTP and file transfers

curl/curl · C · ▲ 13 in 1d Est. 2010 · Latest: curl-8_22_0

The latest release of curl, version 8.22.

0, introduces native support for WebSocket (WS and WSS) protocols, enabling direct real-time bidirectional communication from the command line. This update allows developers to test, debug, and interact with WebSocket servers using familiar curl syntax—such as curl ws://example.com/socket—without relying on external tools or custom scripts.

Beyond WebSockets, curl continues to unify data transfer across 25+ protocols including HTTP/3, SFTP, SMB, and MQTT, all powered by the portable libcurl library. Its single-binary design and MIT-like licensing make it ideal for embedded systems, CI/CD pipelines, and automation workflows where minimal dependencies and broad protocol coverage are essential. The project’s C implementation ensures high performance and low-level control, while maintaining backward compatibility across decades of use.

Despite its maturity, curl remains actively maintained, with the latest commit just days ago and a steady stream of protocol refinements. The addition of WS/WSS reflects growing demand for lightweight, scriptable tools in real-time web applications and IoT edge computing.

The catch: While curl’s WebSocket support enables basic client interactions, it lacks advanced features like subprotocol negotiation, per-message compression control, or binary frame handling found in full-featured WebSocket libraries, limiting its use in complex application-level scenarios.

Why this leads today curl remains essential infrastructure, with ongoing maintenance ensuring reliability in developer workflows where silent failures can disrupt critical automation and API interactions.

Use Cases
  • Developers testing WebSocket APIs in CI/CD pipelines
  • Automation scripts transferring data over SFTP and SMB
  • IoT devices exchanging telemetry via MQTT over constrained networks

Source: curl/curl — based on the README and release notes.

More on the Front Page

n8n-MCP bridges AI assistants to 2,616 workflow nodes for smarter automation 🔗

Latest release hardens workflow write path with node allowlist and rollback verification

czlonkowski/n8n-mcp · TypeScript · 22.8k stars Est. 2025

The czlonkowski/n8n-mcp project provides a Model Context Protocol server that gives AI assistants like Claude Desktop, Cursor, and Windsurf deep access to n8n’s ecosystem of 2,616 workflow automation nodes—832 core and 1,784 community. By exposing node properties, operations, documentation, and real-world examples, it enables AI to understand and build n8n workflows safely.

Developers deploy it via npm, Docker, or directly at dashboard.n8n-mcp.com to augment AI-assisted automation without leaving their IDE.

Release v2.81.0 focuses on reliability: it hardens the workflow write path with a node allowlist, settings retry logic, and rollback verification to prevent unintended changes. The update comes from a single contributor and one commit, reflecting maintenance rather than feature expansion. The project now serves tens of thousands of users, with sponsorship supporting ongoing compatibility with n8n releases.

The catch: Despite 99% node property coverage, only 66.5% of node operations and 86% of documentation are mapped, limiting AI’s ability to execute complex actions or access full guidance without manual verification.

Use Cases
  • Developers generate n8n workflows using AI in Cursor
  • Teams audit automation logic via Claude Code suggestions
  • Beginners learn node configurations through AI-powered examples

Source: czlonkowski/n8n-mcp — based on the README and release notes.

Reverify Grounds AI Reverse Engineering in Binary Verification 🔗

Deterministic tools validate language model hypotheses against actual binary bytes

2akouwu/reverify · Python · 671 stars 3d old

Reverify pairs AI with pure-Python reverse engineering tools to prevent hallucinated results. The model proposes structural claims about binaries; deterministic engines disassemble, emulate, or pattern-match to verify them.

Outputs are tagged VERIFIED, REFUTED, or INCONCLUSIVE with observed bytes. It supports PE/ELF/Mach-O, x86/x64/ARM/ARM64, and generates Frida hooks. Install via pip install reverify; full features require optional engines like Capstone or Unicorn.
The catch: Pure-Python core lacks performance of native disassemblers, limiting speed in large-scale binary analysis.

Use Cases
  • Malware analysts verifying AI-suggested struct offsets
  • CTF players validating algorithm reconstructions from binaries
  • Researchers generating Frida hooks for interoperability testing

Source: 2akouwu/reverify — based on the README and release notes.

Tailwind Class Merger CN Replaces Tailwind-Merge and Clsx with 30× Speed Gain 🔗

Zero-dependency engine works across frameworks and runtimes with identical APIs

shadcn-ui/cn · TypeScript · 802 stars 3d old

The cn library from shadcn-ui is a TypeScript engine for merging Tailwind CSS classes that replaces tailwind-merge and clsx with full API parity and zero dependencies. Benchmarks show it runs 30–37× faster than the prior combo while outputting 26 KB minified JavaScript.

It operates in React, Vue, Svelte, Solid, Astro, and plain templates across Node, Bun, Deno, browsers, and edge runtimes. Migration requires importing cn and replacing the wrapper in @/lib/utils, then aliasing legacy imports.
The catch: As a v0.2.4 release with five open issues, long-term stability and adoption in large-scale enterprise projects remain unproven.

Use Cases
  • React developers replacing clsx/tailwind-merge in component utilities
  • Vue teams optimizing Tailwind class handling in single-file components
  • Svelte creators reducing bundle size in utility-first styling workflows

Source: shadcn-ui/cn — based on the README and release notes.

One-click Rust tool installs leaked DLSS 5 in any DX11/DX12 game 🔗

Single Windows exe adds neural rendering via ReShade or OptiScaler fork

faisalkindi/DLSS5oneclick · Rust · 288 stars 3d old

Faisalkindi/DLSS5oneclick delivers a native Rust executable that one-click installs the leaked DLSS 5 neural-rendering build into any DirectX 11 or 12 game, working even on titles without native DLSS support. It bundles ReShade with the RenoDX add-on or OptiScaler’s DLSSNR fork, extracting DLLs and configuring manifests for clean removal.

The tool auto-detects games via Steam IDs or folder names to optionally apply RenoDX HDR mods. Latest release v0.10.3 fixes driver misdiagnosis by verifying NGX Core first.
The catch: Targets unpatched models; RTX 20–40 support is untested and may fail on older hardware.

Use Cases
  • Gamers add DLSS 5 to legacy DX11 titles
  • Modders test neural rendering without driver hacks
  • Streamers boost FPS in unsupported DX12 games

Source: faisalkindi/DLSS5oneclick — based on the README and release notes.

VoiceStudio Adds Crash-Isolated Voice Generation for Apple Silicon 🔗

New release prevents fatal MPS memory exits from taking down the local backend during dubbing workflows

debpalash/VoiceStudio · Python · 15.6k stars 4mo old

VoiceStudio v0.5.

1 isolates OmniVoice generation on Apple Silicon in a crash-resistant child process, stopping MPS memory faults from killing the backend. The update also improves GPU exhaustion handling, returns actionable dubbing errors, and prevents overlapping WhisperX calls that caused Windows crashes. Developers can now trigger local dictation or connect via HTTP, WebSocket, JSON-RPC, CLI, or MCP. Model catalogue UI scales to desktop width without clutter.
The catch: Intel Macs still cannot run the local Python backend and must use a remote workaround.

Use Cases
  • Developers dubbing video locally without cloud dependencies
  • Audiobook creators generating multilingual narration on-device
  • Teams integrating voice cloning via native CLI or RPC transports

Source: debpalash/VoiceStudio — based on the README and release notes.

CrowdStrike Falcon Bypass Exploits Office Macro Remediation for Privilege Escalation 🔗

Proof-of-concept targets Windows 11 25H2 and Server 2025 with Phase 3 Optimal Protection enabled

MSNightmare/FalconFlank · C · 259 stars 0d old

MSNightmare/FalconFlank is a C-based 0day that escalates privileges by abusing CrowdStrike Falcon’s malicious Office macro remediation. It triggers when Falcon attempts to quarantine a macro-laden document, hijacking the DLL load process to execute arbitrary code with SYSTEM rights.

The exploit works on fully patched Windows 11 25H2 and Windows Server 2025 systems running Falcon Sensor with Phase 3 Optimal Protection and the “Microsoft Office file malicious macro removal” feature active.
The catch: Reliance on outdated bypass techniques means CrowdStrike likely has detections already, requiring exclusions or obfuscation for testing.

Use Cases
  • Red teams testing endpoint detection evasion
  • Security researchers validating macro-based attack surfaces
  • Adversary simulation for Falcon Sensor bypass scenarios

Source: MSNightmare/FalconFlank — based on the project README.

Open Source Shifts Toward Agent-First Tooling for LLMs 🔗

Projects now focus on composable, verified agent skills over monolithic model releases

Trendllm-tools
anthropics/commerce-agents2akouwu/reverifyRyze-AI-Adgent/open-seo-mcp-skillsuseagenthq/useagentPhiloLabs/fable51-worldscarloslfu/slotstreamlnkiai/m3e-canvasHuman-Agent-Society/reeflobehub/lobehubczlonkowski/n8n-mcpczlonkowski/n8n-skillslangchain-ai/open-sweFrancyJGLisboa/agent-skills-platformray-project/rayAI4Finance-Foundation/FinGPTAOrbitron/Eridanusjustrach/codedbddalcu/mlx-serve

A clear pattern is emerging in open source: the rise of llm-tools that treat language models as orchestratable components rather than endpoints. Instead of chasing bigger models, projects are shipping deterministic, verifiable, and composable tooling that grounds agent behavior in real systems.

reverify anchors reverse engineering in binary validation, eliminating hallucinations by checking outputs against actual code. useagent and lobehub provide frameworks for AI coworkers with persistent cloud computers, tool access, and team-wide scheduling — turning agents into deployable coworkers. czlonkowski/n8n-mcp and n8n-skills let Claude build and validate workflows through MCP, while open-seo-mcp-skills ties SEO actions directly to live Search Console and GA4 data via MCP, with DataForSEO integration. fable51-worlds enables world generation through code, and agent-skills-platform introduces a governed marketplace for tested, versioned agent skills with rollback and quarantine. Even infrastructure like ray and mlx-serve is adapting to serve as low-overhead, API-compatible backends for agent workloads. The shift is clear: value now lies in toolchains that make agents reliable, auditable, and actionable — not in the models themselves.

The catch: Much of this tooling remains fragmented across incompatible agent protocols (MCP, custom wrappers, direct API calls), with limited interoperability and real-world validation beyond demos; many projects prioritize novelty over operational maturity, leaving critical gaps in error handling, security, and long-term maintainability that skeptics rightly question.

Use Cases
  • Developers build verified shopping agents using Claude and retail APIs
  • Teams deploy AI coworkers that manage spreadsheets, reports, and PRs autonomously
  • Marketers automate SEO audits and rank tracking using live analytics data via MCP

Open Source Data Infrastructure Shifts Toward AI-Native, Modular Pipelines 🔗

Projects integrate AI agents, streaming LLMs, and zero-trust security into composable data workflows

Trenddata-infra
Ryze-AI-Adgent/open-seo-mcp-skillsuseagenthq/useagentcarloslfu/slotstreamSimplifyJobs/Summer2027-Internshipslightdash/lightdashslothflowlabs/duckletobymao/sqlglotSnailclimb/JavaGuidemicrosoft/ML-For-Beginnersmicrosoft/qlibray-project/rayquestdb/questdbOpenNHP/opennhpredis/redisobsproject/obs-studiocurl/curlClickHouse/ClickHouseskypjack/entt

A clear pattern emerges in open-source data infrastructure: the rise of AI-native, modular systems that treat data pipelines as programmable surfaces for agents and models. DuckDB-based duckle delivers no-code ETL with MCP integration, letting AI agents trigger pipelines via natural language.

Lightdash pushes agentic BI, where analytics dashboards generate and refine themselves through code-driven agents. Slotstream enables running massive MoE models like Qwen3.8-Flash-Next on modest hardware by streaming experts from SSD, lowering the barrier to local LLM use in data tasks. Meanwhile, OpenNHP applies zero-trust cryptography to infrastructure, securing data-in-motion and -at-rest in AI-driven environments. Useagent extends this by giving AI coworkers persistent cloud computers and tool access to execute multi-step data work — from reports to PRs — using Claude Code or Codex. These projects reject monolithic stacks in favor of composable, secure, agent-ready layers where data movement, transformation, and insight generation are orchestrated by AI. The trend reflects a shift from passive data tools to active, intelligent infrastructure that anticipates user needs and operates with minimal manual intervention.
The catch: Much of this remains experimental — MCP adoption is still nascent, agent reliability varies under complex workflows, and streaming LLM approaches like slotstream trade latency for accessibility, raising questions about real-world throughput in production data pipelines at scale.

Use Cases
  • Data engineers deploy self-healing ETL pipelines triggered by AI agents
  • Analysts generate real-time BI reports via natural language to agentic dashboards
  • Teams run local LLMs for data enrichment without exceeding memory limits

Web frameworks evolve beyond UI into full-stack AI and infra orchestration 🔗

New tools blend MLX streaming, GraphQL federation, and terminal workspaces to redefine full-stack development

Trendweb-frameworks
carloslfu/slotstreamlnkiai/m3e-canvaswundergraph/cosmonyakang/nyatermChromeDevTools/devtools-frontendJakubantalik/Librariesppy/osu-webAOrbitron/Eridanusdanhumphrey/page-modellermaurosoria/dirsearchcurl/curlddalcu/mlx-servecocos/cocos-engine

Open source web frameworks are shifting from UI-centric libraries to integrated systems that handle AI inference, backend orchestration, and developer tooling. Projects like carloslfu/slotstream enable running massive MoE models like Qwen3.

8-Flash-Next on Macs by streaming experts from SSD using MLX and Swift, exposing an Ollama-compatible API — turning local hardware into viable AI backends. Similarly, ddalcu/mlx-serve delivers a native Zig-based LLM server for Apple Silicon with OpenAI/Anthropic API compatibility, eliminating Python dependencies. On the backend, wundergraph/cosmo provides an open-source alternative to Apollo Studio for managing GraphQL Federation at scale, offering schema registry, routing, and collaboration tools. Developer experience is also being reimagined: nyakang/nyaterm delivers a modern remote terminal workspace in the browser, while lnkiai/m3e-canvas lets designers sketch Material 3 Expressive UIs and export them as vibe-coding prompts for AI-assisted development. Even DevTools are being extended — danhumphrey/page-modeller is a browser extension for modeling web pages to enable automation, blurring the line between inspection and code generation. These repos signal a trend where web frameworks are no longer just about rendering views but about composing AI, infra, and dev tooling into cohesive, locally executable systems.
The catch: Much of this remains experimental — slotstream and mlx-serve depend on Apple Silicon and MLX, limiting cross-platform reach; cosmo lacks the enterprise tooling of GraphOS; and vibe-coding tools like m3e-canvas trade precision for speed, risking brittle AI-generated output. Integration complexity and fragmented standards threaten to undermine the promise of seamless full-stack AI dev.

Use Cases
  • Developers run local LLMs on MacBooks via SSD-streamed experts
  • Teams manage federated GraphQL schemas without vendor lock-in
  • Designers generate AI prompts from browser-based Material 3 sketches

Deep Cuts

Open-source team rebuilds Microduck robot from Pollen's CAD and code 🔗

Full mechanical and electronic reconstruction enables custom bipedal experimentation

fanhao375/microduck-replica · Python · ▲ 79 in 1d

Fanhao375/microduck-replica reverse-engineers Pollen Robotics' Microduck using official MJCF models and Rust source, delivering CAD assemblies and a complete electronics plan in Python. Builders can now study, modify, and 3D-print a legally accessible version of this dynamic bipedal platform without needing the original hardware.

The project includes motor drivers, sensor integrations, and control logic adapted for reproducibility. It lowers the barrier to legged robot research by providing a documented, open foundation for gait testing and sensor fusion.
The catch: It's still early-stage, with limited documentation and no simulation integration, making plug-and-play use challenging for newcomers.

Use Cases
  • Researchers testing custom walking gaits on open hardware
  • Educators teaching legged robotics with reproducible builds
  • Hobbyists adapting Microduck for sensor payload experiments

Source: fanhao375/microduck-replica — based on the project README.

Open SEO Skills Give Claude Real Search Data for Free 🔗

Runs keyword research, audits, and rank tracking via MCP with DataForSEO integration

Ryze-AI-Adgent/open-seo-mcp-skills · Shell · 380 stars

Ryze-AI-Adgent/open-seo-mcp-skills turns Claude into an SEO analyst by connecting it to your live Search Console, GA4, and ads data through the Model Context Protocol. With DataForSEO built in, it delivers keyword research, backlink analysis, site audits, rank tracking, and competitor gap reports — all as installable skills.

No API keys to manage, no cost, and fully open-source under MIT. Builders can automate SEO workflows inside Claude chats, turning conversational AI into a proactive optimization agent. It works today with any Claude instance supporting MCP servers.
The catch: It's early-stage and niche, requiring technical setup that deters non-developers despite its powerful, free capabilities.

Use Cases
  • SEO auditors automate site health checks using live Search Console data
  • Content teams uncover keyword gaps via Claude-driven competitor analysis
  • Agencies track rankings and backlinks without switching tools or paying fees

Source: Ryze-AI-Adgent/open-seo-mcp-skills — based on the project README.

Quick Hits

dlss5-visual-enhancer Enhances video and image quality using DLSS 5 neural networks with frame interpolation for smoother, sharper visuals in real-time applications. 338
reef Provides infrastructure for continual learning, enabling AI agents to self-improve over time through adaptive knowledge retention and updates. 245
lobehub Orchestrates AI agents as a 24/7 Chief Agent Operator by hiring, scheduling, and reporting on your entire AI team for autonomous operations. 82.2k
fable51-worlds Generates interactive worlds directly from code using Fable 5.1, letting developers build and simulate environments programmatically. 275
m3e-canvas Lets you sketch Material 3 Expressive UIs in the browser and convert designs into vibe-coding prompts for rapid frontend development. 247
commerce-agents Offers a reference blueprint for building shopping and merchant agents powered by Claude, with cross-industry examples in retail, telecom, and entertainment. 1.1k
noty Delivers encrypted, local-only sticky notes that reside at the screen’s edge, built with native macOS SwiftUI and AppKit for seamless, private use. 335
Who shipped it

The Business Desks

Fresh on Hugging Face

Model Drops

The newest model releases builders are picking up right now.
Beyond GitHub

The AI Wire

What builders are reading today — the headlines, papers, and announcements that aren't trending repos.

From the labs & arXiv

ComfyUI Adds MiniMax Music 3 for Unified Audio-Visual Workflows 🔗

Latest release enables dynamic VRAM by default on ROCm 7.14+, boosting accessibility for AMD GPU users

Comfy-Org/ComfyUI · Python · ▲ 147 in 1d Est. 2023 · Latest: v0.34.0

Comfy-Org/ComfyUI’s v0.34.

0 release integrates MiniMax Music 3, allowing users to generate and synchronize audio tracks directly within its node-based interface. The update adds dedicated nodes for anchoring image and audio guides at any frame, streamlining multimodal creation for video and music projects. Developers benefit from sped-up Gemma4 text generation and improved OpenAPI contract synchronization, reducing latency in prompt handling and backend communication.

A key infrastructure shift enables dynamic VRAM allocation by default on ROCm 7.14 and higher, lowering the barrier for AMD GPU adopters who previously needed manual configuration. This change, paired with fixes for minimax music compatibility on non-dynamic VRAM setups, broadens hardware flexibility without sacrificing performance. The project continues to support Python 3.14, with test updates ensuring stability amid evolving language semantics.

Despite its modular strength, ComfyUI’s reliance on a complex node graph can overwhelm newcomers, and its Python/PyTorch stack demands significant VRAM for high-resolution outputs — a trade-off builders must weigh against simpler, less flexible alternatives.

The catch: The node-based workflow, while powerful, presents a steep learning curve for users accustomed to traditional prompt-only interfaces, potentially slowing adoption in fast-paced creative pipelines.

Previously in The Times “covered” — Aug 31

Use Cases
  • Visual artists generating synchronized music videos
  • AI researchers experimenting with multimodal diffusion models
  • Developers building custom audio-visual generation APIs

Source: Comfy-Org/ComfyUI — based on the README and release notes.

More Stories

Microsoft’s Generative AI Course Adds Offline Model Execution for Local Development 🔗

Foundry Local enables running models without cloud subscriptions, expanding accessibility for builders

microsoft/generative-ai-for-beginners · Jupyter Notebook · 119.1k stars Est. 2023

The microsoft/generative-ai-for-beginners repository now includes Foundry Local lessons, allowing developers to run AI models fully offline on personal devices. This update removes dependency on Azure OpenAI Service or internet connectivity during experimentation.

Learners can execute models locally using Python or TypeScript examples across 21 lessons covering prompt engineering, semantic search, and multimodal AI with DALL-E and GPT integrations. The course remains structured around “Learn” and “Build” lessons, with code samples and “Keep Learning” resources.
The catch: Offline execution is limited to smaller models; advanced workloads still require cloud-based Foundry Models or Azure OpenAI for full performance.

Previously in The Times “covered” — Aug 27

Use Cases
  • Students learning generative AI without cloud costs
  • Developers prototyping AI features in air-gapped environments
  • Educators teaching AI fundamentals with offline-ready materials

Source: microsoft/generative-ai-for-beginners — based on the project README.

Microsoft’s ML-For-Beginners curriculum updates with Copilot tips for data science 🔗

12-week, 26-lesson course now includes GitHub Copilot guidance for Scikit-learn workflows

microsoft/ML-For-Beginners · Jupyter Notebook · 90.1k stars Est. 2021

The Microsoft/ML-For-Beginners repository added a Discord-based "Learn with AI" series running September 18–30, 2025, offering tips on using GitHub Copilot for data science tasks. The curriculum covers classic machine learning with Scikit-learn across 26 lessons, 52 quizzes, and project-based assignments in Jupyter Notebook format.

It avoids deep learning, focusing instead on foundational algorithms applicable to global datasets. The catch: Last updated over five years ago, the core content may not reflect recent Scikit-learn API changes or newer ML best practices.

Use Cases
  • Students learn Scikit-learn through hands-on projects
  • Educators adopt structured ML curriculum for classrooms
  • Developers refresh classic ML concepts with Python tools

Source: microsoft/ML-For-Beginners — based on the project README.

Microsoft Qlib Integrates RD-Agent for Automated Quant Research Workflows 🔗

New release adds Parquet data support and Pydantic-based MLflow configuration

microsoft/qlib · Python · 48.2k stars Est. 2020

Qlib, Microsoft’s open-source AI-driven quantitative investment platform, now integrates RD-Agent to automate factor mining and model optimization in quant research. The latest release v0.

9.7 introduces Parquet data handling, Pydantic-settings for MLflow config, and a unified BaseDataHandler interface. It supports supervised learning, market dynamics modeling, and reinforcement learning across the full ML pipeline—from data processing to order execution. The project sees active maintenance, with commits daily and over 7,600 forks.
The catch: Despite frequent updates, 474 open issues suggest ongoing challenges in scalability and usability for production-grade deployments beyond research prototypes.

Use Cases
  • Quant researchers automate factor discovery using ML models
  • Financial engineers backtest trading strategies on historical data
  • Investment teams optimize portfolios with reinforcement learning agents

Source: microsoft/qlib — based on the README and release notes.

Quick Hits

FinGPT FinGPT provides open-source financial large language models with trained weights on HuggingFace, enabling developers to build and deploy AI-driven financial analysis tools. 21.2k
hermes-agent Hermes-Agent is an adaptive AI agent that learns and evolves with user interactions, offering personalized, context-aware assistance without manual retraining. 240.5k
ray Ray delivers a unified distributed computing framework that scales AI and ML workloads seamlessly across clusters, from laptops to data centers. 43.7k
openai-cookbook OpenAI Cookbook offers practical, ready-to-run Jupyter notebooks demonstrating real-world applications of the OpenAI API for developers. 75.7k
pytorch PyTorch enables flexible, GPU-accelerated deep learning with dynamic computation graphs and intuitive tensor operations for research and production. 102.7k

Kornia v0.8.3 Boosts PyTorch Vision with Float16 Support and New Feature Detectors 🔗

Release adds XFeat, ALIKED, batched VisualPrompter, and 2x–3x speedups in core geometry ops

kornia/kornia · Python · ▲ 3 in 1d Est. 2018 · Latest: v0.8.3

Kornia, the differentiable computer vision library built on PyTorch, released v0.8.

3 with meaningful performance and feature upgrades. The update introduces half-precision (float16/bfloat16) support, enabling faster training and inference on compatible hardware without sacrificing accuracy. New local feature detectors and descriptors — XFeat and ALIKED — expand Kornia’s toolkit for tasks like image matching and SLAM. VisualPrompter now supports batched inference, streamlining AI-driven image editing workflows. Core geometry operations saw significant speedups: depth_to_normals runs ~3x faster, while transform_points and denormalize_points_with_intrinsics improved ~2x. Documentation and docstring coverage were broadened, and several deprecated utilities in kornia.utils now emit DeprecationWarnings ahead of removal. The release integrates ~175 commits since v0.8.2, with test infrastructure enhanced via Pixi for reproducible CPU half and float32 validation. Kornia remains a go-to for developers needing differentiable image processing, augmentation, and pre-trained models like SAM, LoFTR, and MobileViT directly in PyTorch pipelines.
The catch: Despite its breadth, Kornia’s heavy reliance on PyTorch limits flexibility for teams using TensorFlow, JAX, or standalone C++ inference, and some advanced operators still lack ONNX export fidelity.

Previously in The Times “covered” — Aug 27

Use Cases
  • Train robust vision models with differentiable augmentation pipelines
  • Deploy real-time feature matching for robotics and AR applications
  • Integrate pre-trained segmentation and detection models into PyTorch workflows

Source: kornia/kornia — based on the README and release notes.

More Stories

QuestDB 10.0.1 refines time-series database with improved protocol and query handling 🔗

Release strengthens QWP table-update caching and high-cardinality window queries for live data workloads

questdb/questdb · Java · 17.3k stars Est. 2014

QuestDB 10.0.

1 improves reliability in its QuestDB Wire Protocol (QWP), ensuring table-update caching respects schema changes like DROP/recreate and column-type updates. The release also enhances high-cardinality window queries and Live View refreshes, while enabling stream processing of ordered UNION ALL without full sorts. JDK 24+ idle-worker allocation is reduced, and Web Console updates to version 2.0.3. These changes support low-latency ingestion and real-time analytics on time-partitioned, memory-mapped columnar data.
The catch: Despite performance gains, the database remains Java-centric with native dependencies, potentially limiting deployment in strictly managed or minimal runtime environments.

Use Cases
  • Financial analysts streaming tick data for real-time market analysis
  • Industrial IoT systems ingesting sensor telemetry at scale
  • Developers building low-latency dashboards with SQL and Arrow integration

Source: questdb/questdb — based on the README and release notes.

NASA JPL Open Source Rover v4.0.0 enables Mars-like terrain navigation with COTS parts 🔗

Rocker-Bogie suspension and differential pivot allow all six wheels to maintain ground contact over obstacles

nasa-jpl/open-source-rover · Prolog · 9.6k stars Est. 2018

The NASA JPL Open Source Rover project released v4.0.

0, its first stable major redesign in over five years, featuring updated parts lists generated via GitHub Actions and a revised endcap using a 2-hole beam. Built entirely from consumer off-the-shelf components—primarily GoBilda—the rover replicates Mars rover mechanics including Rocker-Bogie suspension and differential pivot for rugged terrain traversal. Designed as a learning platform, it requires no prior skills and supports expansions like a head display or robot arm. Despite its educational focus, the project shows signs of maintenance strain with 16 open issues and reliance on automated documentation updates.
The catch: The rover’s dependence on GoBilda parts may limit accessibility for builders outside regions with reliable shipping or those seeking lower-cost alternatives.

Use Cases
  • Students learn mechanical assembly using GoBilda components
  • Educators teach robotics through Mars-relevant suspension systems
  • Hobbyists test custom expansions like sensor arrays or robotic arms

Source: nasa-jpl/open-source-rover — based on the README and release notes.

CADAM v0.3.0 adds parametric controls for browser-based text-to-CAD generation 🔗

Users now adjust dimensions via interactive sliders while exporting to STL, SCAD, or DXF formats in-browser

Adam-CAD/CADAM · TypeScript · 5.1k stars Est. 2025

CADAM’s latest release introduces real-time parametric controls, letting users tweak model dimensions with sliders after generating CAD from natural language or images. Built with TypeScript and WebAssembly, it runs entirely in the browser, leveraging BOSL, BOSL2, and MCAD libraries for mechanical precision.

The tool supports exporting complex assemblies like V8 engines or turbofans as STL, SCAD, or DXF without installation. Recent commits show active agent updates and bug fixes, though documentation lags behind feature growth.
The catch: Complex models with many parameters can slow browser performance, limiting usability on lower-end devices.

Previously in The Times “covered” — Aug 26

Use Cases
  • Mechanical engineers rapidly prototype engine components from text descriptions
  • Robotics hobbyists generate custom gearboxes and export directly to STL for printing
  • Educators demonstrate parametric design principles using browser-based CAD without software setup

Source: Adam-CAD/CADAM — based on the README and release notes.

Quick Hits

ros2_documentation Provides the official documentation for ROS 2, enabling builders to understand and implement robot middleware effectively. 998
ros2_control Offers a flexible, hardware-agnostic control framework for ROS 2, simplifying robot actuation and sensor integration. 992
dingtalk-plugin Enables Jenkins to send build notifications via DingTalk, streamlining team communication in CI/CD pipelines. 363
ardupilot Delivers a mature, open-source autopilot suite supporting drones, rovers, submarines, and planes across diverse robotic platforms. 15.8k
ros2_control_demos Supplies hands-on examples demonstrating ros2_control usage, helping builders learn and prototype robot control systems quickly. 841

Wazuh v4.14.7 drops legacy database daemon for leaner security stack 🔗

Manager cluster stability improves with better buffer handling and API thread limits

wazuh/wazuh · C++ · ▲ 10 in 1d Est. 2015 · Latest: v4.14.7

Wazuh’s latest release removes the deprecated wazuh-dbd daemon and its database_output configuration, streamlining the manager’s architecture by eliminating legacy database overhead. The update strengthens cluster reliability with improved payload buffer allocation, archive decompression limits, and file path validation across worker nodes.

API authentication gains stability through bounded thread pools, regex timeouts, and payload size limits, reducing crash risks under load. Python dependencies like aiohttp, cryptography, and PyJWT are updated to patch known vulnerabilities. On the agent side, fixes resolve AWS SQS subscriber misconfigurations in cross-account setups, correct eBPF-based file integrity monitoring gaps on older kernels, and prevent false disconnections after system clock rollbacks. These changes refine Wazuh’s core functionality: real-time log analysis, intrusion detection via signature and anomaly checks, and file integrity monitoring that tracks changes in content, permissions, and ownership—now enhanced with user and application attribution. The platform continues to unify XDR and SIEM capabilities across cloud, container, and on-prem workloads, forwarding agent data to a central manager for rule-based correlation and alerting via the Wazuh Indexer.
The catch: Despite its breadth, Wazuh’s C++-agent architecture can be resource-intensive on constrained endpoints, and its rule complexity may require significant tuning for low-false-positive operations in dynamic environments.

Previously in The Times “covered” — Aug 29

Use Cases
  • Detect malware and rootkits on Linux servers
  • Monitor cloud workloads for misconfigurations and breaches
  • Audit file changes and user activity in containerized apps

Source: wazuh/wazuh — based on the README and release notes.

More Stories

Sherlock Project Cuts False Positives in Latest OSINT Tool Release 🔗

v0.16.0 adds Debian/Ubuntu community packages and automated site validation to reduce noise

sherlock-project/sherlock · Python · ▲ 59 in 1d Est. 2018

The Sherlock Project released v0.16.

0, significantly reducing false positives through automated testing of social network targets. Community-maintained packages now support Debian and Ubuntu 22.10+, easing installation for Linux users. New flags like --json and --no-txt enhance output flexibility for forensic workflows. The tool drops Python 3.8/3.9 support, requiring ^3.10 for future development. Contributors expanded validation logic via upstream exclusions to improve accuracy across 400+ platforms.
The catch: Despite reduced noise, regional variations and evolving site structures mean false positives persist, requiring manual verification in high-stakes investigations.

Previously in The Times “covered” — Aug 27

Use Cases
  • Infosec analysts checking username availability across platforms
  • Red teams gathering open-source intelligence during engagements
  • Developers automating OSINT reconnaissance in security toolchains

Source: sherlock-project/sherlock — based on the README and release notes.

Shannon 3.0 debuts agentic SAST for real exploit validation 🔗

Open-source AI pentester now generates SARIF reports and native CI/CD integration

KeygraphHQ/shannon · TypeScript · ▲ 71 in 1d 11mo old

KeygraphHQ’s Shannon 3.0 launches agentic static application security testing (SAST) that analyzes source code, identifies attack paths, and executes real exploits to confirm vulnerabilities.

The TypeScript-based tool outputs SARIF-format findings and integrates natively with CI/CD pipelines via GitHub Actions. Users launch via an interactive CLI to run local pentests, with only verified exploits appearing in professional PDF reports. Shannon remains distinct from Keygraph’s commercial platform, offering full open-source access for self-hosted security testing.
The catch: Active exploit execution requires a running target application, limiting use in pre-deployment or air-gapped environments without live endpoints.

Previously in The Times “covered” — Aug 28

Use Cases
  • Developers validate API security before merging pull requests
  • Security teams automate penetration testing in CI/CD workflows
  • Auditors generate SARIF-compliant reports for compliance tracking

Source: KeygraphHQ/shannon — based on the README and release notes.

Infisical Adds Warning Before Root Key Change Prevents Admin Lockouts 🔗

Latest patch v0.165.2 alerts users before deactivating encryption keys in self-hosted deployments

Infisical/infisical · TypeScript · ▲ 29 in 1d Est. 2022

Infisical’s v0.165.

2 release introduces a safety check for administrators managing self-hosted instances. Before forcing deactivation of a root encryption key, the system now warns users about potential data loss. This addresses a risk where premature key rotation could render secrets inaccessible across environments. The change appears in pull request #7912, part of ongoing admin UX hardening. Teams using Infisical for secret syncing with GitHub, AWS, or Kubernetes benefit from reduced misconfiguration risk during infrastructure updates. The feature reflects a focus on operational safety in open-source secret management tools.
The catch: Open issues remain high at 768, suggesting ongoing stability or feature gaps despite recent usability improvements.

Previously in The Times “covered” — Aug 27

Use Cases
  • DevOps teams syncing PostgreSQL credentials across staging and production
  • Security engineers rotating AWS IAM keys via automated secret workflows
  • Platform admins preventing accidental lockout during encryption key maintenance

Source: Infisical/infisical — based on the README and release notes.

Quick Hits

SafeLine Self-hosted WAF and reverse proxy that protects web apps from attacks and exploits in real time. 22.5k
opennhp Lightweight open-source toolkit enforcing Zero Trust security for infrastructure, apps, and data using cryptography in AI-driven environments. 13.9k
dirsearch Fast Python-based web path scanner that discovers hidden directories and files on web servers. 14.7k
keepassxc Cross-platform, community-driven password manager that securely stores and manages credentials across devices. 28.7k
vuls Agent-less vulnerability scanner for Linux, FreeBSD, containers, WordPress, libraries, and network devices with minimal setup. 12.2k

ClickHouse v26.3 LTS Boosts Real-Time Analytics with Cloud-Native Enhancements 🔗

Release adds distributed query optimizations and improved embeddings support for AI workloads

ClickHouse/ClickHouse · C++ · 49.6k stars Est. 2016 · Latest: v26.3.29.7-lts

ClickHouse’s latest long-term support release, v26.3.

29.7-lts, sharpens its edge in real-time OLAP by refining distributed query execution and tightening integration with AI pipelines. The update improves parallel processing across shards, reducing latency for complex aggregations on petabyte-scale datasets—a key upgrade for teams running continuous analytics on streaming data.

Embedded machine learning workflows gain traction through enhanced support for vector embeddings, allowing users to store and query high-dimensional data directly within ClickHouse using SQL extensions. This aligns with the project’s push into the “agentic data stack,” highlighted by upcoming events like the Boston Hands-on training on building agents with ClickHouse and LibreChat.

Cloud-native adoption continues via ClickHouse Cloud, the managed service offering, while self-hosted deployments benefit from refined resource isolation and improved backup orchestration. Recent meetups in Bangkok, Cape Town, and Amsterdam underscore growing global traction, particularly in AI-driven data engineering circles.

The catch: Despite performance gains, ClickHouse remains less ideal for transactional workloads due to its column-oriented design, which can slow point updates and deletes compared to row-based OLTP systems.

Use Cases
  • Power real-time dashboards for ad-tech bidding systems
  • Analyze IoT sensor streams at millisecond granularity
  • Build AI agents with integrated vector search and SQL querying

Source: ClickHouse/ClickHouse — based on the README and release notes.

More Stories

Vercel Labs' Native SDK replaces web runtimes with Zig for true desktop apps 🔗

Declarative .native views and TypeScript logic compile to native binaries without browser overhead

vercel-labs/native · Zig · 7.6k stars 3mo old

Vercel Labs' Native SDK enables developers to build native desktop applications using declarative markup in .native files and TypeScript (or Zig) logic, compiled to native code via a Zig-powered engine.

The toolkit eliminates browser runtimes, WebViews, and JS interpreters from the final binary, delivering true native performance with expressive authoring. Recent work completed macOS distribution signing with notarization and secure timestamping, improving release reliability. Developers can iterate quickly with native dev for live reload and native check for instant validation.
The catch: The project is still early-stage (v0.10.1) with 154 open issues, indicating unresolved stability and platform maturity concerns for production use.

Use Cases
  • Desktop developers building cross-platform apps with native performance
  • Teams avoiding Electron's memory overhead while retaining web-like dev speed
  • Engineers preferring TypeScript or Zig for logic without sacrificing UI expressiveness

Source: vercel-labs/native — based on the README and release notes.

Scrcpy v4.1 adds VP8/VP9 support and fixes encoder constraints 🔗

Update upgrades FFmpeg, SDL, libusb while maintaining low-latency mirroring

Genymobile/scrcpy · C · 148.8k stars Est. 2017

Genymobile/scrcpy v4.1 adds VP8 and VP9 video encoder support, updates terminal title during runtime, and improves size constraint handling for flex display.

The release upgrades FFmpeg to 8.1.2, SDL to 3.4.12, and libusb to 1.0.30, fixing data races, gamepad detection, and colorspace conversions. It enables Android screen mirroring and control via USB or TCP/IP without root, using keyboard and mouse, with audio forwarding on Android 11+. The catch: Open issues remain high at 2896, indicating ongoing maintenance challenges despite recent activity.

Use Cases
  • Developers testing Android apps on desktop
  • Users mirroring phone to PC for presentations
  • Gamers playing mobile titles with keyboard/mouse

Source: Genymobile/scrcpy — based on the README and release notes.

RAGFlow Adds Azure DevOps, You.com, Serply Connectors for Enterprise Data Access 🔗

New release expands data source integrations while improving retrieval API transparency and metadata filtering

infiniflow/ragflow · Go · ▲ 88 in 1d Est. 2023

RAGFlow’s latest release introduces three key data source connectors: Azure DevOps for work item and repository integration, You.com for AI-powered web search in chat and agent workflows, and Serply for lightweight search augmentation.

These join existing Confluence, S3, and Notion sync options, broadening enterprise ingestion paths. Retrieval APIs now expose rerank_candidates_count, knn top_k, and num_candidates for tuning, while metadata filters are pushed down to the index for faster queries. The project remains Go-based with growing agent orchestration features.
The catch: Despite rapid feature growth, 1,577 open issues suggest ongoing stability challenges in complex multi-agent deployments.

Previously in The Times “covered” — Aug 27

Use Cases
  • Developers sync Jira tickets and code repos via Azure DevOps for context-aware coding agents
  • Support teams augment internal knowledge with real-time You.com web search during troubleshooting
  • Enterprises automate cross-platform data ingestion from Confluence, Notion, and cloud storage into RAG pipelines

Source: infiniflow/ragflow — based on the README and release notes.

Quick Hits

moby Assembles container-based systems by enabling collaborative development of container ecosystem tools for builders. 72k
PowerToys Supercharges Windows productivity and customization with a suite of powerful, open-source utilities for power users. 138.3k
redis Delivers ultra-fast, feature-rich caching, data structures, and vector search for real-time data-driven applications. 76.2k
obs-studio Enables free, open-source live streaming and screen recording with professional-grade control and flexibility. 75.8k
kubernetes Provides production-grade container orchestration for scalable, resilient deployment and management of cloud-native apps. 126.1k

RealSense SDK Adds Zero-Copy GPU Frames for Jetson CUDA Workflows 🔗

New CUDA frame access cuts latency on integrated GPUs for robotics and vision pipelines

realsenseai/librealsense · C++ · ▲ 4 in 1d Est. 2015 · Latest: v2.58.4

The librealsense library now enables zero-copy GPU frame access on Jetson platforms, letting CUDA and TensorRT consume depth and color streams directly without host-to-device copying. This update, part of v2.

58.4, introduces rs2_get_frame_gpu_data and the rs2::gpu_frame extension, targeting developers building real-time perception pipelines on NVIDIA’s embedded GPUs.
Beyond GPU efficiency, the release adds Ubuntu 26.04 and ROS2 Lyrical support, expands D401 camera GMSL dual-RGB routing, and exposes device hardware time via rs2_get_device_time_ms. The SDK also rebrands "inference" APIs to "perception," aligning with Intel’s Perception Studio tooling.
Builders using the C++ or web viewers gain a searchable control filter and viewport grid overlay, while JetPack 7.2 compatibility is now baked into the L4T patch script.
The catch: Zero-copy GPU frames require integrated-GPU Jetson hardware and recent L4T builds, excluding discrete GPU workstations and older Jetson modules from the performance gain.

Previously in The Times “covered” — Sep 2

Use Cases
  • Robotics teams fusing depth with CUDA-based object detection
  • Drone developers running low-latency SLAM on Jetson Orin
  • Industrial vision systems triggering actions via GPU-accelerated frame analysis

Source: realsenseai/librealsense — based on the README and release notes.

More Stories

C++ roadmap updates with Miro deprecation notice 🔗

Project shifts focus to draw.io and SVG formats after Miro sync halted

salmer/CppDeveloperRoadmap · Python · ▲ 2 in 1d Est. 2021

The salmer/CppDeveloperRoadmap project now flags its Miro version as obsolete, last synced ~July 2026, directing users to draw.io and SVG outputs generated from a language-neutral source.

Built in Python, the roadmap structures C++ learning paths from fundamentals to commercial-grade skills, covering topics like AI integration, coding standards, and pet-project ideas. It remains a reference for self-study under CC BY-NC-SA 4.0 license.
The catch: Open issues linger unresolved for two years, raising questions about active maintenance despite recent format updates.

Use Cases
  • Junior devs structuring C++ self-study
  • Teams aligning on internal training paths
  • Educators sourcing curated learning resources

Source: salmer/CppDeveloperRoadmap — based on the project README.

Automotive Skills Suite Adds 76 Confirmation Reviewers to Claude for Safety Workflows 🔗

Builder-reviewer skill pairs now automate ISO 26262, cybersecurity, and SPICE deliverables with visual KPI dashboards

jherrodthomas/automotive-skills-suite · Python · ▲ 2 in 1d 4mo old

The jherrodthomas/automotive-skills-suite project expanded its Claude skill library to 152 installable .skill files, pairing each of 76 engineering builder skills with a matching confirmation reviewer.

Reviewers generate Excel-based outputs consumed as stable contracts by downstream skills in the "chain," enabling end-to-end automation of APQP, FMEA, PPAP, and AUTOSAR workflows. Users trigger skills via natural language in Claude Desktop or Cowork, with reviewer outputs including KPI tiles, charts, and findings tables. The chain enforces file-format contracts between skills, reducing manual handoffs in functional safety and V&V processes. The catch: 12 open issues and reliance on Claude’s skill framework raise questions about long-term maintainability and vendor lock-in for enterprise adoption.

Previously in The Times “covered” — Aug 29

Use Cases
  • Systems engineer automates ISO 26262 safety case generation
  • Quality lead runs AIAG-VDA APQP to PPAP workflow
  • Cybersecurity engineer executes TARA to IR plan traceability

Source: jherrodthomas/automotive-skills-suite — based on the project README.

Quick Hits

espectre Detects motion through Wi-Fi signal changes using CSI analysis and integrates with Home Assistant for smart home automation. 9.3k
CatSniffer Sniff, communicate with, or attack IoT devices across multiple radio bands using a portable USB stick with multi-protocol hardware. 937
Simple-SolarTracker-Arduino Automatically adjusts solar panel orientation using Arduino-compatible controllers for optimal energy capture without internet or cloud dependency. 39
LibreHardwareMonitor Monitors real-time PC hardware metrics including temperature, fan speed, voltage, load, and clock speeds via free, open-source software. 9k
litex Enables rapid hardware design and FPGA-based system creation through a Python-driven, open-source hardware construction framework. 4.1k
Simple-Hydroponics-Arduino Automates hydroponic grow systems with Arduino-compatible controllers for precise nutrient, light, and water management — offline and fully configurable. 95

1-Click DLSS 5 Injects Neural Upscaling Into Any PC Game With One Click 🔗

Supports DLSS 5, DLAA, and ray reconstruction across DX11, DX12, and Vulkan titles on RTX 20–50 series GPUs

reiluisii/1-Click-DLSS5 · HLSL · 126 stars 5d old · Latest: v2.6.0

1-Click DLSS 5 simplifies neural rendering injection by automating game detection, API inspection, and payload deployment through a native 64-bit executable. It scans Windows Registry paths for Steam, Epic, GOG, and Xbox apps, then uses PE import table inspection to identify D3D11, D3D12, Vulkan, or OpenGL calls without relying on local DirectX DLLs.

Once detected, it redirects upscaling requests—whether from native DLSS, FSR 2/3, or XeSS—into NVIDIA’s DLSS-NR neural model via proxy redirection.

The tool offers three injection modes: native DLSS support, FSR/XeSS redirection, and Lumenite Kernel-based optical flow for games lacking any upscaler. It includes 100% native DLAA feeder mode, OptiScaler bridge integration, and synthetic DLAA scaling. A streamlined HUD v2 provides real-time feedback and automated Mode 1 selection for compatible titles.

Built in HLSL and released as 1-Click-DLSS5.exe, it embeds a high-res icon, Windows 11 manifest, and Per-Monitor V2 DPI scaling. The latest v2.6.0 added the executable and addressed various fixes.

The catch: Despite broad API support, injection relies on proxy DLL techniques that may trigger false positives in anti-cheat systems, limiting safe use to single-player or offline titles.

Use Cases
  • Enable DLSS 5 in legacy DX11 games lacking native support
  • Replace FSR 2/3 with DLSS-NR for improved image quality
  • Add ray reconstruction to Vulkan titles without developer integration

Source: reiluisii/1-Click-DLSS5 — based on the README and release notes.

More Stories

EnTT v4.0.0 drops legacy helpers, adds concepts for modern C++ ECS 🔗

Breaking changes streamline core APIs while maintaining header-only, dependency-free design

skypjack/entt · C++ · ▲ 4 in 1d Est. 2017

EnTT’s v4.0.

0 release removes deprecated utilities like entt::identity and ENTT_CONSTEXPR, replacing them with concepts such as enum_bitmask and cvref_unqualified. The update refines type_info and component_traits to favor concepts over SFINAE, improving compile-time safety. View-to-view conversion now works across all view types, and storage destruction sees performance gains for trivially destructible types. Despite these changes, EnTT remains a header-only, no-dependency ECS used in production by Mojang and Esri. The catch: Concept-heavy refactors may increase compile times and require C++20, limiting adoption in legacy codebases.

Use Cases
  • Game studios build high-performance entity systems
  • Graphics engines manage component data efficiently
  • Simulation tools iterate over entities with views and groups

Source: skypjack/entt — based on the README and release notes.

Babylon.js Adds WebXR Persistent Anchors and Room Capture in Latest Release 🔗

Release 9.23.0 enhances immersive web experiences with improved spatial mapping and shader management

BabylonJS/Babylon.js · TypeScript · ▲ 8 in 1d Est. 2013

Babylon.js 9.

23.0 introduces WebXR persistent anchor lifecycle and room capture support, enabling more stable AR/VR placements and environmental understanding. The update also adds budget-driven, multi-camera-aware LOD for streaming Gaussian Splatting, improving performance in complex scenes. Shader loading state is now shared across material instances, reducing redundant compilations. WebGPU integration preserves WebXR light estimation for consistent lighting in mixed reality. These features target developers building immersive web apps requiring real-world spatial awareness.
The catch: WebXR features remain limited to browsers with experimental support, and production CDN use is discouraged, requiring self-hosting for deployment.

Previously in The Times “covered” — Aug 26

Use Cases
  • Develop AR/VR apps with persistent object placement in real-world spaces
  • Render large-scale 3D scenes efficiently using adaptive level-of-detail streaming
  • Build cross-platform immersive experiences targeting WebGL, WebGPU, and WebXR devices

Source: BabylonJS/Babylon.js — based on the README and release notes.

Godot 4.7.2 fixes bugs, improves stability across desktop, mobile, and web exports 🔗

Maintenance release addresses usability issues without breaking compatibility with prior 4.x versions

godotengine/godot · C++ · ▲ 18 in 1d Est. 2014

Godot Engine’s 4.7.

2 update resolves stability and usability bugs reported since 4.7.0, refining the editor and export pipelines for Linux, macOS, Windows, Android, iOS, and HTML5. The release includes fixes for scene inheritance glitches, input handling on touch devices, and WebGPU rendering inconsistencies. Developers can upgrade directly from any 4.x version without project conversion. Despite active maintenance, the engine’s C++ core retains complexity that can challenge newcomers seeking lightweight 2D workflows.
The catch: While exports are one-click, mastering advanced features like custom shaders or multiplayer networking still requires significant engine-specific learning.

Previously in The Times “covered” — Aug 29

Use Cases
  • Indie developers shipping 2D platformers to Steam and itch.io
  • Mobile studios building casual games for Android and iOS
  • Educators teaching game design with a free, open-source engine

Source: godotengine/godot — based on the README and release notes.

Quick Hits

cyclopsLevelBuilder Lets you block in levels directly inside the Godot editor for faster, visual level design. 1.6k
comedot Provides a component-based framework and template to streamline 2D game development in Godot. 526
bevy Offers a refreshingly simple, data-driven approach to game development using Rust. 48k
OpenGamepadUI Delivers an open-source, gamepad-native launcher and overlay for seamless game access. 945
netfox Adds multiplayer functionality to Godot projects with easy-to-use networking addons. 1.1k
The Git Times AI Desk
Ask about today's stories — or hit “Ask about this” on any article to focus on one.

Unlock the Git Times AI desk to ask about today's stories and the AI model market.

Upgrade to Premium
Answers by the Git Times AI desk · verify before you ship