Design
Preset
Background
Text
Font
Size
Width
Account Pricing

The Git Times

AI Models
Claude Opus 5.5 $20/M —GPT-6.1 Sol $10/M —Nano Banana 2.1 $7.50/M —Grok 4.7 $6/M —DeepSeek V4 Pro 0813 $1.98/M —Qwen3.8 Max Prime $12/M —Kimi K3 $10/M —
Full Markets →

Lightpanda’s Zig-Powered Browser Outpaces Chromium for AI Automation 🔗

Delivers 16x lighter, 9x faster performance via CDP without Chromium forks or patches

lightpanda-io/browser · Zig · ▲ 72 in 1d Est. 2023 · Latest: 1.0.0

Lightpanda, the headless browser written in Zig, is gaining traction among developers building AI agents and automation workflows. Unlike Puppeteer or Playwright, which rely on Chromium forks, Lightpanda is a ground-up implementation designed specifically for programmatic control.

It exposes the Chrome DevTools Protocol (CDP) server on port 9222, allowing existing automation clients to connect seamlessly while avoiding the bloat of full browser engines. Benchmarks show it requests 933 real web pages on an AWS EC2 m5.large instance with significantly lower memory and CPU overhead.

The project’s latest release introduces meaningful Web API advancements: WebAssembly.compileStreaming, Sanitizer API initial support, TextEncoder.encodeInto, and experimental ServiceWorker integration behind a flag. CORS is now enforced by default with tunable limits, and WebDriver support has been expanded for session management, element interaction, and cookie handling. These updates position Lightpanda as a viable, lightweight alternative for CI/CD pipelines, scraping, and AI-driven browser automation where startup speed and resource efficiency matter.

The catch: Lightpanda lacks native Windows and Android builds, requires glibc (excluding Alpine/musl), and its Web API coverage remains incomplete — critical for complex SPAs or legacy sites needing full browser fidelity.

Why this leads today Lightpanda introduces a headless browser built in Zig specifically for AI automation, offering a new tool that enables developers to script and scale web interactions in ways previous solutions did not support.

Use Cases
  • AI agents scraping dynamic content at scale
  • CI/CD pipelines running fast headless tests
  • Automation scripts needing low-memory browser control

Source: lightpanda-io/browser — based on the README and release notes.

More on the Front Page

jhaizhou-ops/d2r-mobile Brings Switch Diablo II: Resurrected to iPhone, iPad, and Mac Natively 🔗

Runs original Switch game code at 60 fps with full touch controls and Reign of the Warlock expansion on Apple silicon

jhaizhou-ops/d2r-mobile · C · 1k stars 1d old

The jhaizhou-ops/d2r-mobile project is an open-source compatibility layer that enables players to run their own Nintendo Switch copy of Diablo II: Resurrected with the Reign of the Warlock expansion natively on iPhone, iPad, and Mac. Rather than emulating, it executes the game's existing aarch64 binary directly on Apple hardware, rewriting only the handful of instructions incompatible with Apple's kernel while leaving all other code and data unchanged.

Graphics are reimplemented via Metal with offline shader translation, and input, audio, filesystem, and system calls are handled by a minimal host layer. Touch controls are fully rebuilt for edge-to-edge iPhone and true full-screen iPad display, with controller support, power saver mode, and save sync. The latest release, v1.0.6, improves turning feel (r42) by eliminating one-frame stutter on sharp turns and increasing minimum turning speed from 180 to 360 degrees per second when using a gamepad. It also patches small memory leaks from short-lived threads, text-input dialogs, and graphics layer components, and now logs memory warnings on iOS. The no-Xcode build route now supports older devices down to A16 chips.
The catch: The project requires users to provide their own legally obtained Switch game files; it does not include or distribute any game assets, and setup involves manual steps like replacing the shell or running ./setup.sh for Xcode builds, which may pose a barrier for less technical users despite the polished end result.

Use Cases
  • Play Diablo II: Resurrected on iPhone with touch controls
  • Run the Switch version natively on Apple silicon Mac
  • Access Reign of the Warlock expansion on iPad without emulation

Source: jhaizhou-ops/d2r-mobile — based on the README and release notes.

Paperclip v2026.1012.0 lets external AI assistants join agent teams as signed-in collaborators 🔗

Claude Code, Codex, and MCP clients can now review work and delegate tasks via experimental setting

paperclipai/paperclip · TypeScript · 101.4k stars 7mo old

Paperclip’s latest release enables outside assistants like Claude Code and Codex to connect as signed-in users, reviewing agent output and assigning tasks through an experimental flag. The update includes refreshed harnesses for Claude Sonnet 5.

5 and GPT-6.1 Sol, improved crash recovery via SQLite file locks, and reduced default agent instructions. Teams can now orchestrate heterogeneous agents—OpenClaw, Hermes, Gemini—under shared budgets, goal alignment, and heartbeat-driven workflows from a single dashboard.
The catch: Workspace restore locks now require draining all old processes before upgrade, risking downtime in multi-host deployments.

Use Cases
  • Dev teams coordinate Claude Code and Codex agents on shared feature goals
  • Managers track costs and permissions across 20+ autonomous agent terminals
  • Enterprises audit AI agent work via mobile dashboard with role-based access

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

FreeLLMAPI routes 7.4B free tokens monthly through one OpenAI-compatible endpoint 🔗

Smart failover and encrypted key storage now work on legacy x86 CPUs after v0.13.10 fix

tashfeenahmed/freellmapi · TypeScript · 34.5k stars 5mo old

FreeLLMAPI aggregates 34 free LLM providers and 635 model endpoints behind a single /v1 API, routing requests to the best available service while tracking per-key usage to stay within free-tier limits. The latest release v0.

13.10 resolves server crashes on older x86 hardware like Synology NAS boxes by lazy-loading the image-resizing library only when needed. Developers can point any OpenAI-compatible client at a local instance to access free tiers from dozens of providers without managing multiple SDKs or rate limits.
The catch: Personal experimentation only; production use violates providers' free-tier terms and risks abrupt service loss.

Use Cases
  • Developers testing LLM apps without API key management
  • Researchers comparing responses across 474 model families
  • Educators running AI demos on low-cost hardware like Raspberry Pi

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

DSH Ops Toolkit Slashes Token Waste by Optimizing Shell Execution 🔗

Uses standalone Bash and Rust tools to reduce AI overhead on Windows, Linux, and macOS

T-Auto/dsh-ops · JavaScript · 827 stars 5d old

T-Auto/dsh-ops is a DSH plugin that cuts token usage and task time by providing isolated Bash 5.3.

15 and PowerShell 7.6.6 environments, prioritizing Bash to avoid Windows shell conflicts. It integrates native Rust utilities for faster file operations and reduces prompt bloat through progressive tool definition. The plugin maintains DSH’s sandbox and permission model while skipping unnecessary I/O and error-induced token waste. Latest release v0.2.9 fixes bugs in bash payload handling, parameter validation, and WSL fallback paths.
The catch: The 0.3.0-native Rust PTC components require manual build and explicit activation, with platform validation still pending for Linux and macOS.

Use Cases
  • AI developers reducing token costs in DSH workflows
  • Teams standardizing shell execution across Windows, Linux, macOS
  • Builders minimizing shell-related errors in automated tasks

Source: T-Auto/dsh-ops — based on the README and release notes.

Rust DAW SoundCraft delivers full Pro Tools workflow across OS and browser 🔗

v0.6.1 adds shared docking, MIDI sizing, and CLI fixes for audio engineers

storytold/soundcraft · Rust · ▲ 318 in 1d 6d old

SoundCraft reimplements Avid Pro Tools in pure Rust, offering multitrack editing, a full mixer with plugins, MIDI, and browser-based WASM builds. The latest release improves workflow with shared docking persistence, refined MIDI track sizing, and CLI robustness when stdout is closed.

It supports signed builds for macOS, Windows, Linux, FreeBSD, and WebAssembly, enabling consistent audio production across platforms. Every action is scriptable via command line or AI agents, appealing to engineers seeking automation. Built-in synths and AudioSuite processing mirror familiar Pro Tools paradigms.
The catch: Despite 192 open issues and rapid commits, long-term stability and plugin compatibility remain unproven at professional scale.

Previously in The Times “covered” — Oct 9

Use Cases
  • Audio engineers multitracking live bands on Linux studios
  • Composers arranging MIDI scores with built-in synths in browser
  • Podcasters automating vocal chains via CLI scripts on macOS

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

Rust reimplementation of PowerPoint gains CLI and AI agent control 🔗

DeckCraft v0.6.1 adds Azure signing to release workflow and MCP integration

storytold/deckcraft · Rust · ▲ 470 in 1d 6d old

DeckCraft, the open-source Rust clone of Microsoft PowerPoint, now lets CLI scripts and AI agents drive every button via MCP. The v0.

6.1 release refactors CI workflows using shared actions and secures releases with Azure signing. It runs natively on macOS, Windows, Linux, FreeBSD, and in browsers, supporting themes, animations, charts, and full PPTX compatibility. Presentations can be built, edited, and presented with transitions, morph effects, and presenter view. The project remains in alpha, targeting beta per its roadmap.
The catch: As an alpha reimplementation, complex PowerPoint features like advanced SmartArt or macros may render incorrectly or lack full fidelity.

Previously in The Times “covered” — Oct 9

Use Cases
  • Developers automate deck generation via CLI in CI/CD pipelines
  • AI agents create and edit presentations using natural language through MCP
  • Cross-platform teams present decks natively on Windows, macOS, Linux, and FreeBSD

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

PostHog adds mobile speech-to-text and report-saving features in latest release 🔗

Update includes AI-powered session replay insights and task-splitting for client-specific workflows

PostHog/posthog · Python · 40.3k stars Est. 2020

PostHog’s desktop-v0.61.

686 release introduces real speech-to-text for mobile composer dictation and lets users save durable report chat feedback. MobileHog now filters reports by status, renders markdown images, and shows last opened chats in search. Developers can split task sources by client and trace errors to exact code lines. The platform combines analytics, feature flags, and session replay to help teams diagnose issues and ship fixes faster.
The catch: With over 5,500 open issues, maintaining stability across its expanding toolset remains a challenge for adopters at scale.

Use Cases
  • Product teams analyze user behavior with autocaptured events
  • Engineers feature-flag releases to test changes safely
  • Support agents replay sessions to resolve customer issues quickly

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

Self-Hosted Web Frameworks Surge for AI-Enhanced Data Workflows 🔗

Developers prioritize data sovereignty by building local-first tools that integrate AI processing without external dependencies.

Trendweb-frameworks
snapotter-hq/SnapOtterGoogleChrome/modern-web-guidance-srcspree/spreetwentyhq/twentyfast-crud/fast-crudkoel/koelWebKit/WebKitMetaMask/metamask-mobilePostHog/posthogmealie-recipes/mealieLimeSurvey/LimeSurveyNVIDIA/TensorRT-LLMGoogleCloudPlatform/generative-aiNaiboWang/EasySpidercurl/curljstrieb/github-statsddalcu/mlx-servephotobooth-app/photobooth-app

A clear pattern emerges in open source web frameworks: projects are combining self-hosted architectures with AI capabilities to process multimedia and document data entirely on-premises. SnapOtter exemplifies this, offering a TypeScript-based UI and API for local OCR, transcription, and AI-powered file conversion across images, video, and PDFs, ensuring data never leaves the user’s network.

Similarly, fast-crud provides a Vue3-driven, configuration-oriented CRUD framework enabling lightning-fast development of super tables for data-intensive applications, while mealie uses Vue3 for a reactive frontend to manage recipes via URL imports and AI-assisted metadata extraction. Beyond content, twenty positions itself as an open, AI-native alternative to Salesforce, embedding AI agents directly into CRM workflows. Even niche tools like photobooth-app pair Python backends with Vue3 frontends for local media processing, reflecting a broader shift toward full-stack, self-contained systems. These frameworks reject cloud dependency by design, favoring REST APIs, TypeScript SDKs, and modular pipelines that let developers own their data pipelines and avoid vendor lock-in. The trend signals a maturation of edge-capable web tools where AI augmentation happens client-side or on local servers, driven by privacy demands and the need for offline, compliant workflows in regulated industries.
The catch: While promising, many of these tools remain fragmented in their AI integrations—relying on disparate local models or requiring significant setup to run LLMs efficiently—and lack standardized benchmarks for real-world performance, making enterprise adoption risky without deeper tooling maturity.

Use Cases
  • Developers building local AI pipelines for document OCR and transcription
  • Teams creating self-hosted CRM platforms with embedded AI agents
  • Individuals managing personal data like recipes or media without cloud reliance

AI Agents Break Out of Chat Into Real-World Action 🔗

Open source shifts from prompting models to controlling hardware, networks, and enterprise workflows autonomously

Trendai-agents
mhtsec/ARTEXonehumanbeing/esp32-iphone-agenttrhhubsd/ARTEXXpaperclipai/paperclipSnailclimb/JavaGuidepatchy631/ai-engineering-hubGoogleCloudPlatform/generative-aiAI4Finance-Foundation/FinRLcurl/curlddalcu/mlx-serve

The open-source AI agent movement is evolving beyond conversational interfaces into systems that perceive, decide, and act in physical and digital environments. Projects like onehumanbeing/esp32-iphone-agent demonstrate this shift: using a $7 ESP32 chip, it enables any OpenAI-compatible vision model to control a real iPhone via AirPlay and USB HID—no Mac, jailbreak, or developer account required—turning AI into a tangible operator.

In cybersecurity, mhtsec/ARTEX and its fork trhhubsd/ARTEXX showcase autonomous penetration testing agents that plan and execute network attacks without human intervention, having won Baidu’s “agent+” offensive-defensive challenge. Enterprise adoption is rising via paperclipai/paperclip, a TypeScript app designed to manage and orchestrate AI agents in workplace settings, treating them as deployable resources. Meanwhile, educational and integrative efforts like patchy631/ai-engineering-hub and GoogleCloudPlatform/generative-ai provide notebooks for building real-world agent applications using LLMs and RAG, while ddalcu/mlx-serve enables efficient, native LLM inference on Apple Silicon using Zig—stripping away Python dependencies for performance and deployment flexibility. These repos collectively signal a pattern: agents are no longer confined to chatbots; they’re gaining senses, limbs, and authority to act across domains.
The catch: Despite rapid progress, most agents remain brittle in unpredictable environments, lack standardized safety guards, and struggle with long-term coherence—making today’s demos impressive but production deployment risky without significant engineering oversight.

Use Cases
  • Security teams automate network penetration testing
  • Developers control iPhones via low-cost AI vision
  • Enterprises orchestrate multiple AI agents in workflows

Open Source Dev Tools Shift Toward Unified, High-Performance Workflows 🔗

Projects integrate Rust speed, GUI abstraction, and terminal efficiency to streamline coding across stacks

Trenddev-tools
oxc-project/oxcDevolutions/UniGetUIbtwld/fvmgitkraken/vscode-gitlensjunegunn/fzfjesseduffield/lazygitdeathbeam/dotfiles

A clear pattern is emerging in open source dev tools: the convergence of high-performance backends, cross-layer abstraction, and terminal-native workflows to reduce friction in modern development. Rust-powered tooling like oxc-project/oxc delivers lightning-fast JavaScript parsing and transformation, proving that systems-language speed can now underpin frontend toolchains without sacrificing accessibility.

Simultaneously, GUI layers are evolving beyond wrappers — Devolutions/UniGetUI acts as a universal frontend for disparate package managers (apt, brew, choco, etc.), abstracting complexity while preserving backend flexibility. This mirrors the rise of version managers like btwld/fvm, which lets Flutter teams switch SDK versions instantly via CLI, eliminating environment drift.

Terminal efficiency remains a focal point, with junegunn/fzf enabling fuzzy-finding across files, histories, and streams through minimal keystrokes, and jesseduffield/lazygit transforming complex Git operations into an intuitive, keyboard-driven TUI. Even visualization is being reimagined: gitkraken/vscode-gitlens embeds deep Git insights directly into the editor via blame annotations and CodeLens, turning repository navigation into a discovery process. Meanwhile, niche but telling efforts like deathbeam/dotfiles (in GLSL) hint at developers treating environment configuration as code — even shader-based prompts — signaling a broader shift toward programmable, personalized workflows.

The catch: While these tools showcase impressive technical alignment, the trend risks fragmentation — combining Rust backends with Electron GUIs or shell-dependent CLIs creates integration seams, and abstraction layers like UniGetUI may leak complexity when package managers diverge in behavior or permissions, leaving the "unified workflow" ideal more aspirational than realized today.

Use Cases
  • Frontend teams accelerate builds with Rust-based JS tooling
  • DevOps engineers manage multiple package managers through one GUI
  • Flutter developers switch SDK versions instantly per project

Deep Cuts

XRobot Turns Natural Language into Working Robot Code Across Platforms 🔗

Generates firmware and simulation scripts for ESP32, Raspberry Pi, and STM32 from plain English prompts

xrobot-org/XRobot · Python · ▲ 1 in 1d

XRobot lets builders describe robot behaviors in natural language and outputs ready-to-flash code for MCUs, Linux SBCs, and simulators like Webots. It handles sensor integration, motor control, and state machines without manual boilerplate.

The toolkit abstracts hardware differences, letting the same prompt target an ESP32 line-follower or a Raspberry Pi-powered robomaster with minimal tweaks. By embedding domain-specific patterns for embedded robotics, it cuts prototyping time from hours to minutes. Developers avoid wrestling with HAL libraries or simulator APIs—XRobot maps intent to functional C++ or Python output. It’s not magic, but it reduces the cognitive gap between idea and execution on real hardware.
The catch: It’s early-stage, with limited documentation and community support, making adoption risky for production-critical systems despite its promise in labs and classrooms.

Use Cases
  • Hobbyist programs ESP32 line follower using voice-to-code
  • Educator generates Webots sim scripts for classroom swarm demos
  • Engineer prototypes STM32 arm kinematics without writing HAL code

Source: xrobot-org/XRobot — based on the project README.

Quick Hits

twenty Provides a customizable, AI-ready open-source CRM alternative to Salesforce in TypeScript. 58.3k
oxc Delivers high-performance JavaScript tooling for parsing, transforming, and analyzing code in Rust. 23k
gridcraft Offers a fast, pure Rust spreadsheet engine with Excel-like functionality and clean-room design. 4.3k
esp32-iphone-agent Enables AI models to control a real iPhone via low-cost ESP32 hardware using vision and HID interfaces. 1.3k
doory Provides an open-source, TypeScript-based design tool for creating graphics like Canva, self-hosted and extensible. 985
TensorRT-LLM Offers an optimized Python API for building and deploying LLMs with NVIDIA GPU acceleration and runtime orchestration. 14.8k
SnapOtter Lets users process files locally with AI-powered OCR, transcription, conversion, and compression via UI or API without leaving their network. 2.8k
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

Hugging Face Transformers Adds Multimodal EmbeddingGemma2 for Cross-Modal Retrieval 🔗

The new model encodes text, image, audio, and video into a shared 768D space with truncatable dimensions.

huggingface/transformers · Python · ▲ 63 in 1d Est. 2018 · Latest: v5.19.0

The latest release of Hugging Face Transformers (v5.19.

  1. introduces EmbeddingGemma2, a multimodal embedding model from Google based on the Gemma 4 architecture. It processes text, images, audio, and video — individually or combined — into a unified 768-dimensional vector space. This enables cross-modal retrieval, semantic similarity, clustering, and classification tasks. Thanks to Matryoshka Representation Learning, embeddings can be truncated to 512, 256, or 128 dimensions without retraining, offering flexibility for storage and latency-sensitive applications. Visual and video token budgets are configurable, and unused modalities can be disabled at load time to conserve memory. The model integrates seamlessly with existing pipelines and supports inference via the high-level Pipeline API. Transformers continues to act as the central model-definition framework, ensuring compatibility across training frameworks like Axolotl and DeepSpeed, inference engines such as vLLM and TGI, and tools like llama.cpp. With over 1M+ model checkpoints on the Hugging Face Hub, it remains a key conduit for deploying state-of-the-art models.
    The catch: While EmbeddingGemma2 supports flexible dimensionality, its performance on truncated embeddings for niche cross-modal tasks — such as fine-grained audio-video alignment — has not been widely benchmarked, leaving uncertainty for specialized use cases.

Previously in The Times “covered” — Oct 11

Use Cases
  • Developers building cross-modal search over text and images
  • Engineers creating audio-visual content recommendation systems
  • Researchers clustering multimodal datasets for analysis

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

More Stories

Hermes Agent v0.21.7 adds channel control for source update workflows 🔗

Users can now pin to stable releases or track main branch via desktop app settings

NousResearch/hermes-agent · Python · ▲ 321 in 1d Est. 2025

The latest patch shifts source install behavior from tracking every commit to defaulting to tagged releases like `v0.21.

7, reducing update noise. A new *Update channel* toggle in Settings ▸ About lets users switch between stable releases and the main branch without CLI flags. Reasoning models now resume correctly after hermes chat --resume`, fixing prior session breaks. Firewall-related update failures are patched by avoiding release record fetches that triggered 403s. The catch: Open issues exceed 48k, signaling ongoing stability and scalability challenges despite recent fixes.

Previously in The Times “covered” — Oct 7

Use Cases
  • Developers testing agents on serverless infrastructure
  • Teams deploying persistent AI workflows across cloud VMs
  • Individuals syncing agent sessions via Telegram or Discord
  • Researchers generating batch trajectories for LLM evaluation
  • Operators scheduling unattended backups and audits in natural language
  • Engineers spawning isolated subagents for parallel tool pipelines
  • Users avoiding vendor lock-in by switching LLMs via `hermes model`
  • Builders running agents on $5 VPS with hibernating environments
  • Professionals utilizing FTS5 search for cross-session conversation recall
  • Adopters integrating with agentskills.io-standard skill marketplaces
  • Users leveraging voice memo transcription in multi-platform chats
  • Teams utilizing Honcho dialectic modeling for personalized agent behavior

Source: NousResearch/hermes-agent — based on the README and release notes.

AI Engineering Hub delivers 93 production-ready projects for LLM and agent builders 🔗

Repository updated daily with new tutorials on RAG, fine-tuning, and local AI workflows

patchy631/ai-engineering-hub · Jupyter Notebook · 38.3k stars Est. 2024

The patchy631/ai-engineering-hub repo hosts 93+ production-ready projects spanning LLMs, RAG systems, and AI agents, organized by difficulty from beginner OCR apps to advanced fine-tuning pipelines. Built in Jupyter Notebook, it includes working examples like Local ChatGPT with Llama 3.

2 vision and Gemma-3 OCR for structured text extraction. Recent commits show active maintenance, with the last push just one day ago.
The catch: Despite frequent updates, 125 open issues suggest ongoing stability challenges in scaling examples to production environments beyond demo workflows.

Use Cases
  • Learn LLM integration by building local OCR apps with Llama 3.2
  • Develop RAG systems using Qwen 2.5 VL for document processing
  • Experiment with AI agent workflows via DeepSeek Thinking UI implementations

Source: patchy631/ai-engineering-hub — based on the project README.

TensorFlow 2.21.0 drops Python 3.9 support, adds JPEG XL decoding 🔗

Release removes TensorBoard dependency while expanding lite operator type support

tensorflow/tensorflow · C++ · ▲ 19 in 1d Est. 2015

TensorFlow 2.21.

0 drops Python 3.9 support and removes the TensorBoard dependency, signaling a shift toward leaner deployments. The update adds JPEG XL support in tf.image.decode_image and expands int2/int4 type handling across tf.lite operators like SQRT, EQUAL, and SLICE. These changes optimize model size and inference speed for edge devices. Despite active development, the project carries 3,408 open issues, reflecting ongoing complexity in maintaining broad hardware and language compatibility.
The catch: Supporting legacy Python versions and external tools like TensorBoard now requires manual workarounds or older TF releases.

Previously in The Times “covered” — Oct 6

Use Cases
  • Train vision models using JPEG XL for faster image pipelines
  • Deploy int2-quantized models on microcontrollers via TensorFlow Lite
  • Build GPU-accelerated ML apps with CUDA on Ubuntu or Windows systems

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

Quick Hits

ColossalAI ColossalAI optimizes large AI model training and inference to reduce costs, speed up computation, and broaden accessibility for developers and researchers. 41.4k
generative-ai This repository provides hands-on Jupyter notebooks demonstrating how to build and deploy generative AI applications using Google Cloud’s Gemini Enterprise Agent Platform. 17.8k
fastbook fastbook delivers the full fastai deep learning textbook as interactive Jupyter notebooks, enabling practical, code-first learning of modern AI techniques. 25.4k
ai-agents-for-beginners A structured 18-lesson curriculum teaching beginners how to design, build, and deploy AI agents using practical tools and real-world patterns. 77.1k
mcp-for-beginners Learn to build modular, scalable, and secure AI workflows using the Model Context Protocol (MCP) through cross-language examples in .NET, Java, TypeScript, JavaScript, Rust, and Python. 17.4k

Kornia v0.8.3 Adds XFeat, ALIKED Features and Half-Precision Support 🔗

Release boosts geometric vision accuracy and GPU efficiency for PyTorch-based spatial AI pipelines

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

Kornia’s latest patch release introduces XFeat and ALIKED as new local feature detectors and descriptors, expanding its toolkit for robust image matching under varying conditions. The update enables batched inference for VisualPrompter, streamlining prompt-based vision tasks in deep learning workflows.

Core geometry operations now see meaningful speedups: depth_to_normals runs ~3x faster, while transform and denormalize_points_with_intrinsics improve by ~2x, reducing latency in real-time spatial AI applications. Half-precision (float16/bfloat16) support has been strengthened with updated test infrastructure, ensuring numerical stability on modern GPUs and Apple Silicon via MPS fixes. ONNX export reliability has also improved, easing model deployment across platforms. Over 175 commits since v0.8.2 include docstring expansions, RGB↔YUV conversion accuracy fixes, and adaptive 3D interpolation enhancements. Deprecated utilities in kornia.utils now trigger DeprecationWarning, signaling upcoming cleanup.

The catch: Despite broad coverage, Kornia’s reliance on PyTorch limits flexibility for teams using TensorFlow or JAX, and its extensive API surface may overwhelm newcomers seeking lightweight vision utilities.

Previously in The Times “covered” — Oct 8

Use Cases
  • Real-time SLAM systems using GPU-accelerated geometric transformations
  • Training vision models with automated, differentiable data augmentation pipelines
  • Deploying feature matching pipelines (LoFTR, LightGlue) via ONNX for edge inference

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

More Stories

Text-to-CAD Plugin Lets Agents Generate 3D Models from Natural Language 🔗

Latest release fixes STEP export crashes and Windows worker console issues in v0.7.21

earthtojake/text-to-cad · Python · ▲ 305 in 1d 5mo old

The earthtojake/text-to-cad plugin enables AI agents to create 3D models in STEP, GLB, STL, or 3MF formats using plain text prompts. It integrates with Claude Code, Codex, Cursor, Gemini, and Grok agents via skills or plugin frameworks, triggering local CAD generation through uv-managed Python runtime.

The tool also performs design-for-manufacturing checks, outputs engineering drawings, and connects to 3D printing, sheet metal, and CNC services. Recent v0.7.21 resolves a STEP tail crash, first-snapshot browser race, and eliminates noisy Windows worker consoles. Telemetry is opt-in, sending only anonymized usage and crash data.
The catch: The plugin requires a network connection on first run to download CAD runtime and assumes agent compatibility with plugin or skills frameworks, limiting use in air-gapped or restricted environments.

Previously in The Times “covered” — Oct 9

Use Cases
  • Mechanical engineers generate brackets from text descriptions
  • Robotics teams prototype grippers via agent commands
  • Designers export STL files directly for 3D printing workflows

Source: earthtojake/text-to-cad — based on the README and release notes.

iDynTree Enables Floating-Body Robot Dynamics Without Model Reloading 🔗

Latest release v15.1.0 patches Python CI and formatting after 12+ years of active development

gbionics/idyntree · C++ · 234 stars Est. 2014

iDynTree provides C++ multibody dynamics algorithms tailored for free-floating robots like humanoids, with Python and MATLAB bindings. Its undirected graph model lets engineers switch base links for kinematics and dynamics without reloading URDF/SDFormat files or altering serializations—a key advantage for whole-body control research.

The library supports URDF/SDFormat I/O, mixed/body/inertial representations, and joint torque estimation without collocated sensors, as used in the iCub humanoid. Recent work focused on CI hygiene and version bumping to v15.1.0.
The catch: 196 open issues suggest unresolved bugs or feature gaps despite recent activity.

Use Cases
  • Humanoid robot control base-link switching
  • URDF-based dynamics parameter identification
  • Joint torque estimation without physical sensors

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

Pinocchio Adds URDF v1.2 Support for Acceleration and Jerk Limits 🔗

New ModelTpl fields and CMake options improve robot dynamics modeling fidelity

stack-of-tasks/pinocchio · C++ · ▲ 1 in 1d Est. 2014

The latest release v4.1.

0 extends Pinocchio’s URDF parser to interpret <limit> tags for acceleration and jerk, enabling tighter motion constraints in simulations. Users can now specify min_acceleration, max_acceleration, min_jerk, and max_jerk when adding joints, with corresponding fields in ModelTpl. A new CMake option, PINOCCHIO_BUILD_BINDING_WITH_PCH, accelerates Python binding compilation via precompiled headers. Fixes include proper handling of -inf/+inf values in model loading and corrected Viser visualizer mesh scaling. Despite its maturity, Pinocchio’s C++ core demands expertise for real-time tuning, and its dependency on Eigen and FCL may complicate embedded deployment.
The catch: Real-time performance tuning requires deep familiarity with its templated algorithms and build system.

Use Cases
  • Simulating humanoid robots with joint jerk limits
  • Gradient-based control using analytical derivatives
  • Prototyping motion planners via Python-Conda bindings

Source: stack-of-tasks/pinocchio — based on the README and release notes.

Quick Hits

stable-baselines3 Implements reliable reinforcement learning algorithms in PyTorch for training agents in complex environments. 13.9k
Robots Enables creation and simulation of industrial robot programs from ABB, KUKA, UR, and Staubli in C#. 389
openpilot Provides an open-source operating system that enhances driver assistance in over 300 supported vehicles. 63.9k
ros2_rust Offers Rust bindings for ROS 2, enabling high-performance robotics development with memory safety. 1.5k
autoware_universe Delivers a comprehensive open-source autonomous driving platform for perception, planning, and control in C++. 1.8k

Nuclei v3.11.1 Adds Lua Script Args for Deeper Vulnerability Logic 🔗

Community-driven scanner gains thread-safe metadata caching to reduce false positives in parallel scans

projectdiscovery/nuclei · Go · ▲ 26 in 1d Est. 2020 · Latest: v3.11.1

ProjectDiscovery’s nuclei vulnerability scanner updated to v3.11.

1 with a key enhancement: Lua scripts now accept arguments and values, enabling more dynamic and context-aware detection logic. This allows security teams to parameterize checks—for example, varying payloads based on target technology or environment—without duplicating templates. The release also includes a bug fix that reuses metadata cache across thread-safe scans, improving efficiency and reducing redundant requests during large-scale assessments. These changes support nuclei’s core promise: fast, accurate scanning via YAML templates that simulate real-world exploit steps. Written in Go, nuclei scans applications, APIs, networks, DNS, and cloud configs using protocols like HTTP, TCP, DNS, and SSL. It integrates with CI/CD pipelines and tools like Jira, Splunk, and GitLab, backed by thousands of community-contributed templates. Despite its speed and flexibility, nuclei remains a CLI-first tool; running it as a service is discouraged due to potential security risks, requiring extra hardening for production use.
The catch: While nuclei excels at automated detection, complex logic still relies on user-written Lua or JavaScript templates, creating a barrier for teams without scripting expertise who may prefer point-and-click scanners with built-in vulnerability logic.

Previously in The Times “covered” — Oct 10

Use Cases
  • Security engineers automate CVE scanning in CI/CD pipelines
  • Red teams simulate real-world attack steps across internal networks
  • AppSec teams validate cloud config drift using custom YAML templates

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

More Stories

Cilium’s eBPF dataplane replaces kube-proxy for scalable Kubernetes networking 🔗

Latest release v1.20.2 fixes AWS IPAM bugs and policy-deny-response metrics

cilium/cilium · Go · 25.6k stars Est. 2015

Cilium delivers eBPF-based networking, security, and observability for Kubernetes, replacing kube-proxy with efficient hash tables for near-unlimited scale. It enforces L3-L7 network policies via identity-based models and supports service mesh, gateways, and bandwidth management.

The v1.20.2 patch resolves AWS IPAM allocation failures and restores metrics for policy-denied egress traffic. The catch: Open issues exceed 1,100, indicating ongoing complexity in debugging eBPF integration across diverse kernel versions.

Use Cases
  • Platform engineers securing multi-cluster pod communication
  • SREs replacing kube-proxy to reduce latency at scale
  • Teams enforcing L7 HTTP policies without sidecar proxies

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

RustScan Delivers Faster Port Scans with Automated Nmap Piping 🔗

Latest release adds UDP/TCP benchmarks and Rust library for embedded use

bee-san/RustScan · Rust · 20.6k stars Est. 2020

RustScan 2.4.

1 enhances speed guarantees through automated CI benchmarks for UDP and TCP scans, ensuring no pull request degrades performance. The scanner now includes a Rust library, enabling developers to embed port scanning directly into applications. It maintains its core strength: scanning all 65,535 ports in ~3 seconds and piping results into Nmap or custom scripts (Python, Lua, Shell). Adaptive learning refines scans over time using basic math, not machine learning. Docker and ARM support broaden deployment flexibility.
The catch: Open issues persist at 18, and the project’s adaptive learning lacks transparency in how scan history improves future scans.

Use Cases
  • Network engineers scanning internal networks for open ports
  • Security teams automating reconnaissance before penetration tests
  • Developers integrating port scanning into custom security tools via Rust library

Source: bee-san/RustScan — based on the README and release notes.

SWE-agent v1.1.0 adds multilingual support and SWE-smith training data 🔗

Breaking changes include trajectory format updates and tool bundle renames for compatibility

SWE-agent/SWE-agent · Python · 20.5k stars Est. 2024

SWE-agent v1.1.

0 introduces multilingual evaluation support and integration with SWE-smith, a new project generating tens of thousands of training trajectories. These updates enable SWE-agent-LM-32b to achieve open-weights state-of-the-art performance on SWE-bench verified. The release includes breaking changes: the messages field in trajectory data is replaced by query, several tool bundles using "windowed" file viewers are renamed, and review_on_submit is removed in favor of review_on_submit_m. The project remains focused on enabling language models to autonomously fix GitHub issues, find vulnerabilities, or tackle custom tasks via a configurable YAML file.
The catch: Despite performance gains, the project’s complexity and frequent breaking changes may hinder adoption for teams seeking stable, long-term tooling.

Previously in The Times “covered” — Oct 6

Use Cases
  • Developers automating GitHub issue resolution using LLMs
  • Researchers benchmarking agent performance on SWE-bench
  • Security teams testing offensive cybersecurity capture-the-flag challenges

Source: SWE-agent/SWE-agent — based on the README and release notes.

Quick Hits

pyWhat pyWhat identifies emails, IP addresses, and other data types from text or .pcap files, giving builders instant insight into unknown inputs. 7.3k
ProxmoxVE Proxmox VE Helper-Scripts simplify deployment and management of Proxmox VE environments with automated community-driven tools. 29.8k
quivr Quivr turns continuous content streams into searchable, monitorable knowledge with durable ingestion, hybrid search, and plugin-extensible alerts. 39.6k
opencti OpenCTI provides a centralized platform for collecting, structuring, and analyzing cyber threat intelligence to power proactive defense. 10.1k
caddy Caddy delivers automatic HTTPS and extensible HTTP/1-2-3 serving across platforms, making secure web deployment effortless for developers. 77.9k

ClickHouse 26.7 Release Call Highlights Lakehouse and AI Workload Gains 🔗

July 23 session showed real-time analytics scaling with Langfuse agent observability in cloud-native deployments

ClickHouse/ClickHouse · C++ · ▲ 21 in 1d Est. 2016 · Latest: v26.7.29.12-stable

ClickHouse’s July 23 release call detailed performance gains in v26.7 for lakehouse architectures and AI-driven workloads.

Engineers demonstrated columnar storage optimizations that cut query latency by up to 40% on semi-structured data when paired with object stores like S3. The session highlighted integration with Langfuse for tracing LLM agent operations, enabling developers to correlate prompt throughput with backend database response times. Cloud-native deployments now support automated scaling groups via Kubernetes operators, reducing manual cluster tuning. Official docs were updated with step-by-step guides for setting up agent observability in New York and Menlo Park trainings slated for October. The project maintains its C++ core while experimenting with Rust-based extensions for custom table functions, though these remain in early access.
The catch: Real-time analytics at scale still require significant operational expertise in tuning merge trees and managing disk I/O spikes during heavy ingest, posing a barrier for teams without dedicated database administrators.

Previously in The Times “covered” — Oct 10

Use Cases
  • Power real-time dashboards for financial trading platforms
  • Enable lakehouse analytics on semi-structured IoT sensor data
  • Trace LLM agent performance with integrated observability tools

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

More Stories

Go Language Team Merges Major Compiler Optimization for Faster Builds 🔗

Latest commit improves inlining logic, reducing binary size by up to 15% in cloud-native workloads

golang/go · Go · ▲ 43 in 1d Est. 2014

The Go project’s main branch received a significant update 1 day ago, refining the compiler’s inline heuristic to prioritize small, frequently called functions. This change, part of the ongoing Go 1.

24 development cycle, aims to cut build times and memory usage in microservices and serverless environments. Contributors addressed over 200 open issues related to performance regressions in large module graphs. The update does not alter language syntax or break existing code.
The catch: Optimization gains vary by workload, with minimal impact on CPU-bound numerical code.

Use Cases
  • DevOps engineers building containerized APIs
  • Backend teams deploying high-frequency trading systems
  • Cloud architects optimizing function-as-a-service pipelines

Source: golang/go — based on the project README.

Fzf patches security flaws in shell integration and tmux popup handling 🔗

Latest release fixes command injection and focus event bugs in fuzzy finder

junegunn/fzf · Go · ▲ 10 in 1d Est. 2013

Fzf v0.74.

5 patches two security issues: fzf-tmux no longer evaluates layout size via bash arithmetic, preventing command injection, and zsh completion now uses private temp directories for FIFO files. The update also resolves a tmux 3.8+ popup unzooming issue and fixes focus event triggering after reloads when items share keys. Despite its speed and broad shell integrations, the tool remains complex to configure for advanced workflows.
The catch: Deep customization requires learning fzf’s event-driven architecture and key binding syntax, which can deter builders seeking plug-and-play terminal enhancements.

Use Cases
  • Developers fuzzy-finding files in large codebases
  • Sysadmins searching command history with preview
  • Neovim users creating custom interactive menus

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

Quick Hits

zed Zed enables real-time, collaborative code editing with low-latency performance, built by Atom and Tree-sitter creators for faster, smarter development workflows. 91.7k
RuView RuView uses commodity WiFi signals to deliver real-time spatial mapping, vital sign tracking, and presence detection — no cameras or video required. 97.7k
rtk rtk is a zero-dependency Rust CLI proxy that cuts LLM token usage by 60–90% on routine dev commands through intelligent prompt optimization. 83.2k
codegraph codegraph maintains a locally hosted, auto-updating code knowledge graph that reduces token consumption and tool calls for AI coding agents across major platforms. 73.9k
llama.cpp llama.cpp provides efficient, native C/C++ LLM inference — enabling high-performance model execution on consumer hardware without external dependencies. 131k

OpenIPC firmware expands chip support beyond HiSilicon to 11 SoC families 🔗

Nightly build adds Anyka, GrainMedia, SigmaStar, and XiongMai to growing hardware compatibility list

OpenIPC/firmware · C · ▲ 2 in 1d Est. 2021 · Latest: nightly

OpenIPC’s alternative IP camera firmware now runs on chips from Ambarella, Anyka, Fullhan, Goke, GrainMedia, Ingenic, MStar, Novatek, SigmaStar, XiongMai, and HiSilicon. The project, active since 2021, shifted from HiSilicon-only support to multi-SoC enablement as developers ported drivers and Buildroot configurations.

Last night’s build (sha: 86ccd6f) compiled a web UI hashed at cf35234e, reflecting ongoing nightly integration. Contributors add features via patches, improve documentation, or fund development through Open Collective. Paid commercial support offers prioritized bug fixes and maintenance for business users, while free help flows via Telegram chat. The firmware replaces vendor-locked software with an open stack including U-Boot, Linux, and custom RTSP/ONVIF services.
The catch: 181 open issues and reliance on community patches mean stability varies across newer SoCs like SigmaStar and Anyka, requiring builders to validate hardware-specific features like night vision or PTZ control before deployment.

Previously in The Times “covered” — Oct 12

Use Cases
  • Developers flash open firmware on Anyka-based IP cams
  • Businesses deploy paid-supported OpenIPC on GrainMedia chips
  • Hobbyists add ONVIF support to legacy XiongMai devices

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

More Stories

Automotive Skills Suite adds real-time KPI dashboards to Claude workflows 🔗

Builder-confirmation reviewer pairs now generate visual metrics tables for safety and quality artifacts

jherrodthomas/automotive-skills-suite · Python · 3.2k stars 5mo old

The jherrodthomas/automotive-skills-suite project updated its skill pairs to include live KPI dashboards in reviewer outputs, displaying charts and findings tables for ISO 26262, ISO/SAE 21434, and AIAG-VDA deliverables. Each builder skill now triggers a matching confirmation reviewer that validates structured Excel outputs and renders them as interactive tiles within Claude Desktop.

Skills remain installable .skill files triggered by natural language prompts, maintaining the chain’s file-format contract between upstream and downstream tasks. The catch: Open issues number nine, with no recent commits addressing core workflow integration gaps like cross-tool traceability beyond Excel artifacts.

Previously in The Times “covered” — Oct 9

Use Cases
  • Automotive engineers validate ISO 26262 safety cases using Claude
  • Quality teams automate APQP/PPAP documentation review in workflows
  • Cybersecurity analysts generate TARA reports via skill-triggered analysis

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

Arduino OLED add-on lets MiSTer users display custom text and images via USB 🔗

Supports monochrome and greyscale modes on SSD1322 displays with XBM graphics

venice1200/MiSTer_tty2oled · C · 70 stars Est. 2021

The MiSTer_tty2oled project enables builders to connect an Arduino-driven OLED display to their MiSTer FPGA system over USB, showing real-time terminal output or custom graphics. Using C firmware, it renders text and XBM-format images on SSD1322-based screens, with support for both monochrome and greyscale modes.

Hardware schematics and software sketches are provided for DIY assembly, and graphics can be created using GIMP via documented workflows. Despite its age, the project sees rare but consistent maintenance, with the last commit just days ago and a greyscale update released in v0.2.
The catch: Reliance on Arduino USB host mode limits compatibility to specific boards like ESP32 or ESP8266 with USB OTG support, excluding simpler Arduino models.

Use Cases
  • Retro gamers displaying live MiSTer terminal logs on OLED
  • DIY builders showing custom game stats or artwork during play
  • Developers debugging FPGA cores with real-time text output

Source: venice1200/MiSTer_tty2oled — based on the README and release notes.

Photobooth app adds printer monitoring and Python 3.14 support 🔗

New release v9.0.0 improves camera standby and webcam stability on macOS

photobooth-app/photobooth-app · Python · 328 stars Est. 2022

The photobooth-app now monitors printer status to prevent failed print jobs, logging errors when devices are unavailable. Backend updates include Python 3.

14 compatibility, improved gphoto2 backend customization for ISO, shutter speed, and additional settings per mode, and camera standby mode that halts streaming during inactivity to save power and reduce overheating on DSLRs and high-end webcams. WLED plugin gains new events for thrill_still, thrill_video, and thrill_multicam to trigger custom animations during countdowns. Webcam stability on macOS is fixed, and emojis in print statements are removed to avoid crashes on systems with misconfigured locales. The catch: Printer monitoring remains experimental and may not reliably detect all printer models or networked devices.

Use Cases
  • DIY builders create wedding photobooths with DSLR and Raspberry Pi
  • Event organizers generate GIFs, collages, and boomerangs via touchscreen
  • Developers extend functionality using WLED for LED ring countdown signals

Source: photobooth-app/photobooth-app — based on the README and release notes.

Quick Hits

vdbrink.github.io Provides documentation tips and tricks for home automation tools like NODE-RED and Home Assistant using CSS for clean, accessible guides. 47
maker.js Enables 2D vector line drawing and shape modeling for CNC and laser cutters through a TypeScript-based parametric design library. 2k
glasgow Offers a versatile Python-powered hardware toolkit for electronics prototyping, debugging, and testing — like a Swiss Army knife for makers. 2.2k
hackrf Delivers an affordable, open-source software-defined radio platform for experimenting with wireless signals across wide frequency ranges. 8.2k
pxt-maker Enables MakeCode programming for maker-focused microcontroller boards, extending visual coding to hardware like micro:bit and similar platforms. 130

bgfx Adds WebGPU Support via Dawn Native for Cross-Platform Rendering 🔗

New backend enables modern web graphics on desktop and embedded without browser constraints

bkaradzic/bgfx · C++ · ▲ 2 in 1d Est. 2012

The bgfx rendering library now includes a WebGPU backend powered by Dawn Native, expanding its API-agnostic reach beyond Vulkan, Metal, and DirectX. This addition allows developers to write rendering code once and deploy it across Windows, Linux, macOS, and embedded targets using a modern, low-overhead graphics API traditionally confined to browsers.

The backend is exposed through bgfx’s standard interface, requiring no changes to engine or framework integration.
Support for WebGPU complements existing backends like Direct3D 12 and Vulkan, reinforcing bgfx’s “Bring Your Own Engine” promise for studios targeting multiple platforms from a single codebase. Recent commits show active validation on desktop and Wasm/Emscripten builds, with shader translation handled via SPIR-V cross-compilation.
The catch: WebGPU support relies on Dawn Native, which adds build complexity and may lag behind browser implementations in feature coverage, limiting access to cutting-edge WebGPU extensions.

Previously in The Times “covered” — Oct 7

Use Cases
  • Game studios deploy one renderer across PC, console, and web
  • Engine developers abstract graphics APIs without performance loss
  • CAD tools target desktop and embedded via unified graphics layer

Source: bkaradzic/bgfx — based on the project README.

More Stories

Godot 4.7.2 patch fixes bugs without altering core engine behavior 🔗

Maintenance release ensures compatibility while addressing stability and usability concerns

godotengine/godot · C++ · ▲ 83 in 1d Est. 2014

Godot Engine’s latest maintenance release, 4.7.

2-stable, resolves accumulated bugs and usability issues from prior 4.x versions. It maintains full backward compatibility, letting developers upgrade without reworking projects. The engine continues to support one-click exports to desktop, mobile, web, and consoles under the MIT license. Recent commits show active community involvement, with the last push just hours ago. Despite its maturity, the project remains driven by contributor feedback and transparent roadmap planning.
The catch: Over 18,000 open issues signal ongoing challenges in prioritizing fixes amid feature demands.

Previously in The Times “covered” — Oct 9

Use Cases
  • Indie devs shipping 2D platformers to Steam and itch.io
  • Studios prototyping 3D simulations for architectural walkthroughs
  • Educators teaching game design with zero licensing costs

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

libGDX 1.14.2 fixes font cache and click listener bugs 🔗

Patch resolves GlyphCount reset and UI state issues in Java game framework

libgdx/libgdx · Java · 25.5k stars Est. 2012

libGDX 1.14.

2 addresses two specific bugs: BitmapFontCache.clear() now properly resets glyphCount, and ClickListener no longer gets stuck in an over state when cancelled. The release also improves Delaunay triangulation via Shewchuck’s exact predicates and fixes Tiled map template handling for multiple tilesets. These changes refine core rendering and input systems without altering the framework’s cross-platform Java foundation. Developers using bitmap fonts or interactive UI elements benefit directly from the fixes. The project remains Apache 2.0 licensed and Gradle-friendly for setup.
The catch: Open issues remain at 343, indicating ongoing maintenance needs despite recent activity.

Previously in The Times “covered” — Oct 6

Use Cases
  • Android devs building 2D games with Java and OpenGL ES
  • Desktop creators targeting Windows/Linux/macOS from one codebase
  • Web developers deploying HTML5 games via libGDX’s browser backend

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

gdUnit4 adds mocking and scene testing for Godot 4 developers 🔗

Latest release v6.2.2 fixes memory leaks and improves inspector usability

godot-gdunit-labs/gdUnit4 · GDScript · ▲ 1 in 1d Est. 2022

gdUnit4 enables test-driven development in Godot 4 with embedded test inspection for GDScript and C#. It supports fluent assertions, parameterized tests, and automatic test discovery via the editor context menu.

Recent fixes resolve out-of-memory errors during large string comparisons and allow resizing the inspector dock. The framework integrates directly into Godot’s FileSystem and ScriptEditor panels.
The catch: Open issues exceed recent commits, suggesting maintenance may lag behind feature demand.

Use Cases
  • Validate GDScript game logic with automated unit tests
  • Test C# scripts using fluent assertions and mocking
  • Run scene tests directly from the Godot editor inspector

Source: godot-gdunit-labs/gdUnit4 — based on the README and release notes.

Quick Hits

ShaderToHuman Provides HLSL/GLSL/Slang tools to visualize and debug shader behavior for faster iteration and troubleshooting 680
gozen Offers a lightweight, Godot-powered video editor with a clean interface for simple video editing tasks 469
bevy Delivers a flexible, data-driven game engine in Rust that simplifies game development with modular systems 49k
Depth3D Enables depth-based 3D effects and advanced post-processing shaders for immersive visual experiences 759
Sundial-Lite Implements a free, open-source GLSL version of Sundial for realistic time-of-day lighting and shadows in graphics applications 163
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