Design
Preset
Background
Text
Font
Size
Width
Account Pricing

The Git Times

AI Models
Claude Fable 5.1 $50/M GPT-6 Astra $50/M Gemini 3.1 Pro Preview $12/M Grok 4.6 $6/M DeepSeek V4 Pro 0813 $2.95/M ▼7%Qwen3.8 Max (0902) $6/M Kimi K3 $13.28/M
Full Markets →

Fly Brain Simulation Trades Crypto Using Real Neural Reinforcement Signals 🔗

Simulated dopamine responses from portfolio P&L modify synaptic connections in a retained Drosophila connectome model

nftechie/stonkfly · Python · ▲ 55 in 1d 4d old

The nftechie/stonkfly project simulates a fruit fly’s brain to drive cryptocurrency trading decisions through biologically inspired reinforcement learning. It maps live Coinbase BTC-USDC price data into RGB and color-channel inputs that stimulate 3,335 brightness and 811 color sensors tied to a retained male CNS connectome comprising 166,700 neurons and 25.

6 million synaptic connections. Neural activity generates buy, sell, or hold signals, which are validated and executed via a custom Coinbase AgentKit ActionProvider interfacing with Coinbase Advanced. Profitable trades trigger artificial dopamine release in 15 PAM11 cells, reinforcing synaptic changes from Kenyon cells to mushroom body output neurons (MBONs), while losses activate two aversive PPL101 cells. These engineered reinforcement signals update connection weights in real time, simulating a form of experiential learning without claiming profitability or cognitive modeling of pain or reward. The simulation runs locally with paper trading by default, using a $100 simulated balance and public market data, requiring no API keys. For live trading, users must configure a dedicated Coinbase Advanced portfolio with limited funds and scoped API credentials, then manually opt in. The system logs sensory inputs, neural states, and portfolio changes to enable resumable runs. Built in Python 3.11 with a C++17 dependency for performance-critical components, it demands ~16 GB RAM and several gigabytes for datasets. Despite its novelty in bridging neuromorphic simulation and automated trading, the project explicitly states that profitable learning has not been demonstrated, and synaptic changes do not equate to effective trading strategy acquisition.
The catch: The simulation has not demonstrated profitable learning, and its reinforcement mechanism remains an unvalidated analogy to biological reward systems rather than a proven adaptive trader.

Why this leads today It offers a concrete, biologically inspired framework for testing autonomous decision-making in high-stakes financial systems, bridging neuroscience and practical AI agent safety.

Use Cases
  • Researchers testing neuromorphic models in financial decision loops
  • Developers exploring biologically inspired reinforcement learning systems
  • Educators demonstrating neural simulation interfacing with real-world APIs

Source: nftechie/stonkfly — based on the project README.

More on the Front Page

PostHog’s desktop update smooths AI observability and session replay for developers 🔗

The latest release fixes IPC log forwarding and pty output coalescing in the PostHog Desktop app

PostHog/posthog · Python · 39.8k stars Est. 2020

PostHog’s desktop client now handles terminal output more reliably, preventing garbled logs when AI agents generate traces during LLM-powered app debugging. The update coalesces pty output before IPC transmission and stops forwarding main logs to Windows in production, reducing noise in session replays and error tracking.

Developers using PostHog’s AI observability suite can now see cleaner LLM trace data alongside feature flags, experiments, and autocaptured events—all steerable from Slack, web, or the MCP. The platform continues to unify product analytics, session replay, feature flags, and data pipelines under one self-hostable Python stack, with a generous free tier for each tool.
The catch: Despite frequent commits, the project’s 5,613 open issues suggest ongoing stability challenges in integrating its broad toolset, particularly around real-time data pipeline performance at scale.

Use Cases
  • Debug LLM apps with unified traces and session replays
  • Test feature flags using no-code experiments
  • Analyze user behavior via autocaptured product analytics

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

CodeBurn tracks AI token spend across 41 tools locally 🔗

Desktop v0.9.24 adds Windows Capacity Dock and restores chart states on restart

getagentseal/codeburn · TypeScript · 11k stars 5mo old

CodeBurn reads local session files from AI coding tools like Claude Code, Cursor, and Codex to break down token usage and cost by model, project, and task. The CLI runs instantly via npx codeburn with no install, while the desktop app offers a menubar/tray interface and optional web dashboard.

Pricing data comes from LiteLLM, updated daily, and all processing stays on-device—no data leaves your machine. Recent updates include persistent chart states, honest Copilot credit estimates, and visibility for zero-cost providers. The tool surfaces hidden spend, like excessive conversation tokens or overuse of expensive models. The catch: macOS builds are ad-hoc signed only, requiring manual override to open, and Linux support relies on community-maintained packages.

Use Cases
  • Developers audit AI coding costs per project
  • Teams compare model efficiency across tasks
  • Individuals optimize token usage in CLI workflows

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

SGLang adds day-0 support for Qwen3.8 and Spark2.5 models 🔗

Release v0.5.19 brings 786 PRs from 214 contributors, expanding autoregressive and diffusion workloads

sgl-project/sglang · Python · 35.9k stars Est. 2024

SGLang’s latest release integrates Qwen3.8 (2.

4T-A95B) and Spark2.5 for autoregressive inference, alongside updated cookbooks for Granite 4.2 and Ling-3.0 variants. The framework now serves trillions of tokens daily, accelerated by recent TPU and NVIDIA GB300 NVL72 optimizations. With day-0 support for models like Nemotron 3 Ultra and DeepSeek-V4, SGLang maintains its role in high-throughput LLM serving. The catch: Open issues exceed 5,000, indicating ongoing challenges in stabilizing support across rapidly evolving model architectures and hardware backends.

Use Cases
  • Deploy Llama 3.1 70B on GB200 NVL72 for 4.8x decode throughput
  • Run Qwen-Image diffusion workloads on TPUs via SGLang-Jax backend
  • Serve DeepSeek-V3.2 with sparse attention for low-latency agentic tasks

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

ArchiveTune 15.0.0 brings V10 Editorial Player and SponsorBlock to Android 🔗

Major release adds YouTube Music podcasts, lyrics translation, and Google Drive backups

rukamori/ArchiveTune · Kotlin · 6k stars Est. 2025

ArchiveTune 15.0.

0 overhauls YouTube playback with next-song preloading and SponsorBlock integration. The new V10 Editorial Player improves artwork-driven theming and mini-player interactions. Users gain Spotify/TIDAL Canvas support, Last.fm scrobbling, and CSV playlist export. Built with Kotlin and Jetpack Compose, it prioritizes privacy and offline local file playback. Despite 110 open issues, the project maintains active development with commits within the last day.
The catch: Geographic restrictions require a VPN or proxy for YouTube Music access in unsupported regions, adding complexity for global users.

Use Cases
  • Stream YouTube Music ad-free with background playback on Android
  • Play local audio files with EQ and crossfade controls
  • Sync listening history to ListenBrainz and Last.fm accounts

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

GPT-6 Astra Enables Zero-Shot Control for Robots in Simulation and Real World 🔗

Project groups 32 embodied AI workflows by publication date, newest first

zjwzcx/Awesome-Astra-Embodied-AI · Unknown · 505 stars 2d old

zjwzcx/Awesome-Astra-Embodied-AI curates demos where GPT-6 Astra performs high-level planning for robotics tasks, from pick-and-place to Rubik’s Cube manipulation. It integrates with simulators like Isaac Sim and MuJoCo, translating natural language goals into low-level trajectories via planners such as GEAR-SONIC.

The repo organizes 10 simulation, 10 real-world, and additional cases including agentic policy calls and real-to-sim replay. Builders can study how Astra enables zero-shot dexterous control and navigation without task-specific training. The catch: All demos are simulation-based or limited to narrow real-world trials, with no evidence of long-term autonomy or cross-robot generalization.

Use Cases
  • Robotics researchers test zero-shot control in Isaac Sim
  • Engineers validate navigation trajectories on Unitree G1
  • Developers benchmark dexterous manipulation with HumanCLAW-Bench

Source: zjwzcx/Awesome-Astra-Embodied-AI — based on the project README.

Apache Spark scales data pipelines with unified engine for SQL, ML, and streams 🔗

Recent commits show active maintenance despite project's 12.6-year age and Scala core

apache/spark · Scala · 44k stars Est. 2014

Apache Spark powers large-scale data processing with a unified engine supporting SQL, machine learning via MLlib, graph analytics through GraphX, and stream processing. Its high-level APIs in Scala, Java, Python, and R let engineers write complex pipelines once and run them across clusters.

Built with Maven, Spark scales from local testing to production clusters using YARN or Kubernetes. Recent activity includes a commit zero days ago, indicating ongoing development.
The catch: Open issues remain high at 538, suggesting ongoing stability or compatibility challenges in active development.

Use Cases
  • Data engineers run ETL jobs on petabyte-scale datasets
  • Data scientists train ML models using Spark MLlib
  • Analysts query live streams with Spark Structured Streaming

Source: apache/spark — based on the project README.

Lean Formalization Proves Finite-Time Blowup in Navier-Stokes Equations 🔗

Uses Lean 4.34.0-rc2 and Mathlib to certify Millennium Prize Problem alternatives

openai/NavierStokesAndEuler · Lean · ▲ 20 in 1d 6d old

OpenAI’s NavierS and Euler results are now formally verified in Lean 4, certifying that smooth initial data can lead to singularities in finite time for both whole space and periodic domains. The project leverages Mathlib and Lake for reproducibility, with builds triggered via elan.

Formal proofs cover alternatives (C) and (D) of the Clay Institute’s Navier-Stokes problem, including unbounded C¹ norm and divergent vorticity time integral.
The catch: The formalization assumes specific forcing terms and does not yet cover unforced Navier-Stokes blowup in ℝ³.

Use Cases
  • Mathematicians verifying PDE blowup theorems
  • Engineers validating fluid simulation bounds
  • Logicians testing Lean’s applicability to analysis

Source: openai/NavierStokesAndEuler — based on the project README.

Local-First Web Frameworks Redefine Developer Autonomy 🔗

Projects prioritize offline functionality, minimal dependencies, and direct hardware access without cloud reliance

Trendweb-frameworks
truanayangi-com/truanayangiTracecatHQ/tracecatgetagentseal/codeburnsensepost/gowitnessPostHog/posthogsuperplanehq/superplanemetorial/metorialzauberzeug/niceguiAOrbitron/Eridanuslissy93/web-checkntop/ntopngDatalux/Osintgramtauri-apps/taurijstrieb/github-statsddalcu/mlx-serveplaycanvas/engine

A clear pattern emerges in open-source web frameworks: a shift toward local-first, self-contained tools that eliminate backend, login, or cloud dependencies. truanayangi-com/truanayangi lets users run a meal planner locally with pnpm, storing cookies automatically—no server needed.

Similarly, nicegui enables Python-based web UIs that run entirely client-side, while gowitness uses Chrome Headless in Go to capture web screenshots offline. PostHog extends this ethos by offering self-hosted analytics and observability, letting teams steer product insights from Slack or desktop without external SaaS. Even AI integration follows this model: metorial connects any LLM to 1200+ tools via MCP or CLI, avoiding proprietary agents, and codeburn tracks AI token usage locally across 37 editors, npx-run with zero setup. This isn’t just about convenience—it’s about sovereignty. Developers are choosing frameworks that run on their terms: no network calls, no vendor lock-in, and direct access to system resources (as seen in tauri’s Rust-powered desktop apps or mlx-serve’s native Apple Silicon LLM inference). The pattern signals a maturing desire for tools that are inspectable, auditable, and operable in air-gapped or constrained environments.
The catch: Many of these projects trade scalability and ecosystem maturity for autonomy; local-first designs often lack built-in collaboration features, and debugging distributed state in offline-first apps remains an open challenge—practical for solo builders, less so for teams needing real-time sync.

Use Cases
  • Developers build offline-first productivity tools
  • Teams self-host analytics without external dependencies
  • Engineers run AI workflows locally with full data control

Open Source LLM Tools Shift from Models to Orchestration and Guardrails 🔗

Projects now focus on integrating, optimizing, and constraining AI behavior across modalities and workflows

Trendllm-tools
openai/NavierStokesAndEulerSpaceDudem/text-humanizermizzlelover/gongwen-gbt9704-skillzjwzcx/Awesome-Astra-Embodied-AIgetagentseal/codeburnvllm-project/vllm-omnisgl-project/sglanghyhmrright/brooks-lintAstrBotDevs/AstrBotpipecat-ai/pipecatdatawhalechina/self-llmAOrbitron/Eridanuscolbymchenry/codegraphjustrach/codedbddalcu/mlx-serve

The open source LLM ecosystem is moving beyond raw model releases toward tooling for training and inference into sophisticated layers of orchestration, verification, and human alignment. Frameworks like sglang and vllm-omni optimize high-throughput, multimodal model serving, while pipecat enables real-time voice agent pipelines.

Tooling is increasingly specialized: codeburn tracks token usage and cost across 37 agents, codegraph and codedb provide low-latency local knowledge graphs to reduce redundant tool calls, and mlx-serve offers native Apple Silicon inference without Python dependencies. Guardrails are emerging as a core concern: text-humanizer attempts to evade AI detection, brooks-lint applies engineering literature to AI code review with citation-backed severity scoring, and gongwen-gbt9704-skill automates compliant government document formatting under GB/T 9704-2012. Agent frameworks like AstrBot and Eridanus unify IM platforms, plugins, and function calling via OneBot and LLM-driven orchestration. Even formal verification is entering the fold, as seen in NavierStokesAndEuler, which provides Lean certificates for mathematical PDE results—hinting at AI-assisted proof validation. Together, these projects signal a shift toward trustworthy, efficient, and domain-constrained AI systems built from composable, open components.
The catch: Much of this layer remains experimental, with overlapping functionality, unclear interoperability standards, and limited real-world validation—especially in safety-critical domains—raising questions about whether the fragmentation reflects innovation or premature proliferation.

Use Cases
  • Enterprises reduce AI inference costs via local token tracking
  • Developers enforce coding standards using literature-grounded AI review
  • Governments automate compliant document generation with open LLMs

Open Source Data Infrastructure Shifts Toward Modular, Interoperable Stacks 🔗

Projects prioritize composability and real-time processing over monolithic platforms

Trenddata-infra
apache/sparkrukamori/ArchiveTunemusic-assistant/serverethersphere/beePostHog/posthogscikit-learn/scikit-learnkeras-team/kerasgradio-app/gradioLightning-AI/pytorch-lightningHumanSignal/label-studioprotocolbuffers/protobufGodotNuts/GodotFirebase

A clear pattern is emerging in open source data infrastructure: a move toward modular, interoperable components that can be mixed and matched across the data lifecycle. Rather than all-in-one suites, projects are focusing on doing one thing well while exposing clean APIs for integration.

This is evident in how Apache Spark continues to evolve as a unified engine for batch and stream processing, now with enhanced support for lakehouse architectures via Delta Lake integration. Meanwhile, tools like PostHog demonstrate how observability, feature flags, and analytics are being bundled into a single self-hostable platform — yet designed to plug into existing data pipelines via webhooks and APIs.

On the edge and device side, Music Assistant and ArchiveTune show how local media management and streaming aggregation are being rebuilt as lightweight, extensible servers, often running on Raspberry Pi or NAS devices, emphasizing local control and plugin-based expansion. Similarly, Label Studio provides a flexible, framework-agnostic backend for data annotation that outputs standardized formats, enabling seamless integration into ML workflows built with scikit-learn, Keras, or Gradio.

The infrastructure layer is also seeing innovation: Ethersphere/Bee powers decentralized data storage and retrieval on the Swarm network, offering a permissionless alternative to traditional object storage, while Protocol Buffers remains foundational for efficient, language-agnostic data serialization across services. Even ML orchestration tools like PyTorch Lightning are abstracting hardware complexity to enable scalable training without locking users into specific infrastructure.

The catch: While this modularity promises flexibility, it risks creating integration overhead — teams may spend more time stitching together compatible versions of tools than building actual data products, especially as compatibility matrices grow and documentation lags behind rapid releases.

Deep Cuts

C Library Gives DIY Drones True Inertial Navigation Precision 🔗

Enables standalone position tracking without GPS or external sensors

jnz/INSLIB · C · ▲ 33 in 1d

jnz/INSLIB is a lightweight, dependency-free C library implementing strapdown inertial navigation algorithms. It fuses accelerometer and gyroscope data to estimate 3D position, velocity, and attitude in real time — critical for drones, robots, or wearables where GPS fails.

Builders get dead-reckoning accuracy previously locked behind expensive IMU firmware or ROS packages. The code is portable, deterministic, and MIT-licensed, making it ideal for embedded systems with tight timing constraints. The catch: It remains early-stage with limited documentation and no sensor fusion filters like Kalman, requiring builders to handle noise and drift mitigation themselves.

Use Cases
  • Embedded systems engineer implements indoor drone navigation
  • Robotics hobbyist tracks rover position in GPS-denied warehouses
  • Wearable tech developer measures athlete motion without satellite signals

Source: jnz/INSLIB — based on the project README.

Quick Hits

AstrBot Build and deploy customizable AI agent assistants across IM platforms with LLM integration, plugin support, and open-source flexibility as an OpenClaw alternative. 40.5k
pipecat Create real-time voice and multimodal AI agents with low-latency processing, supporting diverse LLMs and sensory inputs for interactive applications. 15.5k
vllm-omni Achieve high-throughput, low-latency inference for omni-modality models (vision, language, audio) using optimized tensor parallelism and dynamic batching. 6.8k
superplane Automate one-shot engineering workflows by generating production-ready code, tests, and docs from natural language specs via AI-driven agents. 7.3k
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.

OpenClaw v2026.9.4 sharpens local AI with expanded plugin model support 🔗

Release adds 294 contributors’ fixes, improves cross-platform onboarding, and refines plugin swap mechanics for Claude, Codex, and local models

openclaw/openclaw · TypeScript · ▲ 34 in 1d 9mo old · Latest: v2026.9.4

OpenClaw’s latest release focuses on refinement over revolution, tightening the assistant’s ability to run local AI models across devices without compromising data ownership. The v2026.

9.4 update streamlines the installer’s Node.js provisioning, now recommending Node 26 for macOS, Linux, and Windows deployments while maintaining compatibility with existing Node 24.16+ setups. Contributors improved the onboarding wizard to better verify model access and workspace creation, reducing setup friction for users deploying via npm, pnpm, or Bun.

A core architectural strength remains the clean separation between trusted gateway and untrusted execution: models and agent harnesses (including Claude, Codex, and local LLMs) are hot-swappable plugins requiring no reconfiguration. Prompts still route only to user-defined model providers and chat platforms—Discord, iMessage, Slack, Teams, Telegram, WhatsApp, and 20+ others—with no telemetry by default. The release notes highlight 1,558 pull requests and 294 contributors, signaling active community maintenance despite the project’s ~10-month age.

The Control UI, launched post-onboarding, now offers clearer feedback when validating assistant responsiveness via direct message. Docker, Nix, and alternative deployment paths retain full support, and the installer continues to auto-start onboarding on fresh installs.

The catch: While plugin flexibility is a strength, reliance on external model providers (even local ones) means performance and capability vary significantly based on hardware and model choice—OpenClaw doesn’t optimize or quantize models itself, leaving users to manage trade-offs between speed, accuracy, and resource use on their own machines.

Previously in The Times “covered” — Sep 13

Use Cases
  • Developers testing local LLMs across Discord and Slack workflows
  • Teams deploying private AI assistants on shared Linux gateways
  • Individuals running AI in WhatsApp and Telegram without cloud exposure

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

More Stories

Hugging Face Transformers Adds 780B-Parameter MoE Model HYV4 🔗

New release enables inference on mixture-of-experts models with 49B active parameters per token

huggingface/transformers · Python · ▲ 302 in 1d Est. 2018

The latest v5.17.

0 release of 🤗 Transformers adds support for HYV4, a 780B-parameter mixture-of-experts language model that activates just 49B parameters per token. Built with Multi-head Latent Attention, DeepSeek Sparse Attention, gated attention sinks, and independent hyper-connections, HYV4 targets long-context reasoning with a 1M-token window. The update expands Transformers’ role as a unified model-definition layer across training frameworks like Axolotl and DeepSpeed, and inference engines such as vLLM and SGLang. Developers can now load and run HYV4 checkpoints via the Pipeline API after installing the library in a Python 3.10+ environment with PyTorch 2.5+.
The catch: Running HYV4 requires significant GPU memory—estimates suggest over 100GB VRAM for full precision inference, limiting accessibility to well-resourced labs or cloud instances.

Previously in The Times “covered” — Aug 30

Use Cases
  • Researchers experimenting with ultra-large language models
  • Engineers building long-context AI assistants
  • Teams evaluating mixture-of-experts architectures for efficiency trade-offs

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

Scikit-learn 1.9.1 adds Python 3.15 support with bug fixes 🔗

Project maintains weekly commits despite 16-year age and 2k open issues

scikit-learn/scikit-learn · Python · ▲ 2 in 1d Est. 2010

Scikit-learn released version 1.9.

1, its first to support Python 3.15 while addressing several bugs. The update includes no breaking changes but keeps the library current for modern Python environments. Weekly commits continue, with the last push just days ago, showing ongoing maintenance despite the project's 2010 origins. The catch: Over 2,100 open issues suggest unresolved challenges in scaling or feature consistency for complex ML pipelines.

Use Cases
  • Data scientists building classification models
  • Engineers implementing regression analysis
  • Researchers conducting statistical modeling in Python

Source: scikit-learn/scikit-learn — based on the README and release notes.

Fastai simplifies deep learning with high-level and low-level components 🔗

Latest release 2.8.12 adds test data prefetching for faster iteration

fastai/fastai · Jupyter Notebook · 28.1k stars Est. 2017

Fastai provides practitioners with high-level components for state-of-the-art results in image classification, segmentation, text sentiment, recommendation, and tabular modeling — often in just five lines of code. Researchers gain access to low-level, mix-and-match building blocks via its layered architecture.

Built on PyTorch, it supports GPU acceleration and runs seamlessly in Google Colab or via pip install fastai. The library emphasizes usability without sacrificing flexibility or performance, backed by extensive documentation and tutorials.
The catch: Despite recent activity, 270 open issues suggest ongoing challenges in stability and edge-case handling for production-scale deployments.

Use Cases
  • Train image classifiers with minimal code
  • Build recommendation systems using collaborative filtering
  • Experiment with tabular data models for structured datasets

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

Quick Hits

gradio Build and share interactive machine learning apps entirely in Python, no frontend expertise needed. 43.5k
ultralytics Perform real-time object detection, segmentation, pose estimation, and tracking across multiple YOLO variants in a unified framework. 61.6k
pytorch-lightning Scale AI model training from single GPU to thousands with zero code changes using automated distributed training. 31.3k
label-studio Annotate images, text, audio, and video with customizable workflows and export standardized labels for ML pipelines. 28.3k
langchain Develop, test, and deploy AI agents with modular tools for reasoning, tool use, and memory integration. 146.3k

Drake Robotics Toolkit Integrates SNOPT Solver for Advanced Motion Planning 🔗

v1.57.0 bundles pre-compiled SNOPT to accelerate optimization in robotic control workflows

RobotLocomotion/drake · C++ · 4.2k stars Est. 2014 · Latest: v1.57.0

The Drake robotics platform now ships with a pre-compiled version of the SNOPT solver directly in its Mathematical Program toolbox, eliminating a manual setup step for users tackling complex trajectory optimization and inverse kinematics problems. This integration, credited to Philip E.

Gill and Elizabeth Wong, tightens the feedback loop between modeling and verification in model-based robotics design. Engineers can now define and solve nonlinear programs within Drake’s unified C++ framework without external solver configuration, streamlining workflows for dynamic manipulation and legged locomotion tasks.

Built for high-fidelity simulation and real-time control, Drake continues to serve teams needing rigorous guarantees in motion planning, especially where safety-critical validation is non-negotiable. Its strength lies in combining multibody dynamics, systems theory, and optimization under one auditable codebase — a combination still rare in open-source robotics stacks.

The catch: Drake’s C++-centric design and steep learning curve may deter teams prioritizing rapid prototyping or Python-first workflows, limiting accessibility for smaller labs or educational settings despite its industrial strength.

Previously in The Times “covered” — Aug 29

Use Cases
  • Aerospace engineers validating satellite docking maneuvers
  • Robotics labs optimizing quadruped gait transitions
  • Autonomous vehicle teams testing perception-to-control pipelines

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

More Stories

NiceGUI 3.16.0 patches XSS and memory flaws in Python web UI toolkit 🔗

Modular 3D scene refactor introduces breaking changes ahead of v4.0

zauberzeug/nicegui · Python · 16.2k stars Est. 2021

NiceGUI’s v3.16.

0 release fixes two security flaws: unauthenticated memory exhaustion on socket disconnect and XSS via color input prop injection. The update also refactors ui.scene into a modular three.js object system, enabling custom objects but deprecating legacy import patterns and instantiation methods slated for removal in v4.0. Line-anchor tracking lands in ui.codemirror for stable code references. Built for dashboards, robotics, and ML tuning, NiceGUI lets builders craft browser-based UIs in pure Python with live reload.
The catch: The framework’s tight coupling to its internal webserver limits deployment flexibility compared to standalone frontend tools.

Use Cases
  • Engineers tuning motor controllers via browser
  • Data scientists building live ML parameter dashboards
  • Hobbyists creating smart home control panels in Python

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

PX4 v1.17 Adds Altitude Cruise Mode for Steady Multicopter Flight 🔗

Release improves fixed-wing takeoff behavior and ROS 2 control interfaces for rovers and VTOLs

PX4/PX4-Autopilot · C++ · ▲ 6 in 1d Est. 2012

PX4 v1.17 introduces Altitude Cruise mode, letting multicopters maintain velocity and heading when sticks are released instead of hovering.

Fixed-wing takeoff now sustains climb with level wings during navigation loss, using takeoff waypoints to define loiter position. ROS 2 gains cleaner fixed-wing and rover control via new FwLateralLongitudinalSetpointType. Zenoh middleware matures to rmw_zenoh compatibility, and simulation adds Gazebo Jetty support with Ackermann SIH. Three new INS drivers join, alongside Septentrio GNSS resilience reporting and barometer auto-calibration. The catch: Despite active development, 367 open issues suggest ongoing stability challenges in edge-case sensor fusion and middleware integration.

Previously in The Times “covered” — Aug 30

Use Cases
  • Developers testing autonomous survey drones in GPS-denied areas
  • Rover teams integrating PX4 with ROS 2 navigation stacks
  • VTOL pilots requiring reliable transition between hover and forward flight

Source: PX4/PX4-Autopilot — based on the README and release notes.

RobotWebTools updates roslibjs for ROS web clients in TypeScript 🔗

Latest release adds Action.cancelAllGoals() and bumps key dev dependencies

RobotWebTools/roslibjs · TypeScript · 827 stars Est. 2013

The Standard ROS JavaScript Library, roslibjs, released version 2.1.

0 with a new Action.cancelAllGoals() method for ROS 2 action clients. Dependabot-driven updates upgraded TypeScript, Vite, and jsdom to improve tooling and security. Built in TypeScript, it connects web apps to ROS via WebSocket, enabling browser-based robot telemetry and control.
The catch: Its monorepo structure and ROS 1/2 dual support increase complexity for developers targeting only one ROS version.

Use Cases
  • Web developers building ROS dashboards
  • Robotics engineers enabling browser telemetry
  • Researchers prototyping web-based robot interfaces

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

Quick Hits

DynamixelSDK Provides a robust C++ SDK for controlling ROBOTIS Dynamixel servos via Protocol 1.0 and 2.0, enabling precise motor communication and integration in robotics applications. 605
gz-sim Offers an open-source, high-fidelity robotics simulator (Gazebo’s successor) for testing and developing complex robot behaviors in realistic virtual environments. 1.5k
crocoddyl Implements an efficient optimal control library using DDP-based solvers to compute robot motions under complex contact sequences, ideal for legged and manipulation tasks. 1.3k
rtabmap Delivers a real-time SLAM and mapping solution with loop closure detection, supporting 3D LiDAR, RGB-D, and stereo cameras for autonomous navigation. 4k
IsaacLab Presents a unified Python framework for robot learning, leveraging Isaac Sim to streamline reinforcement learning, imitation learning, and domain randomization workflows. 8.1k

OpenCTI Platform Enhances CI Stability with Faster Backend Test Rebuilds 🔗

Latest release replaces MinIO with pgsty/silo in dev stacks and fixes flaky test timeouts

OpenCTI-Platform/opencti · TypeScript · ▲ 4 in 1d Est. 2018 · Latest: 7.260914.0

The OpenCTI Platform’s latest release streamlines development workflows by rebuilding backend test jobs from scratch three times instead of reusing pre-built images, reducing flaky test failures. It also swaps the MinIO image for pgsty/silo in CI and development stacks to improve object storage reliability.

These changes address intermittent 502 errors under load by introducing configurable keep-alive and header timeouts for ALB/ECS deployments. Bug fixes include resolving silent report drops when published dates default to the epoch and fixing live stream share object filtering in segregated environments.

Built on TypeScript with a GraphQL API and STIX2-based knowledge schema, OpenCTI remains a go-to for structuring and visualizing cyber threat intelligence, integrating with tools like MISP, TheHive, and MITRE ATT&CK.

The catch: Despite active development, over 2,100 open issues suggest ongoing stability challenges, and the Enterprise Edition’s feature gap may limit advanced capabilities for budget-conscious teams.

Previously in The Times “covered” — Aug 29

Use Cases
  • Security analysts correlating TTPs with observables
  • SOC teams enriching alerts with threat context
  • Red teams simulating adversary behavior using ATT&CK mappings

Source: OpenCTI-Platform/opencti — based on the README and release notes.

More Stories

Web-check expands OSINT dashboard with WAF detection and vulnerability panel 🔗

Latest update adds subdomain lookup, IPv6 fixes, and config disabling for granular control

lissy93/web-check · TypeScript · 34.8k stars Est. 2023

The TypeScript OSINT tool web-check now includes a WAF detection panel and vulnerabilities section in its dashboard, alongside subdomain lookup and improved UI accuracy. Recent commits fix IPv6 DNS truncation, add OCI labels to Docker, and let users disable specific checks via configuration.

Hosted at web-check.as93.net, it aggregates IP, SSL, DNS, headers, cookies, trackers, and carbon footprint data for any URL. Built for sysadmins and security teams, it deploys via Netlify, Vercel, Docker, or Render.
The catch: Open issues exceed 30, and reliance on third-party APIs like NinjaPear may limit offline or air-gapped use.

Use Cases
  • Security teams audit public-facing sites for misconfigurations
  • Sysadmins verify SSL chains and DNS records post-deployment
  • Developers analyze third-party trackers and performance impacts

Source: lissy93/web-check — based on the README and release notes.

AI pentesting tool Strix adds Exa web search to boost vulnerability discovery 🔗

Latest release improves agent intelligence with real-time external data during scans

usestrix/strix · Python · ▲ 99 in 1d Est. 2025

Strix v1.6.

2 integrates Exa as a web search provider alongside Perplexity, enabling its AI agents to fetch live threat intelligence during penetration tests. The update also fixes CLI import warm-up issues and refines telemetry to classify errors by phase. With 6,830 forks and active development—last commit just one day ago—the tool continues to evolve as a developer-first alternative to manual pentesting. Strix runs locally via Docker using your own LLM key, validating exploits with working proofs-of-concept to avoid false positives. The catch: Reliance on external LLM APIs introduces variable costs and latency, potentially limiting accessibility for teams with strict air-gapped or budget constraints.

Previously in The Times “covered” — Sep 2

Use Cases
  • Developers scanning PRs for runtime vulnerabilities
  • Security teams validating exploits with PoCs
  • Bug hunters automating recon and reporting workflows

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

Osintgram Adds CLI Mode for Faster Instagram Recon 🔗

Output limiter prevents data overload during automated scans of public profiles

Datalux/Osintgram · Python · 14.4k stars Est. 2019

Osintgram, the Python-based Instagram OSINT tool, released version 1.3 with command-line mode and output limiting to streamline reconnaissance.

Users can now run scans non-interactively via python main.py --username target and cap results to avoid flooding terminals. The tool still requires valid Instagram credentials or a HikerAPI token and only accesses public profiles or followed accounts. Recent commits fix JSON parsing and data loss bugs, keeping the seven-year-old project functional. The catch: It cannot bypass private profile restrictions, limiting its utility against non-public targets.

Use Cases
  • Security analysts gather public Instagram data for threat intelligence
  • Researchers map social networks using nickname-based account lookups
  • Pentesters automate reconnaissance during authorized security assessments

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

Quick Hits

Ciphey Automatically decrypts unknown encryptions, decodes encodings, and cracks hashes without needing keys or cipher details. 21.6k
juice-shop Provides a modern, sophisticated insecure web app for hands-on security training and vulnerability testing. 13.9k
ntopng Offers real-time web-based monitoring of network traffic for cybersecurity threats and performance analysis. 8.2k
trufflehog Detects, verifies, and analyzes leaked credentials across codebases and repositories to prevent data breaches. 27.9k
setup-ipsec-vpn Enables quick setup of an IPsec VPN server with multi-protocol support and client configs across major Linux platforms and Raspberry Pi. 28.5k

Curl 8.22.0 Adds WebSocket Client Support for Real-Time Data Streaming 🔗

New WS/WSS protocol handlers enable bidirectional communication without external libraries

curl/curl · C · ▲ 2 in 1d Est. 2010 · Latest: curl-8_22_0

The latest release of curl introduces native WebSocket client capabilities, allowing developers to establish full-duplex connections directly from the command line or via libcurl. This eliminates the need for separate tools like wscat or custom scripts when testing or consuming real-time APIs.

Built on the existing URL syntax framework, users can now run commands such as curl ws://example.com/socket to send and receive messages, with support for secure WSS endpoints and custom headers. The implementation follows RFC 6455 and integrates seamlessly with curl’s existing error handling, verbose logging, and proxy support.

Beyond WebSockets, the release refines HTTP/3 performance with improved QUIC connection reuse and adds stricter validation for SMTP AUTH mechanisms to prevent credential leakage. These updates reinforce curl’s role as a universal transfer tool, particularly valuable in DevOps pipelines where minimal dependencies and broad protocol coverage are essential.

The catch: While curl now supports WebSocket client operations, it does not act as a WebSocket server, limiting its use in scenarios requiring bidirectional service implementation or long-running socket hosting.

Previously in The Times “covered” — Sep 3

Use Cases
  • Developers testing WebSocket API endpoints
  • Automation scripts consuming real-time financial data feeds
  • IoT devices sending telemetry via secure WebSocket connections

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

More Stories

CodeGraph v1.6.0 adds GitHub Copilot support for faster local agent context 🔗

One-command setup now indexes projects and wires agents without prompts

colbymchenry/codegraph · C · 70.8k stars 7mo old

CodeGraph’s latest release integrates GitHub Copilot across VS Code, CLI, and JetBrains IDEs, joining Claude Code, Cursor, and others. The codegraph install --yes --init command auto-configures agents and builds a local knowledge graph in one step.

Indexing stays current with code changes, avoiding drift and disk leaks from past sessions. Agents get surgical context—only relevant files and symbols—reducing token use and tool calls. Works fully offline, no Node.js needed.
The catch: 508 open issues suggest stability gaps in complex monorepos or unsupported languages despite recent fixes.

Use Cases
  • Devs indexing C/C++ projects for instant agent navigation
  • CI pipelines pre-warming CodeGraph for faster PR reviews
  • Teams reducing Claude Code token costs with local context graphs

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

RuView adds USB onboarding for ESP32 firmware setup 🔗

Streamlines device configuration via console RX fix and mainline USB support

ruvnet/RuView · Rust · ▲ 179 in 1d Est. 2025

RuView’s latest release introduces USB onboarding for ESP32 sensors, simplifying initial setup with a mainline firmware update and a fix for C6 console RX handling. The change, merged in PR #1902, reduces friction for users deploying nodes to capture Channel State Information for contactless vital signs and spatial sensing.

By eliminating complex serial flashing steps, it lowers the barrier for integrating WiFi-based presence detection into Home Assistant, Apple Home, or Matter ecosystems. The update supports real-time breathing, heart rate, and occupancy tracking through walls using commodity hardware.
The catch: Long-term model drift and retraining needs across environments remain an open challenge for consistent accuracy.

Previously in The Times “covered” — Aug 30

Use Cases
  • Homeowners monitor sleep breathing via wall-penetrating WiFi sensing
  • Caregivers detect falls or inactivity in elderly residents without wearables
  • Automate lighting and HVAC based on room occupancy and movement patterns

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

Tauri 2.1.0 Alpha Adds Rust Security Audits for Desktop Apps 🔗

Framework maintains tiny binaries while flagging five unmaintained dependencies in latest scan

tauri-apps/tauri · Rust · ▲ 15 in 1d Est. 2019

Tauri’s latest alpha release integrates Cargo Audit to scan Rust dependencies, surfacing five unmaintained crates like fxhash and rustybuzz in its security report. The framework still enables developers to build cross-platform desktop and mobile apps using any web frontend backed by a Rust binary, leveraging WRY for system webviews and tao for windowing.

Builders get smaller, faster binaries with built-in bundlers for .exe, .dmg, and .AppImage formats, plus self-updating and system tray support. The catch: Flagged dependencies raise maintenance concerns despite no immediate exploits, questioning long-term reliability in security-sensitive deployments.

Use Cases
  • Developers building secure internal tools with React and Rust
  • Startups shipping lightweight cross-platform apps via npm
  • Enterprises replacing Electron apps to reduce memory footprint

Source: tauri-apps/tauri — based on the README and release notes.

Quick Hits

warp Warp transforms the terminal into an intelligent, agent-powered development environment that anticipates your workflow and accelerates coding in Rust. 65k
ollama Ollama lets you instantly run and manage diverse LLMs like Kimi-K2.6 and Qwen locally, with zero setup and full model control. 180.9k
llama.cpp llama.cpp enables high-performance LLM inference directly in C/C++ with minimal dependencies, ideal for edge and embedded deployment. 128.2k
TypeScript TypeScript enhances JavaScript with static typing and modern features, compiling to clean, standards-compliant JS for scalable app development. 111k
protobuf Protocol Buffers provides a language-neutral, efficient mechanism for serializing structured data, enabling fast, reliable communication across systems and services. 72k

GDSFactory v9.51.0 Boosts Photonic Chip Design With Preserved Cross-Sections 🔗

Latest release fixes port extension bugs and adds YAML support for custom uPDK parameters in hardware workflows

gdsfactory/gdsfactory · Python · 1k stars Est. 2020 · Latest: v9.51.0

GDSFactory’s v9.51.

0 update refines core layout functions critical for photonic and quantum chip design. The release preserves multisection cross sections during port extension (#4715), preventing geometry loss in complex routing — a common pain point when bundling waveguides or electrical traces. It also fixes default component behavior for add_pads_bot and add_pads_top (#4840), ensuring reliable pad placement in PCB and MEMS designs.

A notable usability upgrade lets users define custom uPDK parameters directly in YAML (#4723), streamlining integration with proprietary process design kits without rewriting Python scripts. The update maintains compatibility with KLayout, Ansys, Lumerical, and SPICE simulators, letting designers simulate and verify layouts natively. Output still spans GDSII, OASIS, STL, and GERBER files for fabrication, with netlists and settings exportable in YAML.

The project continues to support over 40 open PDKs and offers a VSCode-based GUI via GDSFactory+ for schematic capture and DRC/LVS checks. With 4M+ downloads and 116 contributors, it remains a go-to for open-source hardware design — but its Python-first approach may steepen the learning curve for teams entrenched in GUI-only EDA tools.

The catch: Despite active development, 62 open issues suggest lingering edge cases in DRC verification and layer-stack handling that could delay sign-off for safety-critical aerospace or medical chip projects.

Previously in The Times “covered” — Sep 4

Use Cases
  • Photonics engineers design tunable lasers using parametric Python components
  • PCB designers generate GERBER files from YAML-defined layer stacks
  • Quantum researchers simulate qubit couplers before GDSII fabrication

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

More Stories

Curated electronics resource list adds no new tools in latest update 🔗

Maintains 536 forks and 33 open issues after decade-long curation

kitspace/awesome-electronics · Unknown · 8.1k stars Est. 2016

The kitspace/awesome-electronics repository continues as a static catalog of EDA tools, tutorials, and part suppliers for engineers and hobbyists. Its last commit was a day ago, but no new resources were added—only routine maintenance.

The list covers KiCad, Altium, Gerber viewers, and inventory platforms like Octopart.
The catch: No mechanism exists to vet outdated links or remove abandoned projects, risking stale recommendations in fast-moving hardware fields.

Use Cases
  • Engineers find free EDA packages for PCB design
  • Hobbyists locate soldering tutorials and part search engines
  • Students access free engineering course links from NPTEL and Khan Academy

Source: kitspace/awesome-electronics — based on the project README.

Bruce Firmware Adds WPA3 Detection and BLE Stability for ESP32-S3 Red Team Kits 🔗

Latest release improves WiFi handshake capture and fixes ESP32-C5 BLE crashes in offensive security workflows

BruceDevices/firmware · C++ · ▲ 6 in 1d Est. 2024

BruceDevices/firmware v1.16.

1 enhances offensive-security capabilities on ESP32-S3 hardware with WPA3-PSK detection and stabilized BLE spam functions. The update ensures WiFi handshakes are captured reliably by Brucegotchi and prevents ESP32-C5 crashes during BLE operations. SSH/Telnet stability is improved for non-keyboard devices, and station deauthentication now functions correctly. Built for M5Stack, Flipper Zero, and RF REAPER kits, Bruce bundles tools like ARP spoofing, TCP sniffing, and Evil Portal into a single flashable firmware.
The catch: Despite active development, 242 open issues suggest ongoing stability challenges in complex RF and networking scenarios.

Previously in The Times “covered” — Aug 31

Use Cases
  • Red teamers testing WiFi network resilience via beacon spam and deauth floods
  • Hardware hackers conducting NFC/RFID and Sub-GHz signal analysis on embedded devices
  • Pentesters deploying rogue access points and ARP poisoning in localized wireless engagements

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

SonosESP v2.0.7 adds sleep timer and battery display for ESP32-P4 touch panels 🔗

OTA flicker fixed and lyrics now render accented characters correctly on 4" or 7" screens

OpenSurface/SonosESP · C · 55 stars 8mo old

OpenSurface/SonosESP released v2.0.

7 with a sleep timer that sets speaker-native timers from 15 to 90 minutes, plus battery levels now show for Move and Roam speakers in the UI. The update fixes OTA screen flicker traced to flash chip behavior, not the ESP32-P4, and stops incorrect URL-decoding of favorite URIs when adding to queue. Synced lyrics from LRCLIB now properly display accented characters like in Beyoncé or Björk titles. The project runs on ESP32-P4 with LVGL, supports 4" or 7" MIPI-DSI panels, and enables browser-based flashing.
The catch: Only seven open issues remain, but long-term stability under continuous multi-room use is unverified at scale.

Previously in The Times “covered” — Aug 29

Use Cases
  • Home builders install wall-mounted Sonos control in kitchens
  • Renters use desktop panel for bedroom speaker groups without app
  • Developers extend LVGL themes for custom smart-display interfaces

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

Quick Hits

ghdl Simulates VHDL designs across standards (2008/93/87) for accurate FPGA and ASIC verification without hardware. 2.9k
micrOS Provides a lightweight async automation OS in Python for DIY hardware control and task scheduling. 137
tulipcc Enables portable Python-based sound synthesis on embedded hardware via the Tulip Creative Computer and AMYboard. 968
TuyaOpen Delivers a unified AI+IoT framework for rapid integration of intelligent agents across diverse edge hardware platforms. 1.8k
firmware Offers a community-driven, open-source alternative firmware for IP cameras with enhanced features and privacy control. 2.2k

Babylon.js Adds MultiTexture Class and USD Improvements in Latest Release 🔗

New GPU particle controls and FBX loader fidelity boost 3D web rendering for complex scenes

BabylonJS/Babylon.js · TypeScript · ▲ 1 in 1d Est. 2013 · Latest: 9.26.1

Babylon.js 9.

26.1 introduces a new MultiTexture class enabling layered material effects without custom shaders, streamlining workflows for artists targeting WebGPU. The release decouples Gaussian classification to improve tree shaking, reducing bundle sizes for ES6 imports. GPUParticleSystem now supports per-particle sizing with fixed quad offsets, enhancing visual fidelity in fluid simulations. FBX loader gains full format coverage, preserving animation curves, NURBS, and DCC-sourced materials from Maya, 3DS Max, and Blender. USD handling sees upgrades in instancing, morph targets, and skinning, addressing long-standing gaps in 3D asset interchange. These changes arrive alongside a dithered tile fade material plugin for smoother LOD transitions. The project maintains its TypeScript-first approach, with full typing via npm and playground-driven learning. Despite 13 years of evolution, Babylon.js remains under active maintenance, with commits daily and a growing ecosystem of exporters and tools.
The catch: While Babylon.js excels in browser-based 3D, its reliance on WebGPU/WebGL limits standalone desktop deployment compared to Unity or Unreal, and large-scale physics-heavy simulations still require offloading to Web Workers or backend services to avoid frame drops.

Previously in The Times “covered” — Sep 3

Use Cases
  • Game devs creating WebGPU-powered 3D titles with layered materials
  • Architectural visualizers importing FBX/USD with preserved animations and shaders
  • Training sim builders using GPU particles for fluid dynamics in browsers

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

More Stories

SpacetimeDB Adds MCP Support for Direct AI Agent Database Access 🔗

Maincloud now routes Model Context Protocol requests to leader replicas with egress tracking

clockworklabs/SpacetimeDB · Rust · ▲ 3 in 1d Est. 2023

SpacetimeDB v2.10.

0 enables AI agents and MCP-compatible tools to connect directly to Maincloud databases via the /v1/mcp endpoint. The update includes cluster-aware routing that proxies MCP requests to the leader replica and tracks egress for tool calls. C# NativeAOT modules also gain monomorphized dispatch, eliminating virtual call overhead at the module boundary. The TypeScript SDK now correctly routes mid-session WebSocket errors to onDisconnect, fixing stalled clients. Despite 888 open issues, the project powers BitCraft Online’s backend as a single module handling chat, items, and player positions for thousands.
The catch: Reliance on in-memory state limits dataset size to available RAM, posing a barrier for large-scale persistent worlds without sharding.

Previously in The Times “covered” — Aug 29

Use Cases
  • Game devs sync player state in real time
  • AI agents query databases via MCP protocol
  • Full-stack apps deploy as single Rust binaries

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

Tracy Profiler Adds Fiber-Thread Memory Tracking in Latest Release 🔗

v0.14.1 hardens protocol encoding and fixes 32-bit Windows VSync compilation

wolfpld/tracy · C++ · ▲ 2 in 1d Est. 2020

The Tracy profiler now attributes memory events to virtual fiber threads, improving accuracy in concurrent workloads. Release v0.

14.1 also silences MSVC alignment warnings, fixes on-demand mode section setup, and hardens Lua handling. It drops support for pre-0.11 traces and links with libdl for legacy glibc compatibility. Builders use Tracy to profile CPU, GPU, locks, and context switches across C++, Lua, Python, and Fortran, with third-party bindings for Rust, Zig, and C#. The catch: Open issues remain high at 165, and the project’s C++ core may deter teams seeking pure-Rust or managed-language toolchains.

Use Cases
  • Game developers profiling frame times in Vulkan or DirectX12 engines
  • Performance engineers tracking memory allocations across threads
  • Diagnosing lock contention and context switches in real-time simulations

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

PlayCanvas Engine Adds WebGPU Shadow Fixes in Latest Release 🔗

v2.22.2 resolves material texture iteration and shadow shader selection bugs

playcanvas/engine · JavaScript · ▲ 12 in 1d Est. 2014

The PlayCanvas engine updated to v2.22.

2 with two key fixes: correct shadow shader selection when adjusting cascade counts and proper texture map iteration in StandardMaterial that now respects prototype properties. Built on WebGL2 and WebGPU, the engine supports 3D Gaussian Splatting, glTF 2.0 asset streaming, and TypeScript/JavaScript scripting for browser-based 3D apps. It’s used by companies like Disney, Samsung, and Snap for games, ads, and VR experiences via WebXR.
The catch: Despite recent activity, the project has 496 open issues, suggesting ongoing maintenance challenges in a complex, multi-runtime codebase.

Use Cases
  • Build browser-based 3D games with physics and animation
  • Create AR/VR experiences using WebXR and glTF assets
  • Develop interactive visualizations with Gaussian splatting support

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

Quick Hits

magictools Provides curated game development resources to help creators build magical experiences with ease. 17.3k
tiled Offers a flexible, powerful level editor for designing game environments across platforms. 12.9k
internal Delivers internal GLSL shaders and rendering utilities for advanced visual effects in MustardOS. 112
GDevelop Enables anyone to create, share, and play 2D/3D/multiplayer games without coding barriers. 26.5k
material-maker Lets artists procedurally generate textures and paint 3D models directly within the Godot engine. 5.9k
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