Design
Preset
Background
Text
Font
Size
Width
Account Pricing

The Git Times

AI Models
Claude Opus 5 $25/M GPT-5.6 Luna $1.20/M Gemini 3.1 Pro Preview $12/M Grok 4.6 $6/M DeepSeek V4 Pro 0813 $3.56/M Qwen3.8 2.4T A95B $6/M Kimi K3 $15/M
Full Markets →

macOS Harness Lets LLMs Control Your Mac Through Raw Accessibility Primitives 🔗

Enables agents to automate any task by directly manipulating apps without frameworks or predefined tools

browser-use/macos-harness · Python · ▲ 107 in 1d 5d old · Latest: v0.1.2

The browser-use/macos-harness project gives large language models direct, low-level control over macOS systems through a minimal Python interface. Rather than relying on specialized APIs for individual applications, it exposes raw Apple Accessibility and Apple Events primitives, allowing LLMs to see, interact with, and script any visible element on screen.

The harness connects a single Python process to the real desktop environment — including logged-in browsers, file systems, and background applications — enabling agents to perform complex workflows by combining basic actions like coordinate clicks, keyboard input, and window observation.

What makes this approach technically interesting is its rejection of abstraction layers. Instead of providing pre-built tools for Slack, Spotify, or Final Cut Pro, macos-harness gives the model raw capabilities — see, key, type, click, ax, and script — and trusts the LLM to compose them into task-specific behaviors. This mirrors how humans use computers: by perceiving interface states and adapting actions accordingly. The agent installs the package, self-documents its workflow, validates permissions, and maintains connection without moving the physical cursor or activating target apps, preserving user context.

By keeping ordinary Python and subprocess available within the same process, the harness avoids sandboxing limitations while still isolating risky operations. It leverages Browser Harness for authenticated web interactions and includes a doctor command to validate required macOS permissions. Anonymous telemetry tracks only aggregate usage metrics, preserving privacy.

The catch: As an experimental, macOS-only tool with limited documentation and only six open issues, its reliability for production-grade agent workflows remains unproven at scale, and its narrow scope excludes Linux and Windows users seeking cross-platform agent control.

Use Cases
  • Developers testing UI agents on real macOS applications
  • Researchers building general-purpose computer-use models
  • Automation engineers creating adaptive desktop workflows

Source: browser-use/macos-harness — based on the README and release notes.

More on the Front Page

XERJ Slashes AI Agent Token Use by 40x with One-Command Local Search 🔗

Rust-powered engine indexes code, docs, logs and PDFs for RAG and security without burning context windows

xerj-org/xerj · Rust · 1.6k stars 1mo old

XERJ lets AI agents search local data without wasting tokens on repetitive file reads. Its autoindex feature scans folders once, making code, documentation, logs, and PDFs instantly searchable for RAG, security audits, or agent memory.

Built in Rust, it delivers Elasticsearch-compatible queries so existing tools plug in unchanged. The project claims 40x fewer tokens than grep or sed for equivalent searches, a gain validated in a controlled study showing coding agents cut output tokens by 2.7x while maintaining solve rates. Field reports suggest everyday savings of roughly 5x, enabling smaller, cheaper models to outperform larger ones reliant on memory. Installation is a one-prompt command; feedback is a single markdown PR under user-feedback/16-agent-field-reports/, CLA-exempt and version-tracked. Recent work fixed edge cases in autoindexing, correctly distinguishing excluded files from deletions and handling non-UTF-8 filenames.
The catch: XERJ is pre-1.0 (v1.0.0-rc.31) with 45 open issues, meaning stability and performance at scale remain unproven for enterprise workloads.

Use Cases
  • Coding agents reusing open-source solutions
  • Security audits scanning local logs and configs
  • Developers searching docs without context window drain

Source: xerj-org/xerj — based on the README and release notes.

Kiro Crew Persists Development Work Beyond Single Sessions 🔗

Self-improving AI agents handle multi-step tasks unattended across desktop, web, and CLI

kirodotdev/KiroCrew · Python · 3.2k stars 1mo old

Kiro Crew provides a persistent workspace where development work continues across sessions using self-learning agents. It runs locally or remotely via desktop app, web dashboard, or CLI, integrating with Slack and Discord for seamless workflow continuity.

Multi-step tasks execute unattended, recurring jobs run on schedule, and system heartbeats trigger alerts when attention is needed. Built on Python with Node.js 22+ required, it supports parallel Crew Mode for handling multiple threads simultaneously. Session summaries and instant resume improve context retention, while Kiro Crew Apps tailor interfaces to specific jobs using integrated skills and services.
The catch: With 817 open issues and rapid evolution, stability for production-critical automation remains an open question despite active development.

Use Cases
  • Developers automate multi-step debugging across sessions
  • Teams schedule recurring code quality checks unattended
  • Engineers monitor systems via agent-driven heartbeat alerts

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

LoopX Provides Stateful Control Plane for Long-Horizon AI Agents 🔗

Enables durable, governed workflows across Codex, Claude Code, and other harnesses

huangruiteng/loopx · Python · 5k stars 2mo old

LoopX is a Python-based control plane that adds long-horizon state, governance, and recovery to AI agents running in harnesses like Codex App or Claude Code. It keeps objectives, todos, and evidence stable across bounded turns, enabling reviewable, restartable work over days or tool switches.

The v0.5.1 release adds desktop previews for macOS and Windows, Lark Goal Topics for targeted agent routing, and DeepSeek Harness as a first-class module. It supports human-agent collaboration through typed previews and receipts, with state stored locally first.
The catch: Early adoption remains limited, with 35 open issues and narrow real-world validation beyond controlled environments.

Use Cases
  • Developers managing multi-day coding agents across IDEs
  • Teams coordinating long-running AI workflows with human oversight
  • Operators needing recoverable, auditable agent sessions after restarts

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

Claude Code plugin adds plain-English rewrites of assistant messages on screen only 🔗

Uses local Ollama by default with fail-open safety to preserve original transcripts

Leutenegger/claudish-to-english · Shell · 579 stars 0d old

Leutenegger/claudish-to-english is a shell-based Claude Code plugin that displays plain-language rewrites of each assistant message using a local LLM via Ollama (default), Codex CLI, Anthropic, or any OpenAI-compatible API. The rewrite is display-only: Claude’s reasoning and saved transcript retain the original text, ensuring no data loss.

If the model, and nothing works until these are in place. The plugin fails open—if the provider is down, times out, or lacks a key, users see Claude’s original output unchanged. An optional hook can rewrite Markdown files into plain language on write or edit, opt-in and off by default.
The catch: Requires a pre-loaded local model like gemma4:26b-mlx on macOS, with no Windows support without manual model switching.

Use Cases
  • Developers seeking clearer AI responses without altering logs
  • Teams using local LLMs to avoid API costs and latency
  • Writers editing Markdown who want instant plain-language previews

Source: Leutenegger/claudish-to-english — based on the project README.

Orca v1.4.188 improves remote agent workspace recovery after reconnects 🔗

Terminal panes now reliably restore splits and scrollback following host handoffs or app restarts

stablyai/orca · TypeScript · ▲ 50.8k in 1d 5mo old

The latest Orca release enhances reliability for developers running parallel AI agents on remote VPS instances. Workspaces now recover more consistently after network interruptions, host handoffs, or application restarts, reducing context loss during long-running agent tasks.

Terminal panes maintain their split configurations and scrollback history even when parked or waiting for protocol replies, addressing a prior pain point in distributed workflows. Codex account switching also no longer triggers false restart notices after reauthentication. These fixes target real-world instability in orchestrated agent environments where uptime and state persistence are critical.
The catch: Despite frequent updates, Orca’s rapid release cycle means documentation often lags behind features, forcing builders to inspect changelogs or source code to understand new capabilities.

Previously in The Times “covered” — Aug 16

Use Cases
  • DevOps engineers orchestrating Claude Code agents across cloud workstations
  • Mobile developers reviewing agent-generated PRs via iOS Orca Monitor
  • Full-stack teams comparing OpenCode and Pi outputs in isolated git worktrees

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

ThreeUI Open Catalog Delivers Free 3D UI Components for Web Developers 🔗

Community version strips Pro/Beta features while preserving all free variants and controls

MengTo/threeui · HTML · 859 stars 0d old

MengTo/threeui provides an open-source, login-free catalog of 50 Community parent components with 141 free variant records and 111 routes. Built with HTML, React, and Three.

js, it includes live renderers, variant pickers, and source tabs identical to the Pro version but excludes paid and beta components. The MIT-licensed project syncs from a main-project snapshot, filtering out restricted assets and generating public import graphs. Developers can browse, render, and inspect code without authentication or account state. Remote thumbnails load from threeui.com but aren’t redistributed. The catch: The catalog depends on external thumbnail servers and lacks Pro/Beta components, limiting access to advanced UI patterns available only in the paid tier.

Use Cases
  • Frontend engineers building WebGL-based dashboards
  • UI designers prototyping interactive 3D component libraries
  • Educators teaching Three.js and React integration without licensing barriers

Source: MengTo/threeui — based on the project README.

AI Agents Now Orchestrate Full Workflows Beyond Simple Prompting 🔗

Open source projects show agents controlling systems, persisting memory, and auditing themselves for reliable automation

Trendai-agents
browser-use/macos-harnessYuzzyuk/marketing-osvm0-ai/vm0kirodotdev/KiroCrewxerj-org/xerjdolthub/dolthuangruiteng/loopxvolcengine/OpenVikingbojieli/ai-agent-bookakitaonrails/ai-memorytt-a1i/archifyifixai-ai/iFixAiinternet-court/internet-court-skillchaitanyagiri/munder-difflinblader/humanizerayghri/i-have-adhdcalesthio/OpenMontagePrimeIntellect-ai/prime-agentcan1357/oh-my-piWenyuChiou/awesome-agentic-ai-zhvectorize-io/hindsightmagnitudedev/magnitudechuspeeism/dashi-ppt-skillxai-org/grok-buildagegr/pi-webmaka-agent/maka-agenthenryqin1997/statemyizhiyanhua-ai/fireworks-tech-graphK-Dense-AI/scientific-agent-skillsop7418/guizang-ppt-skillalibaba/open-code-reviewrocketride-org/rocketride-serververcel-labs/agent-browserNVIDIA/cosmosiamaisim/ProjectAirSimcurl/curl

The emerging pattern in open source AI agents shifts from reactive assistants to proactive, self-directed systems that manage complex, multi-step workflows. Projects like browser-use/macos-harness grant LLMs direct, unrestricted control over macOS, enabling agents to navigate GUIs and execute tasks autonomously.

Similarly, Yuzzyuk/marketing-os packages an entire marketing department into a Claude skill, with modules for audits, copy grading, ad diagnosis, and campaign launches—all operable within agent interfaces. Persistence is another key theme: kirodotdev/KiroCrew maintains a self-improving workspace across sessions, while huangruiteng/loopx provides a control plane for long-horizon, governed agent work across Claude Code and Codex. Memory and context are being rearchitected—volcengine/OpenViking unifies agent memory, RAG, and skills in a self-evolving database, and akitaonrails/ai-memory enables long-term retention and handoff between agent vendors. Specialized skills are proliferating: fixai-ai/iFixAi offers rapid agent auditing to verify behavior alignment, vercel-labs/agent-browser delivers Rust-based browser automation, and chaitanyagiri/munder-difflin enables local multi-agent harnessing. Even niche domains see agentification, as calesthio/OpenMontage turns coding assistants into full video production studios with 12 pipelines and 700+ skills.

The catch: Much of this remains experimental, with inconsistent interfaces, unclear reliability guarantees, and fragmented standards—agents often work well in demos but struggle with edge cases, error recovery, or true autonomy in production, making widespread adoption premature despite rapid innovation.

Use Cases
  • Developers automate macOS workflows via LLM-controlled GUI interactions
  • Marketing teams deploy agent skills for real-time copy auditing and campaign launches
  • Enterprises audit agent behavior in under two minutes to validate task alignment

Open Source Agents Evolve Into Modular, Interoperable Skill Ecosystems 🔗

Developers chain specialized AI tools across clients using standardized interfaces for persistent, composable workflows

Trendllm-tools
browser-use/macos-harnessLeutenegger/claudish-to-englishYuzzyuk/marketing-osvm0-ai/vm0kirodotdev/KiroCrewsgl-project/sglanghuangruiteng/loopxcathrynlavery/diagram-designTiger3807861189/J-Space-Cognition-Suite-V3.6virgiliojr94/book-to-skillinternet-court/internet-court-skillzhaoxuya520/reverse-skilllidge-jun/opencodexcoreyhaines31/marketingskillslightningpixel/modlyWei-Shaw/sub2apiWenyuChiou/awesome-agentic-ai-zhAlishahryar1/free-claude-codecitrolabs/ego-liteOsmantic/ODSshy3130/tickflow-stock-panelK-Dense-AI/scientific-agent-skillsjundot/omlxMadsLorentzen/ai-job-searchalibaba/open-code-reviewrocketride-org/rocketride-serverZhuLinsen/daily_stock_analysisrohitg00/ai-engineering-from-scratch

A clear pattern is emerging in open source LLM tooling: the rise of interoperable, skill-based agent ecosystems. Projects like kirodotdev/KiroCrew and huangruiteng/loopx provide persistent workspaces and control planes that orchestrate long-horizon tasks across Claude Code, Codex, and Cursor.

Skill repositories are proliferating as modular units of capability—virgiliojr94/book-to-skill converts technical PDFs into usable agent skills, while K-Dense-AI/scientific-agent-skills offers a validated library of 161 science-focused skills for drug discovery and biology research. Interface layers enable flexibility: lidge-jun/opencodex acts as a universal proxy letting any LLM backend power Codex or Claude Code, and internet-court/internet-court-skill introduces agent-to-agent commerce with ERC-7710 permissions and x402 payments. Even niche domains are covered, from lightningpixel/modly’s local 3D model generation to cathrynlavery/diagram-design’s Mermaid-free SVG diagrams for Claude Code. This shift reflects a move from monolithic agents to composable, reusable skills—where agents gain capabilities not through retraining, but by plugging in standardized, open skill modules that work across clients and models.
The catch: While promising, the ecosystem remains fragmented—skill formats vary, interoperability is often aspirational, and many tools rely on fragile prompt engineering rather than robust APIs, raising concerns about reliability and long-term maintainability in production agent workflows.

Use Cases
  • Developers automate multi-step coding tasks using persistent AI workspaces
  • Scientists deploy validated agent skills for literature review and experiment design
  • Marketing teams generate and audit copy via integrated AI skill chains in Claude Code

Rust Replaces C in Core Infrastructure Tools 🔗

Systems projects migrate to Rust for memory safety without sacrificing performance or POSIX compatibility

Trendrust-systems
uutils/coreutilsfastrepl/anarlogkane50613/takumixerj-org/xerjAprilNEA/OpenLogiblock/buzzakitaonrails/ai-memoryzenbu-labs/terminal-browserherdrdev/herdrfirecrawl/anydocAlexsJones/llmfitxai-org/grok-buildRyanCodrai/turbovect8y2/dbxvercel-labs/agent-browsermacro-inc/macrohydra-db/hydradbrohitg00/ai-engineering-from-scratchAdam-CAD/CADAMmytechnotalent/Reverse-Engineeringrust-lang/rustlingsaa-proxy/aa-proxy-rsQuaint-Studios/Reia

A clear pattern is emerging in open source: critical systems infrastructure once dominated by C is being rewritten in Rust. Projects like uutils/coreutils deliver a drop-in, cross-platform replacement for GNU coreutils — ls, cp, grep — with identical behavior but memory-safe execution.

Similarly, xerj-org/xerj reimagines desktop search and RAG indexing, using Rust to achieve 40x token efficiency over grep/sed pipelines while maintaining Elasticsearch compatibility. Low-level hardware interaction is also shifting: AprilNEA/OpenLogi remaps Logitech mouse buttons via HID++ in Rust, eliminating telemetry and accounts. Even agent infrastructure is Rust-native: akitaonrails/ai-memory provides long-term memory for coding agents, and herdrdev/herdr offers a runtime for AI agents to operate securely. Performance-sensitive tools like RyanCodrai/turbovec (vector indexing) and hydra-db/hydradb (graph database on object storage) choose Rust for predictable latency and zero-cost abstractions. This isn’t just about language preference — it’s about achieving C-like control with Rust’s borrow checker preventing entire classes of vulnerabilities in network daemons, file systems, and device drivers. The shift reflects a maturation: open source systems are no longer trading safety for speed; they’re getting both.

The catch: Many of these projects are still in early adoption — feature gaps vs. GNU coreutils linger, agent runtimes lack standardization, and Rust’s steep learning curve slows contributions. Integration with existing C-based toolchains remains friction-prone, and some benchmarks show Rust binaries still lag in startup time or binary size for CLI tools, suggesting the trend, while real, is not yet a wholesale replacement.

Use Cases
  • Developers replace GNU coreutils with memory-safe Rust binaries
  • AI agents index local data efficiently using token-optimized search
  • Hardware peripherals reconfigured locally without vendor telemetry
  • Vector similarity search deployed in Python/Rust hybrid applications
  • Graph databases scale on object storage with ACID guarantees

Deep Cuts

Floating Whale Widget Tracks DeepSeek Balance Live 🔗

Animates balance changes with drag-to-pin, auto-flip, and QQ-bouncy physics

MeteorNOX/DeepSeek-Balance-Whale-Widget · JavaScript · 412 stars

MeteorNOX/DeepSeek-Balance-Whale-Widget drops a playful yet practical floating whale into the DeepSeek Harness interface. It watches your account balance in real time, animating number rolls with smooth transitions and responding to drag with smart edge-snapping and auto-flip on left adherence.

Built in JavaScript, it loads automatically with DSH, needing no manual activation—just install and let the whale swim into view. Developers get instant, glanceable spend awareness without breaking flow, turning opaque usage into a tactile, responsive cue. The catch: It’s early-stage, tightly coupled to DSH, and lacks broader framework agnosticism, limiting adoption outside its niche.

Use Cases
  • Monitor DeepSeek API spend during heavy LLM testing
  • Get real-time balance alerts while coding AI integrations
  • Track usage costs in pair programming sessions without context switching

Source: MeteorNOX/DeepSeek-Balance-Whale-Widget — based on the project README.

Quick Hits

sglang SGLang delivers high-performance serving for large and multimodal language models with optimized inference and batching. 32.2k
SpiderPanel SpiderPanel provides a lightweight, customizable dashboard for monitoring system metrics and containerized applications in real time. 402
youtube-downloader-zen YouTube Downloader Zen enables fast, ad-free downloading of YouTube videos and audio with a clean, minimalist interface. 461
RSSHub RSSHub transforms any web content — from social media to niche sites — into structured, self-hostable RSS feeds. 45.8k
spinel Spinel is a minimal, embeddable 2D graphics rendering engine written in C for lightweight UI and game development. 1.8k
sylva Sylva offers an interactive, procedural Three.js landscape where moss grows and responds dynamically to pointer movement. 295
anarlog anarlog is an open-source alternative to Granola AI, offering local, privacy-first meeting transcription and summarization. 9.1k
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

OpenClaw Adds npm Plugin Updates for Safer Assistant Extensions 🔗

Fix ensures singleton-array metadata compatibility with newer npm clients during plugin installs

openclaw/openclaw · TypeScript · ▲ 108 in 1d 9mo old · Latest: v2026.7.1-2

OpenClaw’s latest release v2026.7.

1-2 resolves a critical npm plugin update issue that blocked official extensions from installing or updating correctly. The fix accepts singleton-array metadata from newer npm clients, restoring compatibility for users managing AI assistant tools via npm, pnpm, or Bun. This addresses a regression where strict metadata parsing prevented plugin lifecycle scripts from running, potentially leaving assistants without updated skills or security patches.

The Gateway architecture remains central: it connects local models to channels like WhatsApp, Slack, and Signal through a unified control plane, with onboarding still verifying model access and workspace setup post-install. Developers using Node.js 22.22.3+ can install the package directly, though the installer still provisions runtimes when needed. Companion apps for voice, camera, and screen actions continue to extend functionality on supported platforms.

The catch: With over 6,000 open issues and rapid fork growth, the project’s maintainability and long-term stability remain uncertain despite active development.

Previously in The Times “covered” — Aug 21

Use Cases
  • Developers extending assistant skills via npm plugins
  • Teams deploying private AI across messaging platforms
  • Users integrating local models with voice and camera tools

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

More Stories

AI Engineering from Scratch launches free certification prep program 🔗

New Claude-certified track adds 33 lessons, labs, and 295 assessment questions for builders

rohitg00/ai-engineering-from-scratch · Python · 47.6k stars 5mo old

The August release of rohitg00/ai-engineering-from-scratch introduces a free certification prep section aligned with Claude’s July 2026 exam blueprints. Builders can now follow structured paths for Associate, Developer, and Architect credentials using npx skills add rohitg00/ai-engineering-from-scratch followed by /claude-certification.

The program includes 33 in-depth lessons, runnable labs, diagnostics, capstones, and 295 original assessment questions, all teachable by AI agents like Claude Code or Cursor that read SKILL.md directories. Lessons remain accessible via /learn and /course-guide commands, streaming content directly from the repo without cloning.
The catch: The certification material is community-built and not endorsed by Anthropic, so it lacks official validation for exam readiness.

Use Cases
  • Prepare for Claude certification exams using hands-on AI-agent tutoring
  • Learn AI engineering by building prompts, agents, and MCP servers from scratch
  • Study multilingual AI lessons with fallback to English for untranslated content

Source: rohitg00/ai-engineering-from-scratch — based on the README and release notes.

Hugging Face Diffusers Adds Joint Video-Audio Generation in v0.40.0 🔗

MiniMax-H3 pipeline creates synchronized video and soundtrack without separate vocoder

huggingface/diffusers · Python · ▲ 3 in 1d Est. 2022

The latest release of 🤗 Diffusers introduces MiniMax-H3, a pipeline that generates video and its audio track jointly using a single transformer denoising process. Built on Qwen3VLForConditionalGeneration, it conditions on text and media while outputting packed video-audio latents, eliminating the need for post-hoc audio synthesis.

This update also graduates Modular Diffusers from experimental to stable, enabling reusable component-based workflows for custom diffusion systems. The project maintains broad support for text-to-image, text-to-video, and latent diffusion models across PyTorch.
The catch: High-fidelity joint video-audio generation demands significant VRAM, limiting accessibility for consumer-grade hardware despite modular efficiency gains.

Use Cases
  • Generate synchronized video clips with ambient soundtracks from text prompts
  • Build custom diffusion systems using interchangeable noise schedulers and pretrained blocks
  • Train latent diffusion models for molecule structure generation in biochemical research

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

Tesseract OCR Engine Fixes Multi-Page XML Output in Latest Patch 🔗

Maintainers resolve closing tag bug affecting PDF and hOCR export accuracy

tesseract-ocr/tesseract · C++ · 76.1k stars Est. 2014

The Tesseract OCR engine, a C++-based open source tool for text recognition, patched a bug in multi-page PAGE XML output that was causing malformed tags in exported documents. The fix, merged in pull request #4506, ensures proper closing tags when processing batches of images — critical for workflows generating searchable PDFs or structured data.

Despite its ~12-year history and 5.5.3 release, the project still relies on external tessdata files for language models and lacks a native GUI. The catch: Users must manually manage traineddata files and image preprocessing for optimal accuracy, adding complexity to automated pipelines.

Use Cases
  • Developers extracting text from scanned invoices
  • Researchers processing historical document archives
  • Teams building automated form digitization tools

Source: tesseract-ocr/tesseract — based on the README and release notes.

Quick Hits

airflow Automates workflow creation, scheduling, and monitoring with code-defined pipelines for scalable data and ML orchestration. 46.6k
cosmos Provides world models, datasets, and tools to build Physical AI for robots, autonomous vehicles, and smart infrastructure. 11.6k
streamlit Enables rapid development and sharing of interactive data apps with minimal code using pure Python. 45.6k
ultralytics Delivers unified computer vision capabilities — detection, segmentation, classification, pose, and tracking — via YOLO models. 60.9k
open-webui Offers an intuitive interface to interact with local and cloud LLMs via Ollama, OpenAI API, and more. 149.5k

Robotic Vision Library RMVL 2.4.0 Delivers Cross-Platform Binaries for Real-Time Control 🔗

Release adds ARM64 Windows and Linux deb packages, expanding deployment beyond x86 build-from-source

cv-rmvl/rmvl · C++ · 110 stars Est. 2023 · Latest: 2.4.0

The cv-rmvl/rmvl project released version 2.4.

0 with precompiled binaries for ARM64 Linux and Windows, alongside existing amd64 Linux tarballs and debs. This update targets robotic manipulation workflows requiring tight integration between OpenCV-based vision processing and OPC UA communication, core to its origin in RoboMaster competitions. Developers can now deploy RMVL directly via package managers on Ubuntu 22.04 and 24.04, or run the ARM64 Windows executable without compiling from C++ source. The library maintains its unified architecture for visual servoing, LPSs (Local Positioning System) integration, and IDL-defined interfaces, reducing boilerplate in vision-control loops.
Despite active maintenance — last commit zero days ago — the project shows limited issue traction with only two open issues and 19 forks, suggesting adoption remains niche within industrial robotics. Its C++-only scope and dependence on specific ROS-adjacent toolchains may deter teams using Python-first ML pipelines or alternative middleware like DDS.
The catch: RMVL’s tight coupling to RoboMaster-era hardware assumptions and limited documentation outside Chinese-language communities create barriers for global teams seeking plug-and-play vision solutions beyond competition-specific setups.

Previously in The Times “covered” — Aug 16

Use Cases
  • RoboMaster teams deploying vision pipelines on Jetson Orin
  • Factory automation engineers integrating OpenCV with OPC UA
  • Research labs prototyping visual servoing on ARM SBCs

Source: cv-rmvl/rmvl — based on the README and release notes.

More Stories

CADAM turns plain language into parametric CAD models in-browser 🔗

Latest release adds agent updates and logo refinements while maintaining core text-to-CAD workflow

Adam-CAD/CADAM · TypeScript · 5k stars 11mo old

CADAM generates 3D models from natural language prompts directly in the browser using WebAssembly and OpenSCAD. Users describe designs like "V8 engine" or "turbofan jet" and receive fully parametric models with interactive sliders for real-time adjustments.

Output exports to STL, SCAD, or DXF formats, supported by BOSL, BOSL2, and MCAD libraries. Recent commits include agent system improvements and visual updates to the CADAM logo.
The catch: Complex assemblies with many moving parts may require manual refinement due to limited agent reasoning depth in geometric constraints.

Use Cases
  • Mechanical engineers prototyping engine components from text descriptions
  • Robotics designers generating parametric brackets and housings rapidly
  • Educators teaching CAD concepts through instant language-to-model conversion

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

Text-to-CAD adds DfAM printability skill for agent-driven CAD workflows 🔗

Release 0.4.21 introduces automated design-for-additive-manufacturing checks in Python

earthtojake/text-to-cad · Python · ▲ 23 in 1d 4mo old

The earthtojake/text-to-cad library now includes a dfam-check skill that evaluates per-process printability for CAD models, enabling agents to assess manufacturability early in design. Built in Python, the tool supports STEP, STL, and URDF formats, integrating with robot-description pipelines and simulation tools like MoveIt2.

Users install skills via the Skills CLI, with npx skills add ensuring new capabilities from releases are not missed — unlike update, which only refreshes locked skills. The project sees active maintenance, with a commit within the last day and over 1,400 forks, signaling strong adoption in agent-based engineering workflows.
The catch: The DfAM skill is currently limited to additive processes and does not evaluate subtractive or hybrid manufacturing constraints, leaving gaps for multi-method fabrication planning.

Previously in The Times “covered” — Aug 17

Use Cases
  • Mechanical engineers validating printability of robot arm brackets
  • CAD agents auto-assessing STEP files for support material needs
  • Robotics teams checking URDF compatibility with fused deposition modeling printers

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

Robotics Wiki Gains 174 Forks as Builders Share Tribal Knowledge 🔗

Community contributions now shape practical guides for drones, fabrication, and UAV systems

RoboticsKnowledgebase/roboticsknowledgebase.github.io · JavaScript · 187 stars Est. 2017

The Robotics Knowledgebase wiki, hosted on GitHub Pages, serves as an open-source reference for robot builders seeking hands-on implementation details beyond textbook theory. Built with Jekyll and the Minimal Mistakes theme, it organizes content by subsystem—sensing, actuation, programming—using Markdown templates and absolute paths for images and links.

Contributors fork the repo, write guides in /wiki, update navigation.yml, and submit pull requests for editor review. Local previews require Ruby and Bundler, with the site rendering at http://localhost:4000 via MathJax for LaTeX. Despite recent activity—last push 0 days ago and 15 open issues—the project’s reliance on Ruby tooling may deter builders preferring JavaScript- or Python-native documentation workflows.
The catch: Dependency on Ruby and Jekyll creates a barrier for contributors unfamiliar with the stack, limiting accessibility for teams using modern JavaScript documentation tools.

Use Cases
  • Robotics researchers sharing drone calibration procedures
  • Hobbyists documenting custom actuator designs for fabrication
  • Engineers deploying UAV swarms using shared programming guides

Source: RoboticsKnowledgebase/roboticsknowledgebase.github.io — based on the project README.

Quick Hits

ProjectAirSim Project AirSim enables builders to train and test autonomous systems in high-fidelity virtual environments with realistic physics and sensor simulation. 780
carla CARLA provides an open-source, modular simulator for developing and validating autonomous driving algorithms under diverse traffic and weather conditions. 14.3k
openpilot openpilot upgrades vehicle driver assistance with open-source robotics software, enabling adaptive cruise control and lane-keeping on supported cars. 63.5k
Robots Visose Robots lets users create, simulate, and validate industrial robot programs for ABB, KUKA, UR, and Staubli arms in a unified C# environment. 379
autoware Autoware delivers a full-stack, open-source autonomous driving framework via Docker, integrating perception, planning, and control for real-world deployment. 12k

OpenAI's Codex Security CLI automates vulnerability fixes with AI-driven patching 🔗

Latest release adds read-only verification and improved SARIF exports for GitHub integration

openai/codex-security · TypeScript · ▲ 38 in 1d 1mo old · Latest: npm-v0.1.16

OpenAI's codex-security CLI and TypeScript SDK now streamline the full vulnerability lifecycle—from detection to patch submission—using AI to validate and fix security flaws in code. The tool scans for vulnerabilities, presents findings in an interactive browser, and allows developers to apply patches directly via --patch --patch-severity high, optionally opening a draft pull request with --create-pr.

Built on Codex’s AI models, it requires Node.js 22.13+, Python 3.10, and access to Codex Security via API key. Environment variables like OPENAI_API_KEY enable CI/CD integration without storing credentials locally. The latest release (v0.1.16) introduces a read-only verify-fix mode to validate patches without applying them and improves SARIF export compatibility for GitHub Advanced Security.

Teams can customize validation prompts with --validation-prompt-file and switch inference providers—such as Amazon Bedrock—by configuring alternative API keys and models. Deep scans are capped at 96 hours unless overridden with --max-time-hours, preserving partial results.

The catch: The tool remains gated behind OpenAI’s Trusted Access for Cyber program, limiting adoption to approved users and creating a dependency on external approval for full functionality.

Previously in The Times “covered” — Aug 16

Use Cases
  • DevSecOps engineers automate high-severity fix validation in CI pipelines
  • Security teams generate GitHub-compatible SARIF reports from AI scans
  • Developers remediate critical vulnerabilities via AI-suggested patches with PR automation

Source: openai/codex-security — based on the README and release notes.

More Stories

OWASP Juice Shop upgrades to Angular 22 and TypeScript 6 🔗

Release v20.2.0 adds four new security challenges and improves cheat detection

juice-shop/juice-shop · TypeScript · 13.7k stars Est. 2014

The OWASP Juice Shop project updated its frontend to Angular 22.x and Angular Material 22.

x, upgraded TypeScript to 6.0.x, and switched QR code libraries in release v20.2.0. Four new challenges were added, including "Misplaced IaC Files" and "Login Cloud Admin," alongside enhanced cheat detection using Dice Coefficient for source file overlap. The app remains a deliberately insecure web application containing OWASP Top Ten vulnerabilities for security training and tool testing. The catch: Its reliance on AI/LLM providers for certain challenges creates a barrier for offline or air-gapped training environments.

Use Cases
  • Security teams run CTFs to practice exploit techniques
  • Developers learn secure coding by breaking the app
  • Vendors test DAST and SAST tools against known flaws

Source: juice-shop/juice-shop — based on the README and release notes.

Reverse-Engineering Tutorial Expands Firmware Analysis for Embedded Systems 🔗

Last updated August 2026 with STM32F401CCU6 setup guide and GDB debugging basics

mytechnotalent/Reverse-Engineering · Assembly · 14.1k stars Est. 2020

The mytechnotalent/Reverse-Engineering project offers a free, multi-architecture tutorial covering x86, x64, ARM, AVR, and RISC-V assembly. Recent updates include a hands-on section on setting up a development environment for the STM32F401CCU6 MCU and using GDB to inspect memory and instructions.

The guide walks through static and dynamic malware analysis techniques, binary and hexadecimal fundamentals, and processor architecture details. It remains a go-to resource for builders learning low-level system inspection without cost.
The catch: The tutorial assumes prior programming knowledge and does not provide structured exercises or assessment tools for self-paced learning validation.

Use Cases
  • Firmware developers analyzing bootloader behavior on ARM Cortex-M4
  • Security researchers performing static analysis on IoT device binaries
  • Students learning RISC-V instruction sets through practical disassembly examples

Source: mytechnotalent/Reverse-Engineering — based on the project README.

Caddy v2.11.4 patches security flaws amid AI-generated spam surge 🔗

Maintainers reject over 75% of bogus reports while fixing path, header, and template bugs

caddyserver/caddy · Go · ▲ 32 in 1d Est. 2015

Caddy’s latest release addresses security-adjacent bugs including Windows path normalization in HTTP matchers, preventing placeholder re-expansion in injected queries, and improving HTML stripping in templates. Header fields with underscores are now ignored to avoid collisions.

The FrankenPHP team contributed PHP-related patches. Maintainers note a surge of low-quality AI-generated security reports, leading to blocks on spam accounts. Despite active development, the project carries 271 open issues.
The catch: Ongoing maintenance burden from AI slop diverts resources from genuine bugs and feature work.

Previously in The Times “covered” — Aug 14

Use Cases
  • DevOps teams deploying zero-config HTTPS sites
  • Enterprises scaling internal services with managed CAs
  • Developers testing HTTP/3 locally without external deps

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

Quick Hits

openzeppelin-contracts Provides secure, audited Solidity libraries for building reliable smart contracts with standardized, battle-tested components. 27.2k
sherlock Enables rapid OSINT investigations by scanning usernames across hundreds of social media platforms to uncover digital footprints. 90k
Learn-Web-Hacking Offers structured, bilingual study notes covering web penetration testing techniques, vulnerabilities, and exploit methodologies for learners. 5.5k
suricata Delivers high-performance, multi-threaded network threat detection and prevention with real-time IDS/IPS and NSM capabilities. 6.6k
MISP Facilitates collaborative threat intelligence sharing through an open-source platform for collecting, storing, and correlating IOCs and malware data. 6.5k

TensorFlow 2.21 drops Python 3.9 support and decouples TensorBoard 🔗

Release adds int2/int4 quantization and JPEG XL decoding for edge ML workloads

tensorflow/tensorflow · C++ · ▲ 1 in 1d Est. 2015 · Latest: v2.21.0

TensorFlow 2.21.

0 removes support for Python 3.9 and externalizes TensorBoard as a standalone dependency, streamlining the core framework for production deployments. The update strengthens tf.lite with int2 and int4 type support for SQRT, EQUAL, NOT_EQUAL, cast, fully_connected, and slice operators, enabling tighter model quantization for resource-constrained devices. Meanwhile, tf.image gains native JPEG XL decoding via decode_image, improving image pipeline efficiency for modern web and media applications. The release also exposes NoneTensorSpec in tf.data to better handle optional elements in dataset pipelines, aiding debugging and type safety in complex data flows. These changes reflect TensorFlow’s ongoing shift toward modularity, edge readiness, and precision optimization without sacrificing its comprehensive ML ecosystem.
The catch: Breaking changes like dropped Python 3.9 support may force upgrades in legacy environments, and the decoupled TensorBoard requires separate installation, adding setup complexity for teams relying on integrated visualization workflows.

Previously in The Times “covered” — Aug 17

Use Cases
  • Train vision models with JPEG XL input on cloud TPUs
  • Deploy int4-quantized LLMs to microcontrollers via TensorFlow Lite
  • Build data pipelines with conditional None handling in tf.data

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

More Stories

React Native v0.87.0 adds Hermes dSYMs for iOS crash debugging 🔗

Release includes debug and release symbol files for Hermes and core dependencies to improve native stack traces

react/react-native · C++ · ▲ 10 in 1d Est. 2015

React Native v0.87.

0 ships Hermes V1 dSYMs and React Native Core dSYMs for iOS, enabling symbolic crash logs when using Hermes engine. These symbol files, hosted on Maven Central, help developers debug native crashes in production apps without rebuilding. The release notes link to an Upgrade Helper and changelog for migration guidance. Despite monthly commits, the project maintains over 1,000 open issues, reflecting ongoing complexity in cross-platform native bridging.
The catch: Debugging native modules still requires manual symbol upload and doesn't fully eliminate guesswork in async bridge errors.

Use Cases
  • Mobile teams building cross-platform apps with shared React logic
  • Startups shipping iOS/Android apps using Expo or bare React Native
  • Enterprises adopting incremental integration into existing native apps

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

Vercel Labs' Native SDK blends TypeScript ease with Zig performance 🔗

Latest release adds JSON manifests and image cropping for texture-atlas rendering

vercel-labs/native · Zig · ▲ 9 in 1d 3mo old

The vercel-labs/native project updates its toolkit for building native desktop apps without browser runtimes. Version 0.

9.5 introduces JSON manifests by default for new projects, simplifying configuration and enabling schema validation. It also adds registered-image source cropping, allowing developers to select specific regions from texture atlases for efficient rendering. Built in Zig, the SDK compiles TypeScript or Zig logic to native code while rendering via its own engine into real OS windows. Developers can iterate quickly with native dev for live reloads or native check for instant validation. The approach avoids WebView overhead but requires learning a new markup format (.native files) and managing a dual-language workflow.
The catch: The toolkit remains early-stage, with 142 open questions and limited real-world adoption beyond examples.

Previously in The Times “covered” — Aug 15

Use Cases
  • Desktop app builders seeking native performance
  • Teams wanting TypeScript for logic with Zig compilation
  • Developers avoiding Electron or WebView runtime bloat

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

RuView Cognitum Spaces adds tenant-scoped memory and governed actions to WiFi sensing 🔗

Release v2301 enables OAuth-secured, multi-user spatial intelligence with policy-controlled actuation in smart homes

ruvnet/RuView · Rust · ▲ 103 in 1d Est. 2025

RuView’s latest release activates Cognitum Spaces, introducing tenant-scoped spatial memory and governed-action boundaries to its WiFi-based sensing platform. Using ESP32 sensors, it extracts Channel State Information to detect presence, vitals, and room activity without cameras.

The update adds OAuth-only CLI/MCP integration via @ruvnet/ruview@0.5.1 on npm, requiring explicit credential-use grants and blocking arbitrary token inputs. Spatial data is now isolated per tenant with provenance tracking, and actions like alerts or automation require separate approval—preventing autonomous actuation from sensing alone. It integrates with Home Assistant, Apple Home, Google Home, and Alexa as a Matter bridge, exposing 21 entities per node including sleep state and fall risk.
The catch: Governed-action policies add complexity for simple automations, and real-world validation of vitals accuracy through walls remains limited to controlled environments.

Previously in The Times “covered” — Aug 17

Use Cases
  • Monitor elderly residents’ breathing and movement through walls
  • Detect room occupancy and trigger lighting via Home Assistant
  • Enable voice-activated presence queries via Siri or Alexa without wearables

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

Quick Hits

ladybird Builds a truly independent, standards-compliant web browser from scratch in C++, free from corporate engine dependencies. 65.8k
terminal Combines modern Windows Terminal features with the original console host in a unified, customizable C++ application. 104.7k
rustlings Offers bite-sized, hands-on Rust exercises to build fluency in reading and writing idiomatic Rust code. 63.9k
zed Delivers a lightning-fast, collaborative code editor in Rust with real-time multiplayer editing and Atom-inspired UX. 89k
FFmpeg Provides a complete, cross-platform solution for recording, converting, and streaming audio and video via FFmpeg’s powerful codec suite. 63.5k

Stack-chan v1.0.0 unifies browser tools, firmware, and MODs for M5Stack robots 🔗

Release adds WebAssembly simulator, Voice synthesis, and split host-MOD execution model

stack-chan/stack-chan · TypeScript · ▲ 3 in 1d Est. 2021 · Latest: v1.0.0

The stack-chan project released v1.0.

0, its first stable version after five years of development. This release integrates firmware, MOD development tools, browser-based utilities, and a WebAssembly simulator into a cohesive workflow for M5Stack-based robots. Users can now flash firmware, edit faces, program with Blockly, and build MODs entirely through a Chrome or Edge browser without installing the Moddable SDK or ESP-IDF.

Key technical changes include a separated host and MOD execution model, introducing context APIs, platform-specific drivers, and manifest inspection. The firmware now supports M5StackChan CoreS3 as standard, with build targets for Stack-chan RT and Takao Core2 + SG90 available from source. Voice synthesis was expanded to include Stack-chan Voice on CoreS3 alongside VOICEVOX and ElevenLabs, and the UI now supports Japanese, English, and Simplified Chinese.

The release notes warn that upgrading from v0.2.1 or v1.0.0rc requires rewriting host firmware and rebuilding custom MODs against the v1.0.0 API and Moddable SDK 8.3.1 due to breaking changes in module names and context API.

The catch: The project’s reliance on Moddable SDK creates a dependency chain that may complicate long-term maintenance if upstream SDK updates lag or diverge from Stack-chan’s needs.

Previously in The Times “covered” — Aug 16

Use Cases
  • Educators teach robotics via browser-based Blockly programming
  • Hobbyists customize robot faces and motions using Web UI tools
  • Developers build and test MODs in WebAssembly simulator before device deploy

Source: stack-chan/stack-chan — based on the README and release notes.

More Stories

Upsy Desky v5.0.2 fixes ESPHome version inheritance for standing desk automation 🔗

Release ensures correct project version reporting when integrating with ESPHome-based home automation systems

tjhorner/upsy-desky · Shell · 817 stars Est. 2022

The tjhorner/upsy-desky project released v5.0.

2 to resolve an issue where inheriting the package in ESPHome displayed a placeholder "dev" version instead of the actual release. This fix, addressing GitHub issue #67, hard-codes the project_version so automation tools and dashboards reflect accurate firmware revisions. Users can now flash updates via OTA using firmware.bin or perform clean installs with firmware-factory.bin. The device connects motorized standing desks to home automation platforms like Home Assistant, enabling height-based automations such as raising the desk after prolonged sitting or when leaving home. Despite active maintenance — last commit zero days ago — the project carries 36 open issues, indicating ongoing challenges in firmware stability or hardware compatibility that builders should evaluate before deployment.
The catch: Open issues suggest unresolved bugs may affect reliability in complex automation setups, requiring community support or manual troubleshooting.

Use Cases
  • Home automation enthusiasts sync desk height with lighting and climate
  • Remote workers trigger standing presets via calendar or occupancy sensors
  • Households share personalized desk height presets through voice assistants

Source: tjhorner/upsy-desky — based on the README and release notes.

Tulip CC and AMYboard offer portable Python synthesis on ESP32-S3 hardware 🔗

Open-source music computers now run on web, desktop, and $59 Makerfabs kits

shorepine/tulipcc · C · 958 stars Est. 2022

The Tulip Creative Computer and AMYboard are self-contained ESP32-S3 devices running MicroPython for real-time music synthesis, graphics, and sensor interaction. Tulip CC features a touchscreen and sells for $59; AMYboard adds audio I/O, CV jacks, and S/PDIF for $29.

  1. Both boot instantly into a Python prompt with LVGL graphics and AMY audio engine support. Users can code music, games, or art locally or via web and desktop ports. The project sees active development with a commit one day ago and 55 open issues.
    The catch: Real-time audio performance depends on ESP32-S3 resources, limiting complex polyphony or heavy DSP without external acceleration.
Use Cases
  • Musicians build portable synthesizers with touchscreen control
  • Educators teach embedded audio programming via MicroPython
  • Developers prototype sensor-driven music installations quickly

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

OpenSurface/SonosESP adds blurred background fix for missing album art 🔗

v1.14.4 clears visual artifacts on tracks without artwork via OTA-updateable ESP32-P4 touchscreen

OpenSurface/SonosESP · C · 49 stars 7mo old

The OpenSurface/SonosESP project released v1.14.

4, fixing a bug where blurred backgrounds persisted on tracks lacking album art. Built for ESP32-P4, the touchscreen controller displays synced lyrics, weather, and multi-room Sonos control on 4″ or 7″ panels. Users flash firmware via browser installer—no toolchain needed—and receive OTA updates for Stable or Nightly builds. The system extracts dominant colors from artwork and renders accented characters correctly in Latin-1 and Latin Extended-A.
The catch: Only two open issues exist, but long-term maintenance depends on a single contributor with infrequent commits outside release cycles.

Use Cases
  • Homeowners control Sonos via wall-mounted touchscreen
  • Renters install browser-flashed Sonos remote without drilling
  • Developers extend LVGL-based UIs for ESP32-P4 audio projects

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

Quick Hits

aa-proxy-rs Enables Android Auto connectivity via wired or wireless proxy for seamless in-car infotainment integration. 454
awesome-home-assistant Curates essential Home Assistant resources to streamline smart home automation setup and expansion. 8.4k
micrOS Provides a lightweight asynchronous OS for DIY automation projects with minimal resource overhead. 137
childrens-clock Offers an engaging, child-friendly IoT clock that teaches timekeeping through interactive design. 55
LuaCAD Empowers parametric CAD modeling using Lua scripting for flexible, code-driven mechanical design. 115

Godot AI Plugin Enables Real-Time Scene Building via MCP Protocol 🔗

Latest release disables verbose HTTP logs to reduce noise during AI-assisted development sessions

hi-godot/godot-ai · GDScript · ▲ 15 in 1d 4mo old · Latest: v3.1.5

The hi-godot/godot-ai project lets AI assistants like Claude Code and Codex directly manipulate Godot 4.5+ scenes through the Model Context Protocol.

Over 120 operations across 43 MCP tools enable programmatic creation of nodes, scripts, signals, UI, materials, animations, and environments — all without manual coding or asset generation. Installation is streamlined via the Godot Asset Library or manual addon placement, with a Python server requiring uv for dependency management. The plugin auto-starts the MCP connection upon activation in Project Settings, allowing immediate AI-driven workflows. Recent update v3.1.5 fixes logging noise by disabling uvicorn HTTP access logs by default, addressing developer feedback about console clutter during extended AI sessions. This follows months of steady adoption, evidenced by 113 forks and active Discord engagement, though the project remains dependent on external Python tooling despite its GDScript core.
The catch: The AI server requires a separate Python uv environment, adding setup complexity for developers seeking a pure GDScript, zero-dependency workflow.

Previously in The Times “covered” — Aug 21

Use Cases
  • Game designers rapidly prototype levels using natural language prompts
  • Technical artists iterate shader and particle effects via AI-assisted node editing
  • Solo developers automate repetitive scene setup tasks with conversational commands

Source: hi-godot/godot-ai — based on the README and release notes.

More Stories

Assimp 6.0.5 refines memory use and format support for 3D asset pipelines 🔗

Latest release fixes Haiku builds, GLTF skinning export, and FBX output precision

assimp/assimp · C++ · ▲ 2 in 1d Est. 2010

The Open-Asset-Importer-Library (assimp) v6.0.

5 reduces memory consumption in vertex joining and adds proper transformation for normalized normals. It extends skinning data export to GLB/GLTF and fixes FBX floating-point output. Builds now work on Haiku, with spelling and implicit conversion errors resolved. Used in game engines and asset pipelines, assimp imports 40+ 3D formats into a unified C++ structure with C and Python bindings. The catch: Active development continues, but 538 open issues suggest ongoing stability challenges in niche formats and platform-specific builds.

Use Cases
  • Game developers importing FBX/GLTF models into Unity or Unreal
  • CAD engineers converting IFC or STL assets for simulation
  • Mobile apps loading 3D assets on Android/iOS via C++ bindings

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

Tracy Profiler Adds Daemon and Merge Tools for Automated Multi-Process Capture 🔗

v0.14.0 introduces `tracy-capture-daemon` and `tracy-merge` to streamline distributed tracing workflows

wolfpld/tracy · C++ · ▲ 5 in 1d Est. 2020

The latest release of Tracy, a C++ frame profiler used in game development and performance analysis, adds two CLI tools to automate trace collection. tracy-capture-daemon listens for Tracy clients over UDP and saves each process’s trace to a separate file, while tracy-merge combines multi-process traces into a single view for easier analysis.

Though the merge tool currently excludes GPU zones, memory events, and call stacks, it preserves zones, messages, and plots. These updates reduce manual setup in complex, multi-threaded, or distributed applications. Documentation and binaries are available in the release assets.
The catch: The merge tool does not reconstruct GPU-related data or call stacks, limiting its usefulness for low-level graphics or deep stack analysis.

Previously in The Times “covered” — Aug 15

Use Cases
  • Game developers profiling frame timing across engine systems
  • Performance engineers analyzing CPU-GPU synchronization points
  • Developers tracing memory allocation patterns in real-time applications

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

melonJS 20.0 Unveils WebGPU Backend With WebGL 2 Baseline 🔗

Retiring WebGL 1, the release adds GPU-accelerated tilemaps and mesh instancing

melonjs/melonJS · JavaScript · ▲ 1 in 1d Est. 2011

melonJS 20.0 launches a complete WebGPU renderer covering 2D and 3D features, verified against WebGL 2.

The engine now prioritizes WebGPU, falling back to WebGL 2 or Canvas2D via video.AUTO. New capabilities include GPU-accelerated tilemap rendering, mesh instancing for 100,000-object scenes at 60 fps, and full glTF/GLB 3D support. Built with ES6 and esbuild, the tree-shakeable engine remains under 150 KB minzipped. Breaking changes require constructing an Application before calling await app.init().
The catch: WebGPU support is still emerging in browsers, and the Canvas2D fallback may lack advanced GPU features like instancing or normal maps.

Use Cases
  • Indie devs shipping 2D/3D HTML5 games
  • Teams using Tiled for level design
  • Developers targeting WebGPU with Canvas2D fallback

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

Quick Hits

SpacetimeDB Enables ultra-low-latency, real-time application development with a Rust-based distributed database that scales like in-memory speed. 25k
vsgExamples Provides ready-to-run GLSL examples demonstrating VSG usage and integration with optional add-on libraries for graphics experimentation. 170
bgfx Offers a cross-platform, API-agnostic rendering engine in C++ that lets developers plug in their own engine or framework without graphics API lock-in. 17.4k
ComplementaryReimagined Delivers a high-fidelity Minecraft resource pack that enhances visuals with detailed textures, lighting, and performance-optimized graphics. 345
godot-mod-loader Simplifies modding Godot games by providing a flexible GDScript-based loader compatible with both 3.x and 4.x versions. 667
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