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 $3.37/M ▼6%Qwen3.8 2.4T A95B $6/M Kimi K3 $15/M
Full Markets →

Marketing OS Consolidates 14 Agent Skills Into One Claude-Compatible Toolkit 🔗

Combines audits, copy grading, ad diagnosis, and GEO with artifact-first outputs and honest uncertainty labeling

Yuzzyuk/marketing-os · Unknown · ▲ 16 in 1d 5d old · Latest: v1.1.0

Yuzzyuk/marketing-os delivers a single Claude skill packing 14 marketing modules — from SEO audits scored 0-100 to an 18-tactic hook engine and copy graded before you see it. Built by reverse-engineering top GitHub marketing skills, it keeps what works and adds three essentials: scores on everything, tangible artifacts instead of advice, and explicit "what I couldn't determine" in every report.

The skill works in Claude Code, Claude.ai, Cursor, and Codex, loading only the needed module on demand to save tokens. When connected to an ad-generation MCP like Arcads, it moves from diagnose → brief → produce in one session. Outputs are pure markdown — no executable code to review — with an honesty spine that labels all scores as heuristics and avoids invented proof. Installation is a one-command plugin add or manual skills directory copy, requiring only a filled brand-context.md for tailored results.
New in v1.1.0: the hooks module now includes a diagnostic funnel and fidelity ladder, GEO evidence pass measures real lifts and llms.txt compatibility, and platform-mechanics diagnostics catch learning-phase breakdowns. Competitor hook clustering and coverage-gated audit grading add strategic depth, while the ad-production handoff streamlines asset generation. MIT-licensed and zero-forks-behind, it’s gaining traction for replacing fragmented marketing agent skills with a unified, transparent system.
The catch: As pure markdown with no execution, it cannot run tests, optimize live campaigns, or learn from real-time performance data — limiting its role to planning and diagnostics.

Use Cases
  • Marketing teams auditing campaign performance with scored heuristics
  • Copywriters generating and grading variants before client review
  • Developers integrating marketing diagnostics into IDE workflows via Claude Code

Source: Yuzzyuk/marketing-os — based on the README and release notes.

More on the Front Page

Natural-language diagram skill adds reliable SVG exports for agents 🔗

Fix ensures cloud icons render correctly across machines in exported files

Agents365-ai/drawio-skill · Python · 7.9k stars 5mo old

Agents365-ai/drawio-skill converts plain-text prompts into editable .drawio diagrams, supporting 11 presets including UML, BPMN, and C4, plus codebase and infrastructure-to-diagram generation for Python, JS-TS, Go, Rust, Terraform, Kubernetes, and SQL schemas.

It works with Claude Code, Cursor, Copilot, and other Agent Skills-compatible tools, leveraging the native draw.io CLI to export PNG, SVG, PDF, or JPG. The skill parses natural language, Mermaid, or structured inputs into laid-out diagrams using Graphviz placement and transitive reduction for clean architecture visuals.
The latest update addresses a critical export flaw: SVG files now correctly embed AWS, Azure, and GCP icons via the --embed-svg-images flag in scripts/svgflow.py and scripts/drawiohtml.py, preventing broken references when diagrams are opened on other machines. A regression test confirms the flag is applied in both export paths.
The catch: The skill depends on having draw.io desktop CLI installed and configured, limiting use in fully headless or restricted CI environments without GUI tooling access.

Use Cases
  • Developers generate architecture diagrams from codebase structure
  • DevOps teams auto-create infrastructure visuals from Terraform
  • Engineers produce ERDs directly from SQL schema definitions

Source: Agents365-ai/drawio-skill — based on the README and release notes.

OpenLogi v0.7.7 fixes macOS permissions and CI builds for Logitech device control 🔗

Native Rust tool remaps buttons, DPI, and SmartShift without accounts or telemetry

AprilNEA/OpenLogi · Rust · ▲ 14.2k in 1d 3mo old

OpenLogi released v0.7.

7 with fixes for macOS permission error diagnostics and CI builds on both macOS legs using Icon Composer-capable Xcode. The local-first, Rust-based alternative to Logitech Options+ remaps mouse buttons, DPI, and SmartShift over HID++, supports gestures on any button, and uses plain-text TOML configs. It runs on macOS, Linux, and Windows with per-application profiles and CLI/GUI dual interfaces. No account or telemetry is required.
The catch: OpenLogi remains under active development with 261 open issues, and features/config may still change before stabilization.

Use Cases
  • Developers remapping MX Master buttons for IDE shortcuts
  • Linux users customizing Logitech mouse gestures without Options+
  • Creators adjusting Litra light brightness via per-app profiles

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

Unsloth Adds Auto Compaction for Extended Local Chat Sessions 🔗

Experimental feature preserves searchable history beyond model context limits via RAG indexing

unslothai/unsloth · Python · 74.4k stars Est. 2023

Unsloth’s latest release introduces Auto Compaction (Experimental), enabling users to conduct long local chats that exceed a model’s context window by rolling older turns into a searchable archive. Evicted messages remain retrievable through Unsloth’s RAG pipeline, preserving fact recall without permanent trimming.

The update also includes Remote & LAN Access (Preview), faster chat streaming, and support for custom llama.cpp builds with toggles for cache and decoding options. Built in Python, the desktop app supports LLMs like Qwen3.8 and Gemma 4, diffusion models, and agent integration with Claude Code and Codex. The catch: Auto Compaction remains experimental, with no guarantees on latency or search accuracy under heavy usage.

Use Cases
  • Developers testing long-context agent workflows locally
  • Researchers fine-tuning LLMs with constrained VRAM
  • Teams deploying private LLMs via OpenAI-compatible API endpoints

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

Proma v0.17.55.0 Unifies AI Agents and Chat in Local-First Workspace 🔗

v0.15.0 adds Pi Agent Runtime, workspace isolation, and cross-platform desktop support

proma-ai/Proma · TypeScript · 2.1k stars 6mo old

Proma delivers a general-purpose Agent experience by combining multi-model chat, task automation, and persistent memory in a local-first desktop app. Built with TypeScript, it runs agents via its native Pi Runtime, supports file operations, browser automation, and skills via MCP.

Users configure AI providers locally, keeping data in ~/.proma/ using JSON files. The app enables collaborative task splitting, remote bot bridging (Feishu, WeChat), and voice input across macOS and Windows. Recent updates extended collaboration wait times and improved agent progress reporting.
The catch: Heavy reliance on local setup and manual AI provider configuration may deter teams seeking zero-config, cloud-native agent platforms.

Use Cases
  • Developers automate multi-step code reviews using Agent mode
  • Analysts compare model outputs with attached research documents
  • Teams deploy shared Skills via organizational licensing for consistent workflows

Source: proma-ai/Proma — based on the README and release notes.

elizaOS Powers Autonomous AI Agents Across Desktop, Mobile, and Cloud 🔗

Latest release adds immutable PR evidence assets for auditable agent workflows

elizaOS/eliza · TypeScript · 19.1k stars Est. 2024

elizaOS provides a TypeScript framework for building autonomous AI agents with plugins for chat, voice, memory, and crypto wallet operations. It supports web, desktop, and mobile targets through its Eliza app and core runtime.

The framework is model-agnostic and extends via first-party plugins for scheduling, automation, and device bridges. Recent commits focus on infrastructure, with the last push showing routine maintenance rather than feature expansion. The catch: Despite active forks, nearly 600 open issues suggest stability and documentation gaps may hinder production adoption.

Use Cases
  • Developers building crypto-native AI agents with wallet approval controls
  • Enterprises deploying autonomous assistants across Slack, Discord, and Telegram
  • Creators orchestrating multi-agent workflows for browser and desktop automation

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

GitHub Agentic Workflows Hardens Sandbox for Safer AI Automation 🔗

Latest release strengthens Docker isolation and MCP gateway security for reliable agent runs

github/gh-aw · Go · 5k stars Est. 2025

GitHub's gh-aw extension compiles Markdown-defined AI workflows into standard GitHub Actions, letting developers run agents like Copilot or Claude Code for tasks such as issue triage or CI failure analysis. The v0.

86.2 release focuses on hardening the Docker sandbox (docker-sbx) and MCP gateway, improving safe-output reliability and threat detection. Agent jobs remain read-only and sandboxed by default, with writes handled via validated safe-outputs jobs. The catch: Open issues exceed 370, indicating ongoing stability and usability challenges despite recent security fixes.

Use Cases
  • Developers automate issue triage using AI reasoning
  • Teams investigate CI failures with agentic workflows
  • Maintainers run dependency analysis via secure agents

Source: github/gh-aw — based on the README and release notes.

LLVM 22.1.8 Release Adds Binary Patches for Linux and macOS Builds 🔗

Project maintains 38,400 open issues while supporting C++ toolchains across platforms

llvm/llvm-project · LLVM · 39.9k stars Est. 2016

The LLVM project released version 22.1.

8 with updated binary distributions for Linux x86_64, Arm64, and macOS Apple Silicon. These patches include the core compiler infrastructure, Clang frontend, libc++, and LLD linker. Builders use LLVM to compile C-family languages into optimized object files via intermediate representation. Despite steady updates, the project carries 38,400 open issues indicating ongoing complexity in its modular toolchain.
The catch: High issue volume suggests maintenance overhead may slow adoption of new features in safety-critical systems.

Use Cases
  • Compiler developers building C++ toolchains for embedded systems
  • Performance engineers optimizing runtime environments for cloud workloads
  • Toolchain creators integrating bitcode analysis into custom linkers

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

Open Source AI Agents Evolve from Tools to Full Operating Systems 🔗

Developers now build autonomous agents with memory, skills, and self-improvement as core primitives

Trendai-agents
Yuzzyuk/marketing-osjaredrhod/fullstack-agentgithub/gh-awAgents365-ai/drawio-skillunslothai/unslothelizaOS/elizaproma-ai/Promavolcengine/OpenVikingakitaonrails/ai-memoryifixai-ai/iFixAibojieli/ai-agent-bookayghri/i-have-adhdchaitanyagiri/munder-difflintt-a1i/archifyblader/humanizerHiThink-Tech/Financial-APIcalesthio/OpenMontagellm-as-a-verifier/llm-as-a-verifierinternet-court/internet-court-skillcan1357/oh-my-pivectorize-io/hindsightPrimeIntellect-ai/prime-agentMemTensor/memmy-agentWenyuChiou/awesome-agentic-ai-zhhenryqin1997/statemvercel-labs/agent-browserchuspeeism/dashi-ppt-skillnuyoah-ai-works/nuyoah-xiezhen-promptalibaba/open-code-reviewTencentCloud/TencentDB-Agent-MemorySnailclimb/JavaGuideGoogleCloudPlatform/generative-aied-donner/agentsob-f/OpenBotcurl/curljustrach/codedbmit-acl/mightythedivergentai/GD-Agentic-Skills

The open source landscape is rapidly shifting toward integrated AI agent ecosystems where agents aren’t just scripts but self-contained systems with persistent memory, multimodal interfaces, and autonomous skill chaining. Projects like elizaOS/eliza offer a TypeScript-based agentic operating system, while proma-ai/Proma delivers a seamless general-purpose agent experience designed for professional workflows.

Memory is no longer an afterthought—vectorize-io/hindsight provides agent memory that learns, and akitaonrails/ai-memory enables long-term context retention for agent coding CLIs. Skill marketplaces are emerging too: Agents365-ai/drawio-skill generates 10,000+ diagram types from natural language, calesthio/OpenMontage turns AI assistants into full video studios with 12 production pipelines, and HiThink-Tech/Financial-API supplies real-time A-share data for agent-driven quant research. Self-improvement is gaining traction via PrimeIntellect-ai/prime-agent, a self-improving RLM agent for coding tasks, and llm-as-a-verifier/llm-as-a-verifier, which delivers SOTA feedback without retraining. Even trust layers are being solved—internet-court/internet-court-skill enables agent-to-agent commerce with escrow and dispute resolution.
The catch: Despite rapid innovation, the ecosystem remains fragmented across languages and frameworks, with many agents lacking robust safety guarantees, standardized evaluation, or clear paths to production reliability—making interoperability and long-term stewardship uncertain.

Use Cases
  • Developers deploy self-improving coding agents for autonomous bug fixes
  • Marketing teams use agent skills to generate and grade ad copy in real time
  • Financial analysts run AI agents that audit trades using live market data

Open Source Tools Turn LLMs Into Modular, Agentic Workforce Extensions 🔗

Projects now package AI capabilities as plug-and-play skills for coding agents and local model runners

Trendllm-tools
Yuzzyuk/marketing-osgithub/gh-awAgents365-ai/drawio-skillunslothai/unslothelizaOS/elizaproma-ai/Promaapache/seatunnelcathrynlavery/diagram-designWei-Shaw/sub2apiTiger3807861189/J-Space-Cognition-Suite-V3.7citrolabs/ego-litezhaoxuya520/reverse-skillvirgiliojr94/book-to-skillAlishahryar1/free-claude-codellm-as-a-verifier/llm-as-a-verifierinternet-court/internet-court-skilllidge-jun/opencodexlightningpixel/modlycoreyhaines31/marketingskillsFailproofAI/failproofaifreestylefly/awesome-gpt-image-2MadsLorentzen/ai-job-searchshy3130/tickflow-stock-paneljundot/omlxMemTensor/memmy-agentWenyuChiou/awesome-agentic-ai-zhmultica-ai/multicabradautomates/claude-videonuyoah-ai-works/nuyoah-xiezhen-promptalibaba/open-code-reviewZhuLinsen/daily_stock_analysisTencentCloud/TencentDB-Agent-MemoryOsmantic/ODSGoogleCloudPlatform/generative-aied-donner/llm_engineeringTingsongYu/PyTorch-Tutorial-2ndaliasrobotics/caijustrach/codedb

A clear pattern is emerging where open source repos package specific LLM-driven capabilities as reusable, interoperable skills—effectively turning language models into extensible toolchains. Rather than monolithic apps, developers are sharing granular AI functions: drawio-skill generates 11 diagram types from natural language using vision self-check; marketing-os delivers a full Claude skill with 14 modules for copy grading, ad diagnosis, and launch planning; book-to-skill converts technical PDFs into instant Claude Code references.

These skills plug into agents like Claude Code, Codex, or Cursor via standardized interfaces. Simultaneously, local enablement tools are rising: unsloth offers a UI to run and train models like Qwen3.8 and Gemma 4 locally; omlx provides SSD-cached LLM inference on Apple Silicon via menu bar; memmy-agent gives agents shared persistent memory across frameworks. Agent orchestration is also evolving: gh-aw enables GitHub Agentic Workflows; multica assigns issues to 18 coding agents as teammates; opencodex proxies any LLM to Codex CLI. This shift treats LLMs not as endpoints but as composable runtimes—where skills are versioned, shared, and combined like libraries. The catch: While promising, the ecosystem remains fragmented across agent protocols (Claude Code vs. Codex vs. OpenClaw), skill formats lack universal standards, and local model performance still lags behind cloud APIs for complex reasoning—raising questions about real-world reliability and long-term interoperability beyond demo-friendly use cases.

Use Cases
  • Developers generate UML diagrams from prompts using AI
  • Marketers audit copy and ads via Claude Code skills
  • Researchers train LLMs locally on consumer hardware

Rust Powers Next-Gen Open Source Infrastructure Across Stacks 🔗

From storage to AI agents, Rust enables high-performance, interoperable tools redefining modern open source foundations.

Trendrust-systems
rustfs/rustfsAprilNEA/OpenLogiakitaonrails/ai-memoryblock/buzzAlexsJones/llmfitherdrdev/herdrfirecrawl/anydoczenbu-labs/terminal-browserfirecrawl/pdf-inspectorgastongouron/ironpressmacro-inc/macrohydra-db/hydradbRyanCodrai/turbovecvercel-labs/agent-browserros2-rust/ros2_rustdenoland/denoFuelLabs/sway

A clear pattern is emerging in open source: Rust is no longer just for systems programming—it’s becoming the foundational language for performance-critical, interoperable infrastructure across the stack. Projects like rustfs/rustfs demonstrate this shift, delivering 2.

3x faster S3-compatible object storage for small payloads than MinIO, enabling efficient coexistence with existing platforms. In the agent ecosystem, akitaonrails/ai-memory and herdrdev/herdr provide long-term memory and runtime environments for AI agent collaboration, while vercel-labs/agent-browser offers CLI-based browser automation for agent workflows. On the productivity front, AprilNEA/OpenLogi delivers a local-first, telemetry-free alternative to Logitech Options+ via HID++ remapping, and firecrawl/anydoc converts complex documents to clean Markdown with bindings for Node.js and Python. Even niche domains are seeing Rust adoption: hydra-db/hydradb builds a fast graph database on object storage, turbovec/turbovec provides a vector index with Python bindings via TurboQuant, and firecrawl/pdf-inspector intelligently classifies PDFs for smart routing. Together, these repos show Rust enabling high-performance, dependency-light tools that prioritize interoperability, local control, and vendor neutrality—from storage and databases to AI agent runtimes and desktop utilities. The pattern reflects a broader move toward infrastructure that is both fast and embeddable, reducing reliance on heavyweight runtimes or proprietary systems.

The catch: While promising, many of these projects remain early-stage, with limited production validation, fragmented ecosystems, and uncertain long-term maintenance—raising questions about whether Rust’s advantages in safety and speed will translate to sustainable, widely adopted infrastructure beyond niche builder enthusiasm.

Quick Hits

repowise Analyzes code health, generates docs, tracks git activity, detects dead code, and captures architectural decisions via MCP for smarter development. 6.2k
lanshu-create-ai-presenter-video Creates verified AI presenter videos from scripts and authorized images using a provider-neutral Codex Skill. 353
rustfs Delivers 2.3x faster S3-compatible object storage than MinIO for 4KB payloads, supporting seamless migration and coexistence with other platforms. 31.3k
fullstack-agent Sets up a full-stack AI agent with memory, voice, face, and hands via an installation wizard — pick components or deploy all at once. 269
register Enables users to claim a personalized '.is-a.dev' subdomain for a clean, memorable online identity. 11.1k
Comfyui_Minimax_h3_latent_Upscaler Upscales low-res latents directly for Minimax H3 video generation, bypassing expensive VAE steps to accelerate high-res output and outperform naive interpolation. 281
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 v2026.7.1-2 fixes npm plugin metadata parsing for stable assistant updates 🔗

Patch resolves singleton-array handling to ensure official plugins install correctly across Node.js runtimes

openclaw/openclaw · TypeScript · ▲ 113 in 1d 9mo old · Latest: v2026.7.1-2

OpenClaw’s latest release patches a critical npm plugin update failure that blocked tracked official plugins from installing correction releases. The fix, merged in pull request #108336, adjusts metadata parsing to accept singleton-array inputs from newer npm clients—resolving a silent failure that left plugins outdated or uninstalled.

This ensures the Gateway, which orchestrates models, tools, and messaging channels like WhatsApp, Slack, and iMessage, stays current with security and feature updates.

The assistant runs locally on macOS, Linux, and Windows, provisioning Node.js 22.22.3+ or newer when needed. After onboarding, users interact via Control UI, CLI, or TUI, connecting to hosted or local LLMs through a unified Gateway. Tools execute on-host by default, with sandboxing optional but undocumented in this release.

The catch: Despite 6,034 open issues and rapid fork growth, the project lacks clear sandboxing defaults or audit logs for tool execution—a gap builders must self-manage before exposing the assistant to untrusted inputs or multi-user environments.

Previously in The Times “covered” — Aug 22

Use Cases
  • Developers syncing AI workflows across WhatsApp and Slack
  • Power users running local LLMs with custom tool plugins
  • Teams testing private assistants on iMessage and Signal without cloud deps

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

More Stories

LLM Engineering Course Repo Gains Practical Traction Through Community Forks 🔗

Over 6,900 forks signal widespread adoption despite 381 open issues and aging codebase

ed-donner/llm_engineering · Jupyter Notebook · 7.1k stars Est. 2024

The ed-donner/llm_engineering repository supports a hands-on mastering LLM engineering course using Jupyter Notebooks. Learners build projects incrementally, starting with Ollama installation to run local models immediately.

Recent activity shows steady maintenance, with the last commit just one day ago, though the project is ~2.0 years old. Its popularity is evident in 6,961 forks, reflecting broad reuse by developers seeking applied LLM skills. However, 381 open issues suggest ongoing challenges in content clarity, tool compatibility, or exercise debugging that may hinder self-paced learners.
The catch: The course assumes tolerance for iterative troubleshooting in a fast-evolving LLM landscape where notebooks can quickly become outdated.

Use Cases
  • Developers learning local LLM deployment with Ollama
  • Educators adopting structured LLM curricula
  • Teams prototyping LLM applications incrementally

Source: ed-donner/llm_engineering — based on the project README.

Agentic AI Course Updates with Free Model Options After 14 Months 🔗

Repository shows recent activity despite no new releases since March 2025

ed-donner/agents · Jupyter Notebook · 6k stars Est. 2025

The ed-donner/agents repository hosts Jupyter Notebooks for a 6-week Agentic AI Engineering course, recently updated to support free local models like Ollama and DeepSeek alongside OpenAI. Last pushed five months ago with commits as recent as one day ago, it reflects ongoing maintenance without formal version releases.

Learners configure API keys for model access, with setup guides for Windows, macOS, and Linux. The course includes video resources, avatar-driven FAQs, and guidance on monitoring usage costs. The catch: Over 200 open issues suggest unresolved user problems, and the lack of versioned releases makes tracking specific changes difficult for adopters.

Use Cases
  • Engineers learn agentic AI using Jupyter notebooks
  • Teams prototype AI agents with local open-source models
  • Developers course-cursor setup via platform-specific guides

Source: ed-donner/agents — based on the project README.

Caveman Skill Cuts Agent Token Use by 65% with New Proxy 🔗

Benchmarks show 33.2% fewer input tokens read by agents using Caveman Proxy v2.2.0

JuliusBrussee/caveman · Go · ▲ 182 in 1d 4mo old

JuliusBrussee/caveman’s latest release adds a proxy layer that shrinks what agents read before provider calls, achieving byte-exact recovery. The skill itself still outputs tight caveman-speak while preserving code and command fidelity.

Installers now use secure temp directories and avoid clobbering other plugins’ hooks. Benchmarks confirm reduced token consumption without breaking agent functionality.
The catch: Open issues exceed 360, suggesting usability gaps in complex agent setups despite performance gains.

Previously in The Times “covered” — Aug 16

Use Cases
  • Developers reduce Claude Code API costs
  • Teams minimize LLM context window overflow
  • Engineers automate token-efficient agent workflows

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

Quick Hits

supabase Supabase provides a dedicated Postgres database with built-in auth, storage, and edge functions to accelerate full-stack app development. 108.3k
gemini-cli Gemini CLI lets you interact with Google’s Gemini AI models directly from your terminal for fast, scriptable AI workflows. 106.6k
firecrawl Firecrawl enables scalable web scraping and data extraction via a simple API, turning any site into structured context for LLMs. 171.1k
julia Julia is a high-performance, dynamic programming language designed for scientific computing, machine learning, and large-scale numerical analysis. 49k
prompts.chat Prompts.chat lets you discover, share, and self-host AI prompts privately — ideal for teams needing secure, collaborative prompt engineering. 167.8k

Newton GPU Physics Engine Gains Vectorized Joint Control for Robot Learning 🔗

v1.5.0 adds batched impedance controllers and MuJoCo sleep optimizations for scalable simulation

newton-physics/newton · Python · ▲ 4 in 1d Est. 2025 · Latest: v1.5.0

Newton v1.5.

0 introduces experimental vectorized joint control via the new newton.controllers module, enabling model-based and caller-supplied impedance controllers for heterogeneous GPU robot batches. This release expands scalable batched simulation workflows with isolated implicit MPM, masked solver resets, and dedicated global-world gravity to prevent cross-world state loss in multi-robot learning. Opt-in MuJoCo Warp sleeping now supports tree sleeping, contact-based wake-up, and initial sleep policies, allowing large numbers of articulated systems to remain inactive without computation overhead. Contact reliability improves through deterministic hydroelastic generation, persistent geometry, and stable box manifolds, while cable mechanics gain shear and twist helpers. Built on NVIDIA Warp and integrating MuJoCo Warp as its primary backend, Newton emphasizes GPU acceleration, OpenUSD support, and differentiability for rapid iteration in robotics research. The project remains Linux Foundation-hosted, Apache-2.0 licensed, and requires NVIDIA GPUs (Maxwell or newer) with driver 545+; macOS runs CPU-only.
The catch: Despite active development (391 open issues, last commit 0 days ago), Newton’s GPU dependency excludes AMD or Intel hardware, and its reliance on NVIDIA Warp locks users into a specific vendor stack, limiting portability for teams requiring cross-platform GPU simulation.

Previously in The Times “covered” — Aug 18

Use Cases
  • Robotics researchers running batched reinforcement learning
  • Simulation engineers optimizing contact-rich mechanical systems
  • Developers extending physics engines with user-defined GPU kernels

Source: newton-physics/newton — based on the README and release notes.

More Stories

UBC-Thunderbots Refine Soccer AI with Wheel Ramping and Bug Fixes 🔗

Simulation gains realistic acceleration as RoboCup SSL team patches kick velocity and network typos

UBC-Thunderbots/Software · C++ · 70 stars Est. 2018

The UBC-Thunderbots updated their RoboCup SSL robot soccer software with simulation improvements and critical fixes. Wheel ramping now models gradual motor acceleration in simulation, enhancing motion realism.

Developers resolved a mirrored ball kick velocity error on inverted fullsystem tabs and corrected a gamecontroller network interface typo. The codebase, primarily in C++, supports autonomous navigation, vision, and strategy for small-size league robots. Despite active development, 229 open issues suggest ongoing challenges in stability and feature completeness.
The catch: High issue count raises questions about production readiness for complex real-world deployment beyond simulation.

Use Cases
  • University teams testing SSL AI algorithms
  • Developers simulating robot soccer physics
  • Contributors refining autonomous robot control systems

Source: UBC-Thunderbots/Software — based on the README and release notes.

Drake v1.56.0 bundles SNOPT solver for robotics optimization 🔗

Pre-compiled solver streamlines mathematical program toolbox in C++ framework

RobotLocomotion/drake · C++ · 4.2k stars Est. 2014

RobotLocomotion/drake released v1.56.

0 with a pre-compiled SNOPT solver integrated into its Mathematical Program toolbox. The update removes manual setup for this optimization backend, accelerating deployment for dynamics and control workflows. Drake enables model-based design, simulation, and verification for complex robotic systems, widely used in research and industry for manipulator and legged robot development. Its C++ core ensures high performance but requires significant build expertise.
The catch: Steep learning curve and build complexity limit accessibility for teams without dedicated robotics engineering resources.

Use Cases
  • Researchers simulating quadruped locomotion dynamics
  • Engineers verifying manipulator control algorithms
  • Academics teaching model-based robotics design principles

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

OpenBot v0.8.0 adds Flutter controller and WebRTC video feed for low-cost smartphone robots 🔗

New release enables cross-platform teleoperation and remote web server control via Android/iOS

ob-f/OpenBot · Swift · 3.4k stars Est. 2020

OpenBot’s latest release v0.8.

0 introduces a Flutter-based controller supporting both Android and iOS, complete with real-time video streaming over WebRTC. The update also adds a remote web server for teleoperation and improves Wi-Fi AP-based connections between robot and controller phones. Built on a $50 electric vehicle body, OpenBot uses smartphones for AI workloads like person following and autonomous navigation. The project maintains active development with 35 open issues and recent commits.
The catch: iOS support remains experimental, with core robotics features like autonomous navigation still limited to Android.

Use Cases
  • Educators teach robotics using smartphone brains and $50 robot bodies
  • Hobbyists build person-following robots with Android and Arduino
  • Researchers test real-time navigation policies via Flutter controller and WebRTC feed

Source: ob-f/OpenBot — based on the README and release notes.

Quick Hits

mujoco Simulates complex multi-joint robot dynamics with contact physics for accurate motion prediction and control testing. 14.6k
ros2_rust Enables Rust developers to build ROS 2 nodes with safe, high-performance bindings for robotics applications. 1.5k
yarp Provides a flexible middleware framework for robot communication, device integration, and real-time control across heterogeneous systems. 602
IsaacLab Offers a unified Python framework for end-to-end robot learning, leveraging Isaac Sim for simulation-to-real transfer. 7.9k
ros-mcp-server Bridges AI models like Claude and GPT to physical robots via MCP and ROS, enabling natural language control of robotic systems. 1.4k

OpenCTI's Latest Patch Fixes STIX Message Handling to Prevent Queue Blockages 🔗

The update resolves a critical backend bug where malformed STIX inputs could indefinitely stall worker processes.

OpenCTI-Platform/opencti · TypeScript · ▲ 2 in 1d Est. 2018 · Latest: 7.260817.0

OpenCTI’s recent patch targets a long-standing reliability issue in its backend worker system. Malformed STIX2 messages were previously not properly acknowledged (nacked), causing message queues to block indefinitely and disrupting threat intelligence pipelines.

Fix #17686 now ensures these messages are correctly rejected, preventing worker stalls and maintaining queue throughput.

The release also includes on-demand deployment for pre-built feature branch images (feat #17620), streamlining CI/CD workflows for teams testing changes in isolation. Additional fixes address case template authorization, Attack Pattern upserts, and OTP handling in authentication streams.

Built on TypeScript with a GraphQL API and STIX2-based knowledge schema, OpenCTI remains a go-to for correlating technical observables with non-technical threat context across tools like MISP and MITRE ATT&CK. Its dual CE/EE model lets organizations start open source and scale to enterprise features as needed.

The catch: Despite active development, over 2,100 open issues suggest ongoing stability challenges, and the platform’s complexity may overwhelm teams without dedicated CTI analysts or integration resources.

Previously in The Times “covered” — Aug 21

Use Cases
  • Security teams enriching threat data with MITRE ATT&CK TTPs
  • Analysts linking observables to reports and attribution sources
  • SOCs automating CTI ingestion from MISP and TheHive via connectors

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

More Stories

Archived CAI framework leaves legacy in AI security research 🔗

Final commit preserves source behind 18 papers and 30+ CVEs, now superseded by CSI

aliasrobotics/cai · Python · 9.8k stars Est. 2025

The aliasrobotics/cai repository, archived in March 2025, holds the complete source of the Cybersecurity AI framework. It powered AI-driven offensive and defensive security automation, cited in over 18 research papers and linked to 30+ CVEs.

Last pushed in August 2026, it remains read-only as a research artifact. All packages, issues, and pull requests are still accessible. Professional users are directed to migrate to Cybersecurity Superintelligence (CSI) for continued support. The catch: No further updates, patches, or support will be provided, leaving existing deployments vulnerable to unaddressed threats.

Use Cases
  • Researchers studying AI-powered penetration testing techniques
  • Organizations auditing legacy AI security toolchains
  • Educators teaching foundations of generative AI in cyber defense

Source: aliasrobotics/cai — based on the project README.

Subfinder v2.16.0 sharpens passive subdomain discovery with source fixes 🔗

New malware source added, result capping, and DNS timeout guard improve reliability

projectdiscovery/subfinder · Go · 14.3k stars Est. 2018

ProjectDiscovery’s subfinder released v2.16.

0, refining its Go-based passive subdomain enumerator. The update honors -max-results across paginated sources and blocks partial host matches in extractors. Untrusted source responses are now size-capped, and DialContext replaces deprecated Dial for safer connections. A new scanmalware passive source expands OSINT reach. Crtsh queries now set statement_timeout to avoid PgBouncer handshake failures. The tool remains lightweight, supports JSON/file/stdout output, and integrates via stdin/stdout. The catch: passive enumeration misses active or recently created subdomains, limiting coverage in dynamic environments.

Use Cases
  • Bug bounty hunters scoping target domains
  • Pen testers mapping external attack surfaces
  • OSINT analysts gathering passive intelligence
  • DevSecOps teams automating recon in CI/CD pipelines

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

Dirsearch v0.5.0 adds Rust backend and async scanning by default 🔗

Release supports Python 3.11-3.14 with improved filtering and Nmap XML import

maurosoria/dirsearch · Python · 14.7k stars Est. 2013

Dirsearch, a Python web path scanner, now defaults to async scanning and offers a Rust-native backend via --request-backend native for faster GET requests and wordlist generation. The v0.

5.0 release adds advanced match/filter controls for status, size, and elapsed time, plus Nmap XML target import and per-target runtime limits. Wordlists gain category-based loading and template placeholders like %API_VERSION% and %DATE%. Pre-built binaries and Docker images remain available.
The catch: Active development relies on just two maintainers, with 32 open issues and no indication of expanded contributor growth despite ongoing use in bug bounty and red teaming.

Use Cases
  • Security teams scanning web apps for hidden directories
  • Bug bounty hunters enumerating subdirectories during recon
  • Red teams using API to automate authenticated scans in custom toolchains

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

Quick Hits

x64dbg A powerful open-source Windows debugger optimized for reverse engineering and malware analysis, offering deep inspection of user-mode processes. 49.2k
setup-ipsec-vpn Quickly deploy a secure IPsec VPN server with multi-protocol support and client configs across major Linux distros and Raspberry Pi OS. 28.4k
ntopng Monitor and analyze network traffic in real time via a web interface, detecting threats and anomalies with Lua-powered extensibility. 8.1k
authelia Provides enterprise-grade SSO and MFA for web apps, now OpenID Certified™, enabling secure, unified access control. 28.7k
bunkerweb Delivers cloud-native WAF protection as an open-source solution, defending web apps against common exploits with flexible rule management. 10.8k

ClickHouse 26.3 LTS Release Adds Rust Integration for Query Acceleration 🔗

New Rust-based UDF framework enables custom analytics functions without recompiling the core engine

ClickHouse/ClickHouse · C++ · ▲ 6 in 1d Est. 2016 · Latest: v26.3.21.7-lts

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

21.7-lts, introduces a stable Rust user-defined function (UDF) interface, allowing developers to write performance-critical analytics logic in Rust and load it dynamically into queries. This avoids the prior necessity of compiling custom C++ plugins against ClickHouse’s internal ABI, reducing extension development friction and improving safety. The feature leverages ClickHouse’s plugin system, first exposed in v22.8, now matured with Rust toolchain support via the clickhouse-udf crate. Early adopters report 2–5x speedups in complex filtering and aggregation workloads compared to equivalent SQL UDFs. The release also includes incremental improvements to the adaptive sparse index and enhanced support for Apache Iceberg v2 tables in lakehouse architectures. Documentation now includes a Rust UDF tutorial with examples for JSON parsing and regex matching, signaling a shift toward extensibility as a core competency.

The catch: Rust UDFs still require manual memory management for complex data structures, and debugging runtime panics remains challenging due to limited integration with ClickHouse’s query tracing system.

Previously in The Times “covered” — Aug 17

Use Cases
  • Financial analysts accelerate fraud detection with custom Rust UDFs
  • Data engineers optimize ETL pipelines using Rust for high-throughput transformations
  • Analytics teams deploy domain-specific functions without rebuilding ClickHouse binaries

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

More Stories

Ollama Cuts First-Token Latency by Half with New Metadata Caching 🔗

v0.32.15 release improves response speed for local LLMs like Qwen and Gemma

ollama/ollama · Go · ▲ 80 in 1d Est. 2023

Ollama’s latest release cuts time-to-first-token from ~995 ms to ~524 ms by caching resolved model metadata between requests. The update also normalizes Qwen 3.

8 system message handling and fixes a parser error that could wedge chat or generate streams. Built in Go, it supports local execution of models including DeepSeek, GLM-5.2, and MiniMax via simple CLI commands. Developers use it to run LLMs offline for private agents, code assistants, and RAG pipelines without API fees or data leakage.
The catch: Performance gains depend on sufficient RAM and SSD speed, limiting effectiveness on low-end hardware.

Previously in The Times “covered” — Aug 16

Use Cases
  • Run local LLMs for private code generation
  • Deploy offline AI agents on developer workstations
  • Integrate models into self-hosted chat applications

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

Deno 2.9.5 Adds Unscoped Package Aliasing and Workspace Task Filtering 🔗

Latest release improves developer control over dependency resolution and task execution in monorepos

denoland/deno · Rust · ▲ 17 in 1d Est. 2018

Deno 2.9.

5 introduces the --unscoped flag to alias packages by their bare names, simplifying imports in large codebases. The new --members flag for deno task lets developers run scripts only on specific workspace members, avoiding unnecessary execution across monorepos. Additional fixes resolve bundling deadlocks, crypto API gaps, and Node.js compatibility issues in file system and streams handling. The project continues to rely on Rust, V8, and Tokio for its secure, permission-by-default runtime. The catch: Despite rapid iteration, Deno’s adoption in production backend systems remains limited compared to Node.js, leaving teams to weigh its DX benefits against ecosystem maturity and hiring familiarity.

Use Cases
  • Backend developers building secure REST APIs with TypeScript
  • Frontend engineers creating isomorphic apps using shared code
  • DevOps teams automating deployment scripts with fine-grained permissions

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

Codebase-Memory MCP Server Fixes Graph Query Accuracy in v0.10.8 🔗

Patch resolves Cypher aggregation undercount and Python import alias resolution

DeusData/codebase-memory-mcp · C · ▲ 2 in 1d 5mo old

The DeusData/codebase-memory-mcp server now correctly counts relationship matches and preserves aliased Python from-imports in its knowledge graph. Cypher count() and collect() operations no longer truncate results prematurely, fixing edge cases in large repositories.

Python aliases like from .gate import execute as bridge_execute now generate accurate CALLS edges to the original function definition. These fixes, merged from 19 community pull requests, improve structural query reliability for AI coding agents using the Model Context Protocol. The tool still indexes average repositories in milliseconds and the Linux kernel in three minutes, producing sub-ms responses via its persistent graph.
The catch: Open issues remain high at 486, indicating ongoing stability and feature work despite recent patches.

Previously in The Times “covered” — Aug 16

Use Cases
  • AI agents querying call chains in monorepos
  • Debugging inheritance graphs in Django projects
  • Visualizing cross-service links locally at localhost:9749

Source: DeusData/codebase-memory-mcp — based on the README and release notes.

Quick Hits

lede Provides the LEDE source code for building customizable, lightweight Linux-based firmware for embedded devices and routers. 31.6k
sway Enables developers to create secure, high-performance smart contracts with Rust-based tooling and formal verification support. 61.5k
codex Offers a lightweight, terminal-native coding agent that assists with code generation, editing, and debugging directly in your shell. 114.2k
rustdesk Delivers a self-hostable, open-source remote desktop solution with cross-platform support and end-to-end encryption as a TeamViewer alternative. 121.5k
ghostty Delivers a fast, GPU-accelerated, cross-platform terminal emulator with native UI integration and rich feature support for power users. 60.1k

OpenIPC Firmware Expands to Support Ambarella and SigmaStar Chips 🔗

Community-driven IP camera firmware now runs on 11 SoC families after years of incremental porting

OpenIPC/firmware · C · ▲ 3 in 1d Est. 2021 · Latest: nightly

OpenIPC’s alternative firmware for IP cameras has quietly added support for Ambarella and SigmaStar system-on-chips, broadening its hardware compatibility beyond the original HiSilicon focus. Built with Buildroot and U-Boot, the C-based project now enables developers to flash custom firmware on devices using chips from Anyka, Goke, GrainMedia, Ingenic, MStar, Novatek, XiongMai, and others.

The latest nightly release, built August 22, includes updated web UI assets and kernel modules tailored to these newer architectures.

Unlike vendor-locked firmware, OpenIPC offers RTSP streaming, motion detection, and ONVIF compliance through a community-maintained codebase. Users can access the system via Telnet or SSH, modify configurations in /etc, and contribute patches through GitHub. The project emphasizes transparency, with all build scripts and device tree overlays publicly available.

Paid commercial support remains an option for businesses needing SLA-backed maintenance, while free support flows through Telegram chat and issue triage. Recent activity shows sustained engagement: 263 open issues, 456 forks, and a commit just hours ago.

The catch: Despite wide SoC coverage, OpenIPC still lacks official support for newer AI-enabled vision processors from companies like Himax or Rockchip, limiting its use in advanced edge analytics scenarios requiring hardware-accelerated inference.

Previously in The Times “covered” — Aug 17

Use Cases
  • Developers flashing custom firmware on legacy HiSilicon IP cameras
  • Hobbyists building DIY FPV drone video links with Anyka-based modules
  • Small businesses deploying ONVIF-compliant cameras without vendor lock-in

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

More Stories

GPU-T Delivers GPU-Z-Style Diagnostics for Linux via .NET AppImage 🔗

Latest release cuts AppImage size nearly in half with .NET 10 migration and HarfBuzz fix

lseurttyuu/GPU-T · C# · 321 stars 7mo old

GPU-T provides detailed AMD and NVIDIA GPU diagnostics on Linux, reading specs, sensors, and feature support directly from sysfs and graphics APIs. Built with .

NET and Avalonia UI, it ships as a rootless AppImage with hardware lookup, real-time monitoring, and ReBAR detection. The v0.2.2 release migrated to .NET 10 LTS, optimized the binary, and fixed a first-launch crash by bundling a unified HarfBuzz library. Intel GPU support remains planned but unimplemented.
The catch: Intel GPU support is still pending, limiting full cross-vendor hardware coverage on Linux.

Use Cases
  • Developers verify GPU specs and driver versions
  • Gamers monitor real-time clock speeds and temperatures
  • System admins log sensor data for performance tuning

Source: lseurttyuu/GPU-T — based on the README and release notes.

ESP32-based AIOsense sensor merges modularity with Home Assistant integration 🔗

Latest release fixes BME280 compatibility for ESPHome 2024.2.0 and updates dependencies

Schluggi/AIOsense · Unknown · 161 stars Est. 2022

Schluggi’s AIOsense project offers an all-in-one, solder-friendly ESP32-C3 sensor board supporting mmWave radar, PIR, BME280, light, humidity, and VOC-equivalent sensing. Designed for DIY smarthome builders, it avoids SMD parts and emphasizes upgradeability—users can swap sensors without replacing the entire PCB.

The latest release, esphome-v3.0.1, addresses BME280 driver updates for ESPHome 2024.2.0 and refreshes dependencies. Power draw ranges from 0.11W idle without mmWave to 0.45W with full sensor load. The catch: 13 open issues remain unresolved, and voice assistant support is still listed as "coming soon," raising questions about real-time readiness for advanced automation.

Use Cases
  • Home automation hobbyists building custom environmental monitors
  • Renters needing non-invasive, removable smart sensors
  • Educators teaching sensor fusion and ESPHome configuration in workshops

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

PiKVM Adds WebRTC Streaming for Sub-50ms Remote Server Access 🔗

Latest update enables low-latency browser-based KVM without plugins on Raspberry Pi 5

pikvm/pikvm · Unknown · 10.3k stars Est. 2019

The pikvm/pikvm project released a WebRTC video stream option in its August update, cutting remote console latency to under 50ms on Raspberry Pi 5 hardware. This replaces older MJPEG and HTTP H.

264 modes with direct browser peer-to-peer transmission, eliminating plugin dependencies. Users now access BIOS-level controls, virtual media, and power management through any modern web UI via WebRTC, H.264-over-HTTP, or fallback MJPEG. The feature builds on PiKVM’s existing IP-KVM foundation, supporting mass-storage emulation, ATX power control, and IPMI/Redfish integration. The catch: WebRTC requires a recent Raspberry Pi OS kernel and may strain older Pi 3/Zero2W boards under sustained 1080p use.

Previously in The Times “covered” — Aug 16

Use Cases
  • Datacenter techs reboot frozen headless servers via browser
  • IT teams reinstall OS remotely using virtual USB flash drive
  • Home lab users manage Pi clusters without dedicated monitors

Source: pikvm/pikvm — based on the project README.

Quick Hits

mighty MIGHTY enables real-time, multi-agent trajectory planning for UAVs and ground robots using Hermite splines, offering smooth, collision-free motion in ROS 2 environments. 367
firmware This ESP32 firmware transforms the chip into a low-power, predatory sensor node for stealthy environmental monitoring and adaptive threat detection. 6.5k
PMSG PMSG turns the Seeed Studio XIAO into a wearable face computer, enabling real-time facial processing and interaction via embedded sensors and lightweight compute. 43
OSHW-reCamera-Series The reCamera series provides open-source, AI-enabled camera modules with modular design for edge vision applications like object detection and gesture recognition. 483
vdbrink.github.io This documentation offers practical, builder-focused tips for integrating Node-RED and Home Assistant, simplifying home automation setup with real-world wiring and automation patterns. 48

Stride 4.3 upgrades to .NET 10 and C# 14 for modern game development 🔗

Release adds mouse wheel support, improves build tooling, and updates Visual Studio 2026 paths

stride3d/stride · C# · 7.8k stars Est. 2018 · Latest: releases/4.3.0.2507

Stride 4.3 brings the open-source C# game engine into alignment with .

NET 10 and C# 14, enabling developers to use the latest language features and runtime improvements. The update includes a refactor to use CollectionsMarshal.SetCount for efficient list resizing and moves BEPU physics asset compilation into Stride.Assets for better modularity.

The release adds practical enhancements: mouse wheel delta is now captured as a virtual button input, improving control flexibility in games and VR apps. Documentation fixes address typos and update MSBuild paths for Visual Studio 2026, while disk space requirements were adjusted from 14 GB to 19 GB to reflect current tooling needs.

Stride remains a multi-platform engine supporting Direct3D and Vulkan, with its visual editor, Game Studio, and CLI tools (stride new, stride sdk, stride upgrade) allowing code-first or visual workflows. Templates are also available via dotnet new for genre starters, demos, and tutorials.

The catch: Despite active development, 674 open issues indicate ongoing stability and feature maturity challenges, particularly around graphics regressions and build consistency across platforms.

Previously in The Times “covered” — Aug 16

Use Cases
  • Indie developers building cross-platform 3D games with C#
  • VR creators leveraging realistic rendering and input handling
  • Teams using CLI tools for automated project setup and version control

Source: stride3d/stride — based on the README and release notes.

More Stories

JoltPhysics Adds GPU Compute for Hair Simulation and Friction Model 🔗

v5.6.0 introduces DX12/Vulkan/Metal compute shaders and a strand-based hair system

jrouwe/JoltPhysics · C++ · 11.4k stars Est. 2021

JoltPhysics v5.6.

0 adds GPU compute interfaces for DX12, Vulkan, and Metal to offload physics work, including a new strand-based hair simulation using Cosserat rods. The update also implements a revised friction model that averages contact points, reducing memory use by 40% and improving speed by 15% in Pyramid tests. These features target game and VR developers needing efficient multi-threaded physics. The library remains C++-only and requires Vulkan SDK for macOS builds.
The catch: GPU compute features are optional and require additional tooling, limiting out-of-the-box use on some platforms.

Use Cases
  • Game studios implementing realistic hair and fur in VR titles
  • Physics engineers optimizing collision queries across threads
  • Developers building deterministic multiplayer simulations with replay support

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

Pixelorama adds SVG export and corner pin effects in v1.2.1 🔗

Built on Godot 4.7.1, the update improves animation timeline usability and shader support.

Orama-Interactive/Pixelorama · GDScript · 10.2k stars Est. 2019

Pixelorama v1.2.

1 introduces SVG image export, converting opaque pixels to <rect> elements, and a new Corner Pin layer effect for perspective adjustments. Right-clicking keyframes now opens a deletion menu, and texture blit shaders can be loaded customly. Timeline enhancements include track folding, frame snapping, and larger current-frame numbers. The tool remains GDScript-based, requiring Godot 4.7.1 to run. The catch: SVG export is basic, merging same-color pixels remains unimplemented, and early access builds via GitHub Actions are explicitly labeled unstable.

Use Cases
  • Game artists creating spritesheets with onion skinning
  • Animators syncing pixel art to audio via timeline
  • Designers exporting pixel art as scalable SVG for web use

Source: Orama-Interactive/Pixelorama — based on the README and release notes.

Quick Hits

Yugens-Terrain-Authoring-Toolkit Enables precise terrain sculpting in Godot using Marching Squares algorithm for customizable, performant 2D landscapes. 602
Greater-Flavor-Mod Deepens historical immersion with expanded provinces, dynamic flavor events, and refined accuracy for grand strategy gameplay. 239
GD-Agentic-Skills Equips Godot 4.7+ AI agents with audited, typed skills and persona orchestration for complex, production-ready behavior systems. 574
GDevelop Provides a free, open-source engine for building and deploying 2D/3D/multiplayer games across platforms with no barriers to entry. 25.9k
Starter-Kit-Racing Offers a ready-to-use Godot-based racing template with vehicle physics, track systems, and input handling for rapid prototyping. 391
cocos-engine Delivers a high-performance, cross-platform engine via Cocos Creator to build and publish 2D/3D games and instant web experiences. 9.8k
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