Design
Preset
Background
Text
Font
Size
Width
Account Pricing

The Git Times

AI Models
Claude Opus 5 $25/M GPT-5.6 Luna $0.60/M Gemini 3.1 Pro Preview $12/M Grok 4.6 $6/M DeepSeek V4 Pro 0813 $0.87/M Qwen3.8 2.4T A95B $6/M Kimi K3 $15/M
Full Markets →

KADATH Evolves Agent Populations to Solve Goals Through Competitive Selection 🔗

Each generation breeds improved autonomous agents by mutating prompts, code, and tools under fixed benchmarks

i3T4AN/KADATH · Python · ▲ 47 in 1d 5d old

KADATH implements an evolutionary runtime that treats agent frameworks as genomes subject to selection pressure. Given a goal, it spawns a population of agents—each with editable system prompts, Python logic, tools, and dependencies—and grades them against a measurable benchmark.

Top performers inform the next generation through mutation and reproduction, while weaker agents are culled. Over epochs, the system converges on high-performing agent designs without relying on manual prompt engineering. In a demonstrated ten-epoch run, the leading cohort’s best score jumped from 18 to 91, with the top-five median rising from 8 to 77, indicating broad population improvement rather than individual outlier success. The kernel preserves evaluation integrity by isolating fitness scoring, scheduling, and lineage from the evolvable agent genome, preventing self-modification of objectives. Organisms remain read-only during epochs, only adapting in post-grade mutation phases. Built in Python and leveraging smolagents for core agentics, KADATH provides a reproducible, containerized pipeline with Git-based lineage tracking and recovery. It shifts focus from optimizing single agents to cultivating adaptive populations that collectively solve complex, open-ended goals through competition.
The catch: As an early-stage project with no open issues and minimal external adoption, its real-world efficacy beyond toy benchmarks remains unproven at scale or in noisy, dynamic environments.

Why this leads today i3T4AN/KADATH provides a new evolutionary runtime enabling autonomous agent swarms to self-improve across reproducible epochs, offering developers a foundational tool for optimizing agent systems in practice.

Use Cases
  • Researchers evolving LLM agents for scientific hypothesis generation
  • Teams automating prompt and toolchain optimization for repetitive tasks
  • Developers benchmarking agent robustness against shifting success criteria

Source: i3T4AN/KADATH — based on the project README.

More on the Front Page

DSH Web UI Plugins Unify Agent Control, Git, and Remote Ops in One Interface 🔗

Version 0.1.10 adds live token stats, mobile pairing, and eight interchangeable skins for DeepSeek Harness

zhu1090093659/dsh-web-ui · TypeScript · 1.3k stars 2d old

The zhu1090093659/dsh-web-ui project consolidates task boards, Git visualization, remote execution, and mobile control into a single TypeScript-based interface for DeepSeek Harness (DSH). Users install individual plugins or an aggregate package to gain a five-column task board (待规划 to 已失败) where clicking “执行” triggers real DSH agent sessions, with status auto-updating on completion.

Tasks support cron-based scheduling—e.g., daily DSH upgrades or weekly report generation—running unattended.

A branch selector above the input box enables Git history navigation, while the integrated Git graph visualizes branch lanes and commits for rapid change tracking in large repos. During agent sessions, a dynamic right-side panel appears: a file tree for workspace browsing, a multi-tab preview pane (supporting Markdown, HTML, code, diff, CSV, PDF, Office, images, and text), and a real-time SCM panel for staging, unstashing, and discarding changes. Panel widths are draggable, with double-click reset and persistent state per project. All eight skins adapt to this panel, ensuring thematic consistency.

A persistent whale娘 pet follows agent state (思考, 等待, 工作, 庆祝), responds to head pats and fish snack feeding to increase intimacy, and evolves from “幼鲸” to “深海羁绊.” Custom names, free dragging, and togglable visibility are supported. Real-time token stats—TPS, LLM latency, context usage, cache hit rate, and I/O counts—display beneath the input box.

Mobile pairing via sidebar phone icon enables remote control after QR code or link pairing; sessions, messages, model switching, and permission presets sync with desktop. The plugin uses SSE for real-time mobile updates but falls back to polling if tunnels like Cloudflare Quick Tunnel or Tailscale Serve block SSE, risking delayed message delivery. Remote operations include SSH host management (key/password auth, ~/.ssh/config import), xterm.js terminals, SFTP transfers, local-only port forwarding, and multi-host cluster execution via aliases, tags, or environments. Agents can execute remote commands via natural language (“连一下 xxx 看看状态”) when host config is shared.

Plugin toggles and parameters live in “设置 > 插件配置,” with instant application. The skin center offers eight preview-then-apply themes: Luna Classic (Windows XP-inspired gradients and Bliss backdrop), Minecraft (pixel skybox, stone buttons, wood sign input), Whale Illustration (indigo palette under translucent panels), and Deep Sea Whale Goddess (atmospheric goddess-and-whale artwork with cobalt cooling tones). A stock ticker skin shows real-time A股/港股/美指数/加密/外汇 data, integrating with dsh-fun-ticker for custom watchlists and dsh-longbridge for broker snapshots.

The catch: Mobile real-time messaging depends on SSE-capable tunnels; common options like Cloudflare Quick Tunnel or Tailscale Serve force polling, introducing seconds of delay for new messages unless users configure named tunnels or custom TCP forwarding— a non-obvious setup hurdle for seamless remote use.

Use Cases
  • Developers schedule automated agent tasks via cron for nightly builds
  • Teams use mobile pairing to monitor and control DSH sessions remotely
  • Operators manage multi-host SSH workflows through a unified web terminal

Source: zhu1090093659/dsh-web-ui — based on the README and release notes.

Unsloth Desktop Enables Local LLM Training With Reduced VRAM Usage 🔗

Supports Qwen3.8, Gemma 4, and DeepSeek-V4 across CPU, NVIDIA, AMD, and multi-GPU setups.

unslothai/unsloth · Python · 71.2k stars Est. 2023

Unsloth Desktop lets builders run and train LLMs and diffusion models locally, including Qwen3.8, Gemma 4, and DeepSeek-V4, with support for fine-tuning via LoRA and QLoRA.

The app delivers up to 2× faster training and 50% more accurate tool calling through self-healing code execution in sandboxed environments. It exports models to GGUF and NVFP4 formats and provides an OpenAI-compatible API for agent integration. Remote access is available via secure Cloudflare HTTPS.
The catch: Despite broad hardware support, performance gains depend heavily on GPU compatibility, and some advanced features remain in beta with limited documentation.

Use Cases
  • Developers fine-tuning LLMs locally to avoid cloud costs
  • Researchers running private RAG with web search and MCP tools
  • Teams deploying agents via OpenAI-compatible API on-premise

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

Awesome-DSH-Plugin Catalogs Community Extensions for DeepSeek Harness 🔗

Lists 211 installable plugins ranging from UI themes to generative components via dsh plugin add

awesome-dsh-plugin/awesome-dsh-plugin · HTML · 464 stars 0d old

The awesome-dsh-plugin repository curates 211 community-developed plugins for DeepSeek Harness (dsh), an open-source agent harness where every component—from models to UI—is a plugin. Builders can extend or replace core agent functionality using dsh plugin add, with plugins declared via `dsh.

bundlemanifests. Categories include UI enhancements likedsh-tianshu-tui(terminal interface),dsh-visualize(generative HTML cards), anddsh-focus-chat(minimal output view), alongside tools such asdsh-at-filefor Codex-style file mentions anddsh-genuifor interactive components like charts and 3D scenes. The list supports discovery viadsh-find-plugin`, enabling agents to locate and install plugins directly. The catch: With 8 open issues and no recent code updates beyond curation, the list’s maintenance and plugin vetting rigor remain unclear for production adoption.

Use Cases
  • Developers customize dsh UI with terminal or sidebar plugins
  • Engineers add file-attaching or annotation features to agent workflows
  • Teams integrate generative UI components like forms and 3D views into chats

Source: awesome-dsh-plugin/awesome-dsh-plugin — based on the project README.

Book-to-skill turns technical PDFs into agent skills for on-demand reference 🔗

Generates structured frameworks and decision rules from books, cutting token use 24–51× vs full context dumps

Leutenegger/book-to-skill · Python · 1k stars 0d old

Leutenegger/book-to-skill converts any technical book PDF or document folder into a unified agent skill. It distills content into structured frameworks, decision rules, anti-patterns, and per-chapter files—not summaries—so agents like Claude Code, GitHub Copilot CLI, or Amp load only relevant sections on demand.

Install via pip install -e ., then run book-to-skill <path-to-book.pdf> to generate the skill. Once installed, agents access precise answers using /your-book-slug <query>, eliminating hallucination and manual PDF digging. The tool works with any host supporting the open Agent Skills standard.
The catch: Generated skills require manual installation per host (Claude, Copilot, Amp) and lack automated cross-platform sync.

Use Cases
  • Developers reference API specs from O'Reilly books during debugging
  • Engineers recall architecture patterns from system design texts while coding
  • Students study textbook chapters via agent queries without re-reading full PDFs

Source: Leutenegger/book-to-skill — based on the project README.

Skitter-Creek-Bath-Salts rewires DRAM to unlock CPU secrets 🔗

Demonstrates physical memory scrambling exposes PSP, SMM, and microcode on AMD 16h

xoreaxeaxeax/skitter-creek-bath-salts · C · 1.2k stars 0d old

Skitter-creek-bath-salts manipulates DRAM controller translation registers to redirect physical memory accesses, bypassing hardware-enforced protections. By altering the final address translation layer, it makes protected regions like the Platform Security Processor and System Management Mode visible and writable from software.

The exploit requires only a single bit-flip in the memory controller to break address translation invariants, turning &x == &x into a conditional falsehood. Tested on AMD Family 16h CPUs, it reveals how memory isolation relies on undocumented, unlockable hardware states. The technique suggests broader applicability across architectures, though current implementation is AMD-specific. The catch: No public exploits exist for post-16h AMD or ARM/RISC-V, leaving real-world impact and mitigations uncertain.

Use Cases
  • Security researchers analyzing firmware attack surfaces
  • Hardware vendors validating memory controller protections
  • OS developers testing kernel isolation boundaries

Source: xoreaxeaxeax/skitter-creek-bath-salts — based on the project README.

shadcn-labs/pdfcn Simplifies PDF Generation with React Components 🔗

Built on Takumi and Forme, it offers zero-config, one-command setup for developers

shadcn-labs/pdfcn · TypeScript · 408 stars 2d old

shadcn-labs/pdfcn provides ready-to-use, customizable PDF components for React, enabling developers to generate PDFs with minimal setup. Leveraging Takumi and Forme, it supports both rendering engines and includes live previews in documentation.

The library is fully compatible with shadcn/ui, using the same registry format and CLI workflow. Components cover tables, forms, charts, invoices, and reports, allowing composable document building. The catch: With only one open issue and recent activity, long-term maintenance and scalability under heavy workloads remain unproven.

Use Cases
  • Developers generate invoices using reusable React components
  • Teams create reports with zero-config PDF generation
  • Builders embed customizable charts in exportable documents

Source: shadcn-labs/pdfcn — based on the project README.

Cordiverse/Paper Unifies Revertible Effects and Reactive Coeffects for Dynamic Composition 🔗

Introduces Cordis meta-framework to manage spatiotemporal composability in plugin systems and agent harnesses

cordiverse/paper · Unknown · 1k stars 0d old

Cordiverse/paper presents a programming paradigm for spatiotemporal composability by formalizing revertible effects and reactive coeffects at runtime. It unifies effect and coeffect contexts into a single context type, enabling components to declare dependencies and fully revert side effects upon removal.

The paradigm is implemented in Cordis, a meta-framework with a core library for effect tracking, coeffect resolution, and a declarative component loader supporting hot module replacement and configuration reconciliation. The project is a preprint under active revision, with content subject to change.
The catch: The paradigm remains theoretical with no proven large-scale adoption or performance benchmarks in production systems.

Use Cases
  • Plugin systems requiring safe dynamic loading and unloading
  • Self-evolving agent harnesses managing inter-component dependencies
  • Runtime environments needing hot module replacement with state rollback

Source: cordiverse/paper — based on the project README.

AI Agents Reshape Open Source Workflows Through Modular Skill Systems 🔗

Developers are building composable, self-improving agents that integrate memory, tools, and reasoning across coding, research, and operations.

Trendai-agents
Leutenegger/book-to-skillSaladDay/pi-from-scratchccch1mneyyy/dsh-TUIayi-ai/nie-grassroots-logici3T4AN/KADATHaipoch/open-scienceunslothai/unslothpytorch/rllightdash/lightdashChachamaru127/claude-code-harnessPrimeIntellect-ai/prime-agentTencentCloud/TencentDB-Agent-Memorycloudflare/cloudflare-osgoogle/skillsayghri/i-have-adhdbojieli/ai-agent-bookifixai-ai/iFixAicorsairdev/corsairyc-software/qmtrycompai/crmhuangruiteng/loopxtt-a1i/archifyopenchamber/openchamberalibaba/open-code-reviewesengine/DeepSeek-Reasonixlangchain-ai/openwikiDevin-AXIS/iPolloWorkblader/humanizercan1357/oh-my-piget-bb/bbcalesthio/OpenMontagealirezarezvani/claude-skillsrocketride-org/rocketride-serveragentplugins/agent-plugins-specHKUDS/Vibe-Tradinglemony-ai/cascadeflowholaboss-ai/holaOSHKUDS/CLI-AnythingPan-Chera/Multi-Agent-CADvectorize-io/hindsightvercel-labs/agent-browserGoogleCloudPlatform/generative-aiGrigorij-Dudnik/RoboCrewjustrach/codedb

Open source is converging around AI agents as programmable workflows rather than monolithic assistants. Projects like book-to-skill transform technical PDFs into executable Claude Code skills, while prime-agent enables self-improving coding loops through reinforcement learning.

Memory systems are becoming standardized: TencentDB-Agent-Memory structures team knowledge into reusable assets (chat logs, skill graphs, code maps), and hindsight offers agent memory that learns from past interactions. Tool integration is accelerating via corsair’s agent integration layer and vercel-labs/agent-browser for Rust-based browser automation. Specialized agents emerge for niche domains: HKUDS/Vibe-Trading for finance, calesthio/OpenMontage for video production, and trycompai/crm for agent-native customer relationship management. Frameworks like i3T4AN/KADATH use evolutionary algorithms to breed and optimize agent populations over epochs, and holaboss-ai/holaOS provides a unified workspace with shared memory across 100+ integrations. Even skills target human factors: ayghri/i-have-adhd prevents agents from burying answers in verbose output, and blader/humanizer strips AI telltales from text. This shift reflects a move from prompting LLMs to engineering agent lifecycles—planning, tool use, memory, and iteration—as first-class, open-source components.
The catch: Much of this remains experimental; agent reliability, safety, and interoperability are unproven at scale, and the rapid proliferation of competing memory schemas, tool interfaces, and agent runtimes risks fragmentation before standards emerge.

Use Cases
  • Developers turn PDFs into interactive coding skills
  • Teams share structured memory across AI agents
  • Researchers automate reproducible workflows with local agents

Web frameworks are fracturing into hyper-specialized, domain-specific UI layers 🔗

Developers are composing purpose-built interfaces for AI tools, analytics, and secure terminals instead of monolithic stacks

Trendweb-frameworks
zhu1090093659/dsh-web-uiccch1mneyyy/dsh-TUIshadcn-labs/pdfcnSmall-tailqwq/dsh-deep-whaleAutomattic/wp-calypsoPostHog/posthogsaas-js/saas-uiemacs-elfeed/elfeedtrezor/trezor-suiteChromeDevTools/devtools-frontendbrightdata/clitalivia-group/taliviaopenchamber/openchamberh4ckf0r0day/obscuradnshe/DNSHE-FreeDomainsGoogleCloudPlatform/generative-aizhamao-robot/zhamao-frameworkmitmproxy/mitmproxybunkerity/bunkerwebntop/ntopngmaurosoria/dirsearchinfoslack/awesome-web-hackingjstrieb/github-statskarlseguin/http.zigbkaradzic/bgfxOrama-Interactive/Pixelorama

A clear pattern is emerging in open-source web frameworks: the rise of tightly scoped, single-purpose UI layers designed to augment specific developer tools or workflows. Rather than building broad application platforms, projects are focusing on polished, embeddable components that solve narrow but critical UX gaps.

For example, zhu1090093659/dsh-web-ui delivers a plugin-rich interface for the DeepSeek Harness, adding task boards, git graphs, and live token stats as modular skins. Similarly, ccch1mneyyy/dsh-TUI provides a full-screen terminal interface inspired by Claude Code, featuring pixel-accurate status lines, streaming thought visualization, and context-aware rollbacks—addressing a void in official CLI tooling. On the analytics side, talivia-group/talivia offers a self-hosted, revenue-focused alternative to traditional web analytics, integrating session replay and attribution directly into founder workflows. Meanwhile, shadcn-labs/pdfcn demonstrates how specialized UI can simplify document handling with zero-config PDF components built on Takumi and Forme. Even security tools are getting bespoke fronts: brightdata/cli enables structured web scraping from the terminal, while bunkerity/bunkerweb delivers a cloud-native WAF with a focused web interface for policy management. This shift reflects a maturation of the ecosystem: developers increasingly prefer composing lightweight, interoperable UIs over adopting heavyweight frameworks. The catch: While this fragmentation enables precision and rapid iteration, it risks creating siloed interfaces that lack consistency, increase cognitive load, and complicate long-term maintenance—especially when specialized UIs diverge from accessibility standards or shared design systems, leaving teams to glue together incompatible pieces.

Use Cases
  • Engineers monitoring AI agent token usage in real time
  • Analysts attributing revenue to specific user sessions
  • Security teams managing WAF policies via terminal UI

Open Source Data Infrastructure Shifts Toward Agentic, Local-First Workflows 🔗

Projects now embed AI agents and self-contained analytics to enable autonomous data pipelines without cloud dependency

Trenddata-infra
aipoch/open-sciencePostHog/posthoghi-manshu/chartylightdash/lightdashbrightdata/clitalivia-group/taliviaZhuLinsen/daily_stock_analysistt-a1i/archifymalisper/pgrustcoreyhaines31/marketingskillstsingyuai/growth-labrepowise-dev/repowisegoogle-ai-edge/mediapipeHumanSignal/label-studio

A clear pattern is emerging in open-source data infrastructure: the rise of agentic, local-first systems that combine AI-driven automation with reproducible, offline-capable workflows. Repos like aipoch/open-science exemplify this shift, offering a TypeScript-based workbench where scientific agents orchestrate model-agnostic research pipelines entirely locally, ensuring reproducibility without external APIs.

Similarly, lightdash/lightdash delivers agentic BI—analytics that generate insights at the speed of code, allowing developers to query and visualize data through programmable interfaces rather than brittle dashboards.

This trend extends beyond analytics into data acquisition and labeling. brightdata/cli enables terminal-based scraping and structured extraction of web data, turning unstructured sources into queryable inputs for downstream agents. Meanwhile, HumanSignal/label-studio provides a standardized, multi-type annotation framework that feeds clean, labeled data into model training loops—critical for agent reliability.

On the storage and processing front, malisper/pgrust rewrites Postgres in Rust for superior performance, signaling a broader move toward infrastructure that’s both faster and more embeddable in agent-driven systems. Complementing this, tsingyuai/growth-lab and talivia-group/talivia demonstrate end-to-end growth engines that autonomously fetch data, run experiments, and attribute revenue—all self-hosted and orchestrated via code.

Even visualization is being reimagined: hi-manshu/charty brings declarative, potion-free graphing to Jetpack Compose, while tt-a1i/archify generates verifiable, motion-rich architecture diagrams as self-contained HTML—ideal for agent-generated documentation.

The catch: While promising, this cluster remains fragmented—many projects lack interoperability standards, and the "agentic" label often overstates current capabilities, with most systems still requiring significant human-in-the-loop tuning for reliable autonomous operation in production environments.

Use Cases
  • Researchers run reproducible AI-driven experiments locally
  • Developers build self-hosted analytics pipelines with code
  • Teams extract and label web data for model training without SaaS tools

Deep Cuts

Grassroots China Logic Toolkit Empowers Agents With Governance Insight 🔗

Translates Nie Huihua’s fieldwork into executable skill modules for AI agents

ayi-ai/nie-grassroots-logic · HTML · ▲ 80 in 1d

The ayi-ai/nie-grassroots-logic project distills sociologist Nie Huihua’s research on China’s local governance into reusable agent skills. Built in HTML, it structures field-tested logic patterns—like issue escalation paths and resource brokerage—into modular prompts agents can invoke.

Developers integrating these skills gain nuanced, on-the-ground decision-making frameworks for policy simulation or community-facing bots. It’s not theoretical; each skill mirrors observed village-level adaptations to central directives. Use it to stress-test AI agents in complex socio-political environments where top-down rules meet informal practice. The catch: It’s early-stage, niche, and assumes familiarity with Chinese administrative realities, limiting broad adoption without contextual grounding.

Use Cases
  • Policy designers simulate local implementation challenges
  • Community agents navigate bureaucratic workflows authentically
  • Researchers test governance models against field-validated logic

Source: ayi-ai/nie-grassroots-logic — based on the project README.

Quick Hits

deepseek-harness DeepSeek Harness enables builders to extend and customize AI workflows through a modular plugin architecture, turning any component into an interchangeable, reusable unit for rapid prototyping and scalable integration. 79.5k
ai-trader-bot The AI Trader Bot combines on-chain smart contracts with off-chain automation to execute real-time arbitrage strategies, offering builders a secure, programmable framework for decentralized finance automation. 2.7k
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

Julia 1.12.6 sharpens HPC and ML tooling for scientific builders 🔗

Release adds GPU memory profiling and faster package precompilation for large-scale simulations

JuliaLang/julia · Julia · ▲ 2 in 1d Est. 2011 · Latest: v1.12.6

Julia 1.12.

6 refines performance for technical workloads, targeting builders in HPC and machine learning. The update introduces CUDA.@profile for granular GPU memory tracing, helping debug kernels in large-scale simulations. Package precompilation now caches dependencies more aggressively, cutting startup latency in complex environments by up to 40% according to benchmarks in the release notes. These changes address long-standing friction points in iterative scientific workflows where REPL responsiveness and toolchain speed directly impact productivity. The release also stabilizes multithreading defaults for linear algebra, reducing tuning overhead for parallel matrix operations common in ML pipelines. Builders benefit from tighter integration with juliaup, which now handles version isolation more reliably across Linux and macOS, simplifying CI/CD matrix testing. Documentation saw targeted updates to the MPI and Distributed computing guides, reflecting community feedback on cluster deployment patterns. Despite these gains, the project continues to manage 4,599 open issues, indicating sustained pressure on core maintainers to balance feature velocity with bug triage.
The catch: Julia’s rapid release cadence can strain package compatibility, requiring builders to pin versions carefully in production systems to avoid breakage from minor updates.

Previously in The Times “covered” — Aug 7

Use Cases
  • Accelerate climate modeling with GPU-aware differential equations
  • Train ML models using native multithreaded linear algebra
  • Deploy scientific APIs with deterministic startup latency via juliaup

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

More Stories

AirLLM Enables 2.8T Kimi K3 Model on Single GPU 🔗

Streams sparse MoE experts, cutting VRAM to 3.72GB on RTX 6000 Ada

lyogavin/airllm · Jupyter Notebook · ▲ 155 in 1d Est. 2023

AirLLM runs massive language models on modest hardware by loading only the experts a token actually uses, avoiding full-layer inference. It supports models like DeepSeek-V3 (671B) on ~12GB and Llama 3.

1 405B on 8GB, using techniques such as MXFP4 weight compression and PCIe-efficient data transfer. The project requires specific dependencies: flash-attn, a CUDA 12 PyTorch build, and transformers 4.56.x. Despite recent activity, including Kimi K3 support in v3.1.0, the project maintains 136 open issues, indicating ongoing stability and compatibility challenges.
The catch: Heavy reliance on niche, version-pinned dependencies complicates deployment in evolving AI stacks.

Use Cases
  • Researchers testing trillion-parameter models on consumer GPUs
  • Developers deploying large LLMs in low-VRAM edge environments
  • Teams fine-tuning MoE models without quantization or distillation

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

GoogleCloudPlatform/generative-ai Adds Gemini 3.7 Flash Notebooks 🔗

Repository expands with Agent Development Kit samples and Vertex AI agent templates

GoogleCloudPlatform/generative-ai · Jupyter Notebook · ▲ 3 in 1d Est. 2023

The GoogleCloudPlatform/generative-ai repository updated its notebooks to include Gemini 3.7 Flash examples and production-ready Agent Starter Pack templates.

It provides ready-to-use agents built with the Agent Development Kit for conversational bots and multi-agent workflows. Resources cover RAG grounding, search, vision, and audio solutions using Chirp and Imagen. Setup guides support Colab and Workbench environments. The catch: Open issues exceed 80, indicating unresolved bugs or feature gaps despite recent activity.

Use Cases
  • Developers build conversational agents using Gemini 3.7 Flash
  • Teams deploy retrieval-augmented generation pipelines on Vertex AI
  • Enterprises create multimodal search engines with Agent Search tools

Source: GoogleCloudPlatform/generative-ai — based on the project README.

Label Studio Adds Vector Annotation Tags for Precise Image Labeling 🔗

Latest release includes interactive task source view and enhanced Data Manager filters

HumanSignal/label-studio · TypeScript · ▲ 14 in 1d Est. 2019

HumanSignal's Label Studio 1.23.

0 introduces Vector and VectorLabels tags for annotating images with polylines, polygons, and skeletons. The update also adds an interactive view for task source JSON, enabling filtering, searching, and selective copying, plus improved Data Manager filters with copy/paste and JSON export capabilities. Built in TypeScript, the tool supports labeling audio, text, images, video, and time series data with export to ML-ready formats.
The catch: Despite active development, 922 open issues suggest ongoing stability and usability challenges for production ML pipelines at scale.

Use Cases
  • ML engineers labeling medical images for tumor detection
  • Data teams annotating customer service transcripts for intent classification
  • Researchers preparing multi-modal datasets for video action recognition models

Source: HumanSignal/label-studio — based on the README and release notes.

Quick Hits

openai-cookbook Provides practical Jupyter notebooks to help developers master OpenAI API integration with real-world examples and step-by-step guidance. 75.2k
google-research Shares cutting-edge Google Research code and notebooks showcasing advances in ML, AI, and systems for builders to explore and adapt. 38.5k
pytorch Offers a powerful, GPU-accelerated Python framework for building dynamic neural networks with flexible tensor computation and autograd. 102.4k
gemini-cli Brings Gemini’s AI capabilities into your terminal as an open-source agent for coding, querying, and automating tasks via natural language. 106.5k
firecrawl Enables scalable web interaction — search, scrape, and automate — through a unified TypeScript context API built for developers. 167.1k

Autoware Universe Enhances Planner Safety With New Boundary Checks 🔗

Latest release adds drivable area validation and trajectory filtering to prevent lane departures

autowarefoundation/autoware_universe · C++ · 1.7k stars Est. 2021 · Latest: 0.52.0

Autoware Universe’s 0.52.

0 release strengthens autonomous vehicle safety through two key planner updates. The behavior_path_side_shift package now includes drivable area checks to prevent unintended lane departures, while trajectory_validator gains an uncrossable boundary departure filter. These changes build on prior work like the boundary_departure checker and reflect a broader push to harden route planning against edge-case failures. Other updates include neural-network-based planner timing metrics, VTL support for reverse maneuvers, and improved sensor integration via the CARLA interface. The project continues to refactor legacy interfaces, removing deprecated planning components and updating message dependencies to autoware_msgs 1.13.0.
Built on ROS 2 and written in C++, Autoware Universe remains a critical extension layer for the Autoware ecosystem, enabling advanced perception, planning, and control workflows for self-driving research and deployment.
The catch: Despite active development, 490 open issues suggest ongoing instability in newer features like the diffusion planner and multi-object tracker, which may require careful validation before production use.

Previously in The Times “covered” — Aug 10

Use Cases
  • Researchers testing boundary-aware path planning in simulation
  • Developers integrating reverse maneuver support into vehicle stacks
  • Teams validating trajectory safety with automated departure filters

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

More Stories

MAVROS 2.15.0 fixes race condition in parameter plugin for safer drone control 🔗

ROS 2 bridge now supports MAVLink v2 signing and refactored mavconn for improved security

mavlink/mavros · C++ · 1.2k stars Est. 2013

The latest MAVROS release patches a critical race condition in the param plugin that previously caused "Promise already satisfied" crashes during concurrent parameter updates. It adds MAVLink v2 signing support via libmavconn, enhancing authentication for secure drone-ground station communication.

Developers benefit from refactored mavconn internals and updated dependencies, including Sphinx and urllib3, improving build stability and documentation. Tested against PX4 and ArduPilot autopilots, MAVROS continues to translate MAVLink telemetry and commands into standard ROS 2 topics and services, enabling familiar robotics programming for UAVs. The bridge remains extensible via its plugin architecture, allowing custom MAVLink message handling without core modifications. The catch: Open issues exceed 400, indicating ongoing maintenance challenges despite recent fixes.

Previously in The Times “covered” — Aug 8

Use Cases
  • Drone developers programming UAVs using ROS 2 instead of raw MAVLink
  • Ground stations proxying commands to flight controllers over UDP/TCP links
  • Simulation workflows testing PX4/ArduPilot with ROS 2 navigation stacks

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

Webots ROS2 adds Husarion and Crazyflie robot support 🔗

Latest release targets Webots R2025a with two new robot-specific packages

cyberbotics/webots_ros2 · C · 542 stars Est. 2019

The cyberbotics/webots_ros2 project released version 2025.0.

0, adding webots_ros2_husarion and webots_ros2_crazyflie packages for simulating those robots in Webots with ROS2 integration. It enables motion planning, navigation, and SLAM simulations using ROS2 messages, services, and actions. Backed by ROSIN and OpenDR funding, the project maintains active development with a commit just one day ago. Despite 51 open issues and nearly seven years of existence, it remains a key bridge between Webots and ROS2 ecosystems.
The catch: Heavy reliance on specific Webots versions may create lag when new simulator releases arrive, forcing users to wait for updates.

Use Cases
  • Robotics researchers simulate Husarion robots in Webots using ROS2
  • Developers test Crazyflie drone navigation and control algorithms
  • Engineers validate SLAM and motion planning in virtual environments

Source: cyberbotics/webots_ros2 — based on the README and release notes.

Quick Hits

spatialmath-python Enables precise 2D/3D spatial math for robotics — manipulate poses, rotations, and transforms in Python with intuitive, production-ready tools. 639
RoboCrew Turn any robot into an autonomous LLM-powered agent using CrewAI/Autogen-like simplicity — no low-level coding required. 127
IsaacLab Accelerate robot learning with a unified, NVIDIA Isaac Sim-native framework for simulation, training, and deployment at scale. 7.9k
drake Design, simulate, and verify complex robotic systems with rigorous model-based methods — from kinematics to control — in C++. 4.2k
URDF-Studio Build, visualize, and export URDF robots in-browser with AI-assisted modeling, hardware integration, and seamless MuJoCo export. 456
blockly Otto Blockly; a fully integrated graphical programming for any type of Arduino projects, including robots, ready to install in your computer, it works offline and also online 136

OpenAI's Codex Security CLI brings AI-powered vulnerability scanning to developer workflows 🔗

New release adds configurable deep-scan time limits and fixes credential handling for CI pipelines

openai/codex-security · TypeScript · ▲ 70 in 1d 1mo old · Latest: npm-v0.1.11

OpenAI's codex-security project offers a TypeScript SDK and CLI tool that uses AI models to detect, validate, and fix security vulnerabilities in codebases. It integrates directly into development workflows, allowing developers to scan for issues locally or in CI without leaving their terminal.

The tool requires Node.js 22.13.0+, Python 3.10+, and access to Codex Security via API key or ChatGPT sign-in, with credentials managed securely and never stored in system keyrings.

The latest release, npm-v0.1.11, introduces configurable time limits for deep scans, letting users set --max-time-hours to any value up to 96 hours, with partial results preserved if the limit is reached. It also resolves several critical bugs, including POSIX path handling on Windows, credential forwarding for external providers like Amazon Bedrock, and race conditions in worker termination during scans. Authentication logic now cleanly separates interactive and noninteractive modes, letting CI systems rely on API keys while local users can opt for ChatGPT login.

For teams adopting DevSecOps, the tool supports scanning repositories, comparing scan histories, and listing unconfirmed findings via findings list [repository]. Findings are preserved across scans, and users can redirect state storage using CODEX_SECURITY_STATE_DIR if the default directory is unwritable.

The catch: The tool remains in early release (v0.1.11), with 141 open issues and a narrow scope focused on Node.js and Python ecosystems, raising questions about its readiness for large-scale, multi-language enterprise use.

Previously in The Times “covered” — Aug 11

Use Cases
  • Developers scanning Node.js apps for vulnerabilities before commit
  • Security teams automating vulnerability checks in CI/CD pipelines
  • DevOps engineers validating fixes across scan versions using history compare

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

More Stories

OWASP MASTG v2.0.0 launches with machine-readable security testing framework 🔗

Guide now offers automated demos and full cross-linking across MASVS, MASWE, and test techniques

OWASP/mastg · Python · 13.1k stars Est. 2016

The OWASP Mobile Application Security Testing Guide (MASTG) released v2.0.

0 after a multi-year refactor to modularize its content into machine-readable components. The update introduces structured metadata, automated demos, and full cross-linking between MASVS controls, MASWE weaknesses, testing techniques, tools, and best practices. Built in Python, MASTG supports static and dynamic analysis for Android and iOS apps, enabling testers to verify mobile security weaknesses systematically. Contributions from MAS Advocates like NowSecure, Guardsquare, and Vulnit helped drive the release, which spans work from January to June 2026. The catch: The guide’s increasing complexity and reliance on structured metadata may pose a steep learning curve for newcomers focused on quick, ad-hoc testing rather than standardized, repeatable assessments.

Use Cases
  • Security engineers reverse-engineering Android binaries
  • Pentesters validating iOS app cryptography implementations
  • Teams aligning mobile app tests with OWASP MASVS compliance requirements

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

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

Maintainers reject over 75% of bogus security claims to focus on real fixes

caddyserver/caddy · Go · ▲ 36 in 1d Est. 2015

Caddy’s latest release addresses path normalization on Windows, query placeholder safety, and HTML stripping reliability. Contributors from FrankenPHP and the community helped patch header collision risks and reverse proxy body handling.

The project now blocks spammy AI-generated vulnerability reports that waste maintainer time. Despite its extensible, dependency-free design and default HTTPS, Caddy requires manual setup for low-port binding on Linux via setcap.
The catch: Production use demands operational awareness of privilege escalation for port 80/443 access, a step often overlooked in quick-start guides.

Previously in The Times “covered” — Aug 8

Use Cases
  • Dev teams deploying zero-config HTTPS sites
  • Enterprises managing internal PKI with local CA
  • Developers extending servers via Go plugins without bloat

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

Quick Hits

dirsearch Dirsearch efficiently scans web directories and files to uncover hidden endpoints and misconfigurations during penetration tests. 14.6k
bunkerweb BunkerWeb delivers a cloud-native, open-source WAF with real-time threat protection and easy deployment for modern web apps. 10.8k
subfinder Subfinder rapidly discovers subdomains using passive sources, helping builders map attack surfaces without noisy active probing. 14.2k
ntopng ntopng provides real-time, web-based network traffic visibility with deep packet analysis for cybersecurity monitoring and anomaly detection. 8.1k
awesome-web-hacking Awesome-Web-Hacking curates a comprehensive, organized list of tools and resources for web application security testing and learning. 7.2k
mitmproxy Mitmproxy enables interactive interception, inspection, and modification of HTTP/HTTPS traffic for debugging, testing, and security assessments. 44.7k

PHP 8.5.9 Patches Critical Security Flaws in Core Interpreter 🔗

Release addresses vulnerabilities while maintaining backward compatibility for web applications

php/php-src · C · ▲ 2 in 1d Est. 2011 · Latest: php-8.5.9

The PHP development team released PHP 8.5.

9 as a security-focused update to the widely deployed interpreter. This version patches multiple vulnerabilities identified in the core engine, including memory safety issues and input validation flaws that could allow remote code execution. All PHP 8.5 users are urged to upgrade immediately via source downloads or platform-specific binaries from php.net/downloads.php.

Built primarily in C, the interpreter requires autoconf, bison, and re2c for source compilation, with additional dependencies like libxml2 and libsqlite3 for default builds. The release notes confirm no functional changes beyond security fixes, preserving the language’s established behavior for web scripting, CLI tools, and embedded systems. Test suites run in parallel by default, leveraging up to 10 logical processors to validate stability post-compilation.

Despite its maturity — over 15 years of active development — the project carries significant technical debt. The codebase shows signs of aging, with 1,974 open issues indicating persistent challenges in modernizing internals without breaking vast ecosystem compatibility.

The catch: The interpreter’s C-based architecture and commitment to backward compatibility limit adoption of radical performance or safety improvements seen in newer runtimes, creating a tension between stability and innovation.

Use Cases
  • Deploy dynamic web applications on LAMP stacks
  • Build command-line tools for system automation
  • Maintain legacy enterprise PHP applications requiring security updates

Source: php/php-src — based on the README and release notes.

More Stories

Deno 2.9.5 adds unscoped package aliasing and experimental QuickJS backend 🔗

Release includes workspace task filtering and fixes for Node.js compatibility streams

denoland/deno · Rust · ▲ 11 in 1d Est. 2018

Deno 2.9.

5 introduces the --unscoped flag to alias packages by their bare name and adds an experimental QuickJS backend alongside V8. Workspace tasks now support a --members flag to run only in selected members. Fixes resolve esbuild protocol deadlocks, respect file permissions during bundling, and improve Node.js stream backpressure handling. The runtime maintains secure defaults and TypeScript-first tooling for server and CLI apps.
The catch: Experimental QuickJS backend may lack full API compatibility and performance parity with V8, limiting production readiness.

Use Cases
  • Backend devs building secure HTTP servers
  • CLI tool authors needing single-binary distribution
  • Teams adopting TypeScript without transpilation setup

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

RAGFlow Integrates Snowball Stemmer for 16 Languages in Tokenization 🔗

Release v0.26.4 adds language-aware tokenization, fixes Docling math formula extraction

infiniflow/ragflow · Go · ▲ 463 in 1d Est. 2023

RAGFlow’s July 2026 release introduces a Snowball stemmer supporting 16 languages, integrating dataset language parameters across its tokenization pipeline and adding Dutch frontend support. The update resolves critical bugs including Docling parser dropping mathematical formulas and MCP server crashes from misaligned chat API responses.

Built in Go, the engine continues to unify RAG with agent capabilities for LLM context layers, now handling multilingual unstructured data with improved precision. The catch: Open issues exceed 1,800, indicating ongoing stability challenges despite rapid feature expansion.

Previously in The Times “covered” — Aug 8

Use Cases
  • Enterprises processing multilingual technical documentation
  • Developers building agents requiring accurate math formula retention
  • Teams deploying LLMs with synchronized data from Confluence and Notion

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

Starship prompt gains git am progress tracking in Rust-powered shell tool 🔗

v1.26.0 adds nix-shell depth and Pixi project name support for faster workflows

starship/starship · Rust · 59.4k stars Est. 2019

Starship, the Rust-based cross-shell prompt, now shows real-time git am operation status in v1.26.

0, helping developers track patch application progress without leaving the terminal. The update also introduces a nix-shell:level variable to display nesting depth and exposes PIXI_PROJECT_NAME for Pixi users. Timezone handling improved via jiff integration, and gcloud region detection now respects CLOUDSDK_COMPUTE_REGION. Despite active development with 1,031 open issues, the project maintains broad shell compatibility.
The catch: Configuration complexity increases with each new module, requiring users to manually enable features in starship.toml for optimal performance.

Use Cases
  • Developers tracking git am patch applications in real time
  • Nix users monitoring shell nesting depth during dev environment setup
  • Pixi users displaying project names directly in their shell prompt

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

Quick Hits

rustdesk Rustdesk offers a self-hosted, open-source remote desktop solution that lets you securely control machines anywhere — no subscription, no vendor lock-in. 120.4k
codex OpenAI Codex in Rust brings AI-powered code generation directly to your terminal, letting you write, edit, and debug code with natural language prompts. 105.9k
zed Zed is a blazing-fast, collaborative code editor built for real-time teamwork — combining Atom’s extensibility with Tree-sitter’s precision for instant, fluid coding. 88.6k
openwrt This OpenWrt mirror provides read-only access to the official firmware source for reference and PR submission, enabling builders to inspect and contribute to embedded Linux routing. 28k
cc-switch CC-Switch unifies Claude Code, Codex, OpenCode, and more into one cross-platform desktop assistant — streamlining AI agent workflows with a single, official interface. 127.2k

Glasgow Hardware Toolkit Gains Momentum After Maintainer Health Recovery 🔗

Project prepares for expanded team and Crowdsupply fulfillment as development accelerates

GlasgowEmbedded/glasgow · Python · 2.2k stars Est. 2018

GlasgowEmbedded/glasgow is a Python-driven hardware interface explorer that functions as a "Scots Army Knife for electronics," enabling developers to debug, prototype, and interact with FPGAs, sensors, and embedded systems through a unified USB-connected toolkit. After years of stalled progress due to founder Catherine @whitequark’s health challenges and external disruptions, recent updates signal renewed activity: the maintainer has relocated to the UK, secured healthcare, and is preparing to onboard additional contributors to the current core team of three.

The project’s last push occurred just two days ago (2026-08-14), marking the first significant commit in months and indicating a shift from maintenance mode to active development. While no formal release has been tagged yet, the commit history shows work on interface explorer stability and documentation improvements. Glasgow supports scripting via Python, allowing automation of complex hardware interactions — from JTAG debugging to SPI/I2C bus probing — all through a single hardware adapter.

Its value lies in consolidating multiple lab instruments into one open-source, extensible platform, reducing the need for proprietary tools. Builders use it to bring up custom boards, reverse-engineer unknown protocols, or validate FPGA designs without juggling multiple dongles and software suites.

The catch: Development remains constrained by maintainer bandwidth, and hardware fulfillment via Crowdsupply continues to lag behind software progress, creating uncertainty around immediate availability for new users despite advancing code.

Previously in The Times “covered” — Aug 12

Use Cases
  • Engineers debugging custom FPGA boards via JTAG and UART
  • Makers reverse-engineering sensor protocols on IoT devices
  • Developers automating hardware validation with Python scripts

Source: GlasgowEmbedded/glasgow — based on the project README.

More Stories

ElatoAI Enables Real-Time Voice AI on ESP32 Using Cloudflare Workers 🔗

Project integrates 100+ LLMs, TTS, and STT models via secure WebSockets for edge-deployed AI toys

akdeb/ElatoAI · TypeScript · 1.9k stars Est. 2025

ElatoAI lets builders run real-time speech-to-speech AI on Arduino ESP32 devices by connecting to Cloudflare Workers AI, Durable Objects, and external LLM APIs like OpenAI, Gemini, and Grok. The system uses a FastAPI server with Pipecat to orchestrate over 100 voice pipelines, supporting local models via MLX on ESP32 or cloud-based STT/TTS from Deepgram and MeloTTS.

Developers can customize agent personalities, voices, and conversation flow using server-side VAD and Opus audio compression. Control is available through a webapp, with demos showing integration across major AI voice platforms. The catch: Reliance on external LLM and voice API keys creates ongoing cost and dependency barriers for long-term, scalable deployment.

Use Cases
  • Hobbyists build conversational AI toys with ESP32 and voice APIs
  • Educators deploy low-latency voice agents for interactive learning kits
  • Prototypers test multi-model speech pipelines on edge hardware

Source: akdeb/ElatoAI — based on the project README.

Silhouette Card Maker automates custom TCG proxy creation 🔗

MTG plugin update adds Universes Beyond preference and language prioritization options

Alan-Cha/silhouette-card-maker · Python · 192 stars Est. 2025

Alan-Cha/silhouette-card-maker generates print-ready PDFs for Silhouette cutting machines to produce custom card game proxies. The Python tool supports Magic: The Gathering, Pokémon, and Yu-Gi-Oh via plugins that fetch card images and layout sheets for standard, poker, and bridge sizes.

Recent updates to the MTG plugin now include --prefer_ub and --ignore_ub flags to control Universes Beyond card versions, alongside improved language prioritization via --prefer_lang. Output requires a Silhouette Studio-compatible cutter and manual image sourcing for non-supported games.
The catch: Open issues number 47, indicating unresolved bugs or feature gaps despite active maintenance.

Use Cases
  • Hobbyists creating custom MTG proxies for home playtesting
  • Designers producing Pokémon card sheets with Silhouette Studio
  • Developers adapting plugins for niche TCG image acquisition workflows

Source: Alan-Cha/silhouette-card-maker — based on the README and release notes.

AIOsense Adds BME280 Support for ESPHome 2024.2.0 Compatibility 🔗

Fixes sensor integration while maintaining modular, no-SMD DIY smart sensor design

Schluggi/AIOsense · Unknown · 161 stars Est. 2022

Schluggi/AIOsense released esphome-v3.0.

1, patching BME280 sensor handling for ESPHome 2024.2.0. The update resolves compatibility issues without altering the core all-in-one sensor concept, which combines temperature, humidity, light, PIR, mmWave, and VOC sensing on a single ESP32-C3-based PCB. Designed for easy soldering and upgradeability, it avoids SMD parts and lets users swap modules as needs change. Power draw remains low at 0.11W idle without mmWave, scaling to 0.45W with full sensor load. The catch: 13 open issues suggest ongoing stability work, and voice assistant features remain "coming soon" after two years of development.

Previously in The Times “covered” — Aug 8

Use Cases
  • DIYers building custom environmental monitors
  • Home Assistant users avoiding commercial sensor costs
  • Hackers prototyping multi-sensor IoT nodes without SMD rework

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

Quick Hits

linorobot2 linorobot/linorobot2: Enables rapid prototyping of autonomous mobile robots with modular support for 2WD, 4WD, and Mecanum drive systems for real-world navigation tasks. 981
hackrf greatscottgadgets/hackrf: Provides an affordable, wideband software-defined radio platform for experimenting with wireless signals from 1 MHz to 6 GHz. 8k
openflight jewbetcha/openflight: Offers a Python library for parsing, manipulating, and generating OpenFlight 3D scene files used in simulation and visualization. 829
awesome-connected-things-sec V33RU/awesome-connected-things-sec: Curates essential security tools, frameworks, and best practices for securing IoT and connected devices across industries. 3.5k
LuaCAD ad-si/LuaCAD: Lets designers build parametric CAD models using Lua scripting, leveraging OpenSCAD’s power with a more flexible, code-first approach. 36

MPV Anime Build Refines UOSC and Track Selection in v4.8 Update 🔗

Adds dedicated Open File button and optimizes controller logic for smoother anime and live-action playback

Chinna95P/mpv-anime-build · GLSL · 174 stars 7mo old · Latest: v4.8

The MPV Anime Build project updated to v4.8 with focused improvements to its user interface and media handling.

A dedicated Open File button now appears in the UOSC controls bar, launching MPV’s native file browser directly within the player to select video and audio files without relying on system dialogs. This browser is configured to exclude image assets like posters and backdrops by setting image_types to empty, keeping the selection clean and media-only.

Under the hood, anime_profile_controller.lua and main.lua were optimized to reduce unnecessary processing and fix minor UI/logic bugs while preserving the existing auto-profile system for anime and live-action content. The track selector received refinements in track-selector.lua, improving reliability during manual and automatic track switching without altering the override workflow. These changes build on the project’s core features: AI upscaling via Anime4K, ArtCNN, and FSRCNNX; dynamic power management; universal HDR support; and intelligent audio processing. The build also maintains its Android variant (mpvEX / Aniyomi) with gesture controls and shader optimizations for mobile.

The catch: Despite active development, the project has three open issues and a narrow scope focused primarily on anime-oriented playback, which may limit appeal for users seeking a general-purpose MPV configuration without anime-specific upscaling or shaders.

Use Cases
  • Anime enthusiasts upscaling 720p content to 4K with AI shaders
  • Developers testing HDR passthrough in MPV on Linux desktops
  • Mobile users configuring gesture controls in mpvEX for anime playback

Source: Chinna95P/mpv-anime-build — based on the README and release notes.

More Stories

Bgfx powers cross-platform rendering with API-agnostic flexibility 🔗

Recent commit adds WebGPU support via Dawn Native, expanding modern graphics reach

bkaradzic/bgfx · C · ▲ 4 in 1d Est. 2012

Bgfx remains a C-based rendering library that abstracts Direct3D, Metal, Vulkan, OpenGL, and WebGPU backends, letting developers target desktop, mobile, console, and web with a single codebase. Its "Bring Your Own Engine" approach suits studios avoiding engine lock-in.

The project’s activity shows steady maintenance, with a commit just 0 days ago and ongoing work on bindings for Rust, Swift, and Zig. Despite broad platform coverage, the library requires manual integration of windowing and input via SDL, GLFW, or native frameworks, adding setup complexity. The catch: bgfx does not include a scene graph, physics, or asset pipeline, leaving core engine systems to the developer.

Use Cases
  • Indie studios building multiplatform 2D/3D games without engine royalties
  • Tool developers creating cross-platform graphics utilities like cmftStudio
  • Graphics researchers testing algorithms across APIs with minimal porting effort

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

Pixelorama 1.2 Adds Keyframe Animation and 3D Layer Editing 🔗

Update enables layer opacity animation and GLTF import while breaking pxo file compatibility

Orama-Interactive/Pixelorama · GDScript · 10.1k stars Est. 2019

Pixelorama’s v1.2 release introduces a keyframe-based timeline for animating layer opacity and effects with interpolation.

The update also adds autotiling support for tilemap layers and a rewritten 3D layer system that allows material property editing, GLTF scene import, and direct drawing on 3D objects using the pencil tool. Built on Godot 4.6.3, these features expand Pixelorama’s toolkit beyond 2D sprite work into more complex animation and 3D-integrated pixel art workflows. However, the 3D layer overhaul breaks backward compatibility: pxo files from v1.2 will load but display empty 3D layers in older versions, requiring users to stick to one version range for project continuity. The catch: The 3D layer changes invalidate existing pxo files’ 3D data, forcing a version lock for collaborative or long-term projects.

Use Cases
  • Indie devs creating animated sprites with synchronized audio
  • Artists designing tilesets using autotiling for seamless repetition
  • Pixel artists experimenting with 3D object drawing in 2D space

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

Quick Hits

Yugens-Terrain-Authoring-Toolkit Enables precise, real-time terrain generation in Godot using Marching Squares for dynamic, customizable landscapes. 590
Greater-Flavor-Mod Expands historical depth with richer provinces, cultural variety, and accuracy for immersive grand strategy gameplay. 238
MakeUpUltraFast Delivers high-fidelity Minecraft visuals on low-end hardware via optimized GLSL shaders without sacrificing performance. 251
fubax-shaders Offers a versatile suite of ReShade post-processing shaders for enhanced lighting, color, and cinematic effects in games. 136
StarDustCFWPack Provides a ready-to-use, all-in-one CFW pack for Nintendo Switch with Atmosphere, SX OS, and essential apps pre-configured. 127
godot-ai Streamlines AI integration in Godot with a production-ready MCP server and tools — easy install, free, and developer-friendly. 1.7k
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