Design
Preset
Background
Text
Font
Size
Width
Account Pricing Wednesday, August 5, 2026

The Git Times

“The information you have is not the information you want. The information you want is not the information you need.” — Neil Postman

AI Models
Claude Opus 5 $25/M GPT-5.6 Luna $0.60/M Gemini 3.1 Pro Preview $12/M Grok 4.5 $6/M DeepSeek V4 Pro $0.87/M Qwen3.8 Max $6/M Kimi K3 $15/M
Full Markets →

Fzf's latest release sharpens fuzzy finder speed for short queries 🔗

Optimizations cut latency by up to 2.4x for single-character searches in terminal workflows

junegunn/fzf · Go · ▲ 79 in 1d Est. 2013 · Latest: v0.74.2

Fzf accelerates interactive terminal tasks by letting developers fuzzy-search files, command history, or any list with minimal keystrokes. Its Go-powered engine scans millions of items in milliseconds, turning slow lookups into instant selections.

The v0.74.2 update targets short-query performance—a common pain point—by optimizing the initial scan phase so single-character inputs now run up to 2.4x faster and two-character queries up to 1.4x faster.

Beyond speed, fzf integrates deeply with shells and editors. It ships bindings for Bash, Zsh, and Fish, enabling Ctrl-R history search or Alt-C directory jumps without leaving the prompt. In Neovim or Vim, its plugin transforms file navigation, buffer switching, and grep previews into fluid, preview-aware interactions. Tmux and Zellij users gain stable floating panes thanks to fixed border-label rendering, while Kitty graphics previews no longer corrupt tmux titles.

Built as a single portable binary, fzf avoids runtime dependencies—a boon for containerized or minimal environments. Its programmable interface lets scripts build custom menus via event-driven callbacks, though this requires learning its specific API patterns.

The catch: While fzf excels at ad-hoc exploration, its library use for complex UIs demands significant boilerplate compared to declarative frameworks like Bubble Tea or Zagreus, limiting adoption in full TUI applications.

Why this leads today fzf’s rapid growth reflects its practical, widespread adoption as a lightweight tool that streamlines terminal workflows across development and system tasks, reducing friction in daily command-line use.

Use Cases
  • Developers fuzzy-finding files in large monorepos
  • Sysadmins filtering logs or process lists interactively
  • CLI tool builders creating custom selection menus

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

Fresh on Hugging Face

Model Drops

The newest model releases builders are picking up right now.

More on the Front Page

Cherry Studio 2.0.0 Unifies LLMs, Agents, and Tools in One Desktop AI Workspace 🔗

The release adds split-view layouts, knowledge base agents, and one-click model configuration for Claude Code and Codex.

CherryHQ/cherry-studio · TypeScript · 49.5k stars Est. 2024

Cherry Studio 2.0.

0 refocuses the AI productivity studio from chat-centric interactions to a task-oriented workspace. Built in TypeScript, it now supports multi-window and split-view layouts, letting users run conversations, code analysis, and document review side by side without context switching. The update strengthens agent capabilities by embedding a core runtime that reduces dependency overhead and enables direct reuse of Claude and Codex subscriptions across OpenAI and Anthropic endpoints. Agents can now read and maintain local knowledge bases, functioning as persistent Wiki-like stores that grow with user input. Code tools were rebranded as Code Mate, offering one-click configuration for model services including Claude Code and Codex, while image generation interfaces now expose dynamic parameter controls per model for tighter output tuning. Performance upgrades target responsiveness: a revised data foundation prevents lag during long conversations, and large folder imports no longer block the UI. The app continues to support over 300 pre-configured assistants, local models via Ollama and LM Studio, and cross-platform use on Windows, macOS, and Linux with no setup required.
The catch: Despite reduced environment friction, the agent runtime’s maturity is unproven at enterprise scale, and 1,240 open issues suggest ongoing stability and feature integration challenges.

Use Cases
  • Developers testing multiple LLMs side-by-side for code generation
  • Knowledge workers building persistent AI-assisted research wikis
  • Designers iterating on image prompts with real-time parameter feedback

Source: CherryHQ/cherry-studio — based on the README and release notes.

Rust Agent Runtime ZeroClaw v0.8.4 Hardens Security and Memory Controls 🔗

Maintenance release expands SOP policies, provider fallbacks, and desktop reliability across 49 contributors

zeroclaw-labs/zeroclaw · Rust · 32.5k stars 5mo old

ZeroClaw v0.8.

4 delivers a hardening update with 262 commits from 49 contributors, focusing on memory and SOP control planes. The release adds opt-in retrieval caching, typed-fact extraction, and auditable recall paths for memory, while SOP execution gains per-SOP admission policies and quorum-based approval brokering. Operator surfaces now include dashboard-driven upgrades and searchable keybinding help. Channels gain Mattermost WebSocket listening and signed Nextcloud Talk replies. Provider reliability improves OpenAI Responses setup and streamed retries. The agent remains a single Rust binary configurable for local LLMs and 30+ channels, running with workspace boundaries and OS-level sandboxes. The catch: With 715 open issues and rapid development, long-term stability and documentation completeness remain uncertain for production-critical deployments.

Use Cases
  • Developers deploy private AI agents on Linux/macOS/Windows
  • Teams automate cross-platform workflows via Discord, Telegram, email
  • Enterprises run self-hosted LLMs with tool access and approval gates

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

Virtual Mac enables macOS apps like Xcode on iPad Pro and Air 🔗

Requires jailbroken iPadOS 16–16.3.1 with M1/M2 chip and Sileo package manager

nfzerox/VirtualMacOniPad · Objective-C · 335 stars 1d old

Virtual Mac lets users run macOS 12–27 on iPad Pro (M1, M2) or iPad Air (M1) by leveraging a jailbreak and custom package repository. After adding the Cydia repo and installing via Sileo, users can launch pro apps like Xcode and Terminal directly on iPadOS.

The project supports macOS versions up to Monterey through Golden Gate, with Ventura to Sequoia recommended. Touchscreen and virtual keyboard enable basic interaction, though Magic Keyboard is advised for optimal use. iCloud sign-in is unsupported, requiring manual setup.
The catch: Experimental support for macOS 26–27 may cause visual or performance issues, and jailbreaking voids warranty and carries security risks.

Use Cases
  • Developers compile iOS apps using Xcode on iPad Pro
  • Sysadmins troubleshoot servers via Terminal on iPad Air
  • Designers test macOS-exclusive tools on tablet hardware

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

Morphicons Enables Universal Stroke Icon Morphing with Spring Physics 🔗

Zero-dependency library auto-rotates and morphs icons via 2D Procrustes math

guillermolg00/morphicons · TypeScript · 564 stars 4d old

Morphicons lets any stroke-based SVG icon morph into any other using spring physics, with rotations emerging automatically from shape similarity via 2D Procrustes analysis. The ~7 KB gzipped TypeScript library requires zero runtime dependencies and works as a drop-in replacement for icon sets like Lucide, Tabler, and Heroicons.

It supports React, Vue, Svelte, and React Native via optional peer bindings, passing through all standard props while ensuring correct SSR and accessibility. Animations trigger on prop changes without manual configuration or AnimatePresence wrappers. The latest release fixes SSR hydration mismatches by quantizing path data to four decimal places.
The catch: Morphicons is limited to stroke-based icons and does not support filled or complex gradient shapes, restricting its use with certain icon libraries.

Use Cases
  • Developers animate UI state changes with automatic icon transitions
  • Design systems implement consistent, physics-based icon morphing
  • React Native apps reduce bundle size with zero-dependency icon animation

Source: guillermolg00/morphicons — based on the README and release notes.

macOS App Simplifies Virtual iPhone Management for Researchers 🔗

SwiftUI tool wraps vphone-cli to boot and control iOS research VMs in one window

zqxwce/vphone-ws · Swift · 343 stars 1d old

zqxwce/vphone-ws is a native macOS app that lets developers browse, create, and boot virtual iPhones using Apple's Virtualization.framework.

Built with SwiftUI, it serves as a graphical front-end for the vphone-cli command-line tool, enabling researchers to manage iOS research VMs without terminal workflows. Users can select security levels via variants like Jailbreak or Experimental, allocate resources, and monitor VM state through a live library view. The app includes lifecycle controls, cloning, exporting, and per-VM settings, with host readiness checks for Apple Silicon Macs running macOS 15 or newer. The catch: As a v0.0.1 release with only one open issue, the tool remains early-stage and may lack stability for production research environments.

Use Cases
  • iOS researchers booting virtual devices for app testing
  • Security analysts examining jailbreak-resistant iOS builds
  • Developers validating cloudOS behavior in isolated VMs

Source: zqxwce/vphone-ws — based on the README and release notes.

doc7 Converts Any Document to AI-Ready Markdown Using Local Vision Models 🔗

Recovers visual elements like charts and diagrams as searchable text without OCR stacks

magicrew/doc7 · Go · 441 stars 2d old

doc7 transforms PDFs, Office files, scans, and screenshots into structured Markdown by leveraging any OpenAI-compatible multimodal model running locally via LM Studio or Ollama. It processes text, tables, formulas, charts, and diagram relationships into a single searchable document, preserving visual context often lost in traditional OCR pipelines.

The tool requires no per-page fees or document-processing service lock-in, and supports batch processing, an interactive CLI, and an asynchronous HTTP service. Installers are available for macOS, Linux, and Windows, with checksum verification and quarantine handling on Apple systems. The catch: As a v0.1.1 release, doc7 depends on users to provide and manage their own multimodal model, which may present a barrier for those seeking a fully self-contained solution.

Use Cases
  • Developers converting technical papers to searchable Markdown
  • Teams processing scanned reports with embedded charts and diagrams
  • Researchers extracting formulas and UI states from screenshots for AI training

Source: magicrew/doc7 — based on the README and release notes.

Open Source AI Agents Shift from Tools to Self-Governing Ecosystems 🔗

Projects now bundle memory, governance, and skills into interoperable agent operating systems

Trendai-agents
KKKKhazix/human-writingzeroclaw-labs/zeroclawcomet-ml/opikmicrosoft/agent-governance-toolkitNano-Collective/nanocoder+30 more

The open source AI agent landscape is rapidly evolving beyond isolated skills toward cohesive, self-sustaining ecosystems. Frameworks like omnigent enable real-time collaboration across Claude Code, Codex, and custom agents while enforcing policy and sandboxing—turning agent orchestration into a plug-and-play infrastructure.

Memory layers are becoming standardized: TencentDB-Agent-Memory provides team-level hubs that convert conversations, docs, and code into governed, reusable assets like Skill and Code-Graph, shareable across frameworks. Governance is no longer an afterthought; microsoft/agent-governance-toolkit delivers zero-trust identity, execution sandboxing, and OWASP-aligned policy enforcement for autonomous agents. Meanwhile, skill marketplaces are exploding—alirezarezvani/claude-skills offers 345 Claude-compatible skills spanning compliance to C-level advisory, while nanocoder and CherryHQ/cherry-studio emphasize BYOM (Bring Your Own Model) and local-first control. Even niche needs are addressed: ayghri/i-have-adhd combats agent output burying with ADHD-friendly formatting, and ifixai-ai/iFixAi enables sub-two-minute autonomous agent auditing. The rise of agent operating systems like unicity-aos/aos-ce (Rust) and agentic CMS platforms like CoreBunch/Instatic signals a shift from tooling to self-contained, modifiable agent societies.
The catch: Despite rapid innovation, fragmentation remains severe—skills often lock into specific agent runtimes (e.g., Claude Code vs. Codex), governance tools lack cross-framework audit trails, and memory systems struggle with semantic interoperability, raising doubts about whether today’s ecosystems will compose seamlessly tomorrow.

Use Cases
  • Developers build auditable trading agents with governed memory
  • Teams deploy local-first coding agents with custom skill marketplaces
  • Enterprises enforce zero-trust policies across multi-agent workflows

Open source LLM tools reshape developer workflows with composable AI agents 🔗

Modular skills, memory hubs, and model-agnostic proxies enable flexible, self-hosted AI coding assistance across environments

Trendllm-tools
openobserve/openobservecomet-ml/opikYishenTu/claudianNano-Collective/nanocoderCherryHQ/cherry-studio+25 more

A clear pattern is emerging in open source around composable, model-agnostic LLM tooling that treats AI agents as extensible, interoperable components rather than monolithic services. Projects like omnigent-ai/omnigent provide a meta-harness to orchestrate Claude Code, Codex, and Cursor agents with policy enforcement and real-time collaboration, while TencentCloud/TencentDB-Agent-Memory introduces a team-level memory hub that converts conversations, docs, and code into reusable, governed assets like Skill and LLM-Wiki for cross-agent sharing.

Model flexibility is reinforced by proxies such as lidge-jun/opencodex and Osmantic/ODS, which let developers swap LLMs—Claude, Gemini, DeepSeek, or local Ollama models—without rewriting agent workflows. Skill ecosystems are rapidly standardizing: alirezarezvani/claude-skills offers 345+ ready-to-use skills spanning engineering to compliance, and virgiliojr94/book-to-skill turns technical PDFs into executable Claude Code skills for on-the-job reference. Even niche domains are covered, from bradautomates/claude-video enabling video understanding in agents to AgriciDaniel/claude-seo providing 25 sub-skills for technical and international SEO. The shift is toward decentralized, self-hosted agent stacks where developers bring their own models (Nano-Collective/nanocoder) and avoid vendor lock-in through open protocols and skill interchangeability.

The catch: Despite rapid innovation, the ecosystem remains fragmented—skills often aren’t portable across agent frameworks, memory formats lack standardization, and many tools assume Claude Code or Codex as a baseline, limiting true model agnosticism in practice. Production-grade reliability, security auditing, and consistent evaluation benchmarks are still lagging behind feature velocity.

Lightweight, zero-dependency web tools gain traction in open source 🔗

Developers favor self-contained frameworks that prioritize performance and simplicity over bloated stacks

Trendweb-frameworks
imsai-sh/zhuzhiliaoguillermolg00/morphiconsopenobserve/openobservePostHog/posthogsurveyjs/survey-library+13 more

A clear pattern is emerging in open source web frameworks: a shift toward minimal, dependency-free tools that deliver focused functionality without overhead. Projects like imsai-sh/zhuzhiliao, a single-file HTML simulation of a traditional toy with real audio sampling and mobile-first design, and guillermolg00/morphicons, a ~7 KB gzipped TypeScript library for physics-based icon morphing with zero dependencies, exemplify this drive for lean, embeddable solutions.

Similarly, surveyjs/survey-library offers a dynamic form builder for React, Angular, and Vue with no required framework lock-in, while opengeos/GeoLibre provides a cloud-native GIS platform that runs entirely in the browser, desktop, or Jupyter — all built in TypeScript with an emphasis on lightweight, accessible deployment. Even observability tools like openobserve/openobserve are embracing this ethos, offering a single-binary, high-performance alternative to heavyweight platforms by optimizing storage and reducing complexity. This trend reflects a broader developer preference for tools that are easy to audit, deploy, and integrate — especially in edge, embedded, or resource-constrained environments. The rise of Zig-based HTTP servers like karlseguin/http.zig and profile-enhancing utilities like jstrieb/github-stats further underscores a appetite for precision-engineered, dependency-light web infrastructure.
The catch: While these minimalist approaches excel in simplicity and performance, they often lack the ecosystem breadth, plugin support, and long-term maintenance guarantees of larger frameworks, risking fragmentation and increased burden on developers to reinvent common features like authentication, state management, or accessibility handling across projects.

Deep Cuts

Ontology Toolkit Automates OBO Lifecycle from Build to Release 🔗

Integrates ROBOT, Protege, and Travis for end-to-end ontology CI/CD pipelines

INCATools/ontology-development-kit · Dockerfile · 355 stars

INCATools/ontology-development-kit provides a Docker-based framework to streamline the entire ontology development lifecycle. It automates reasoning, validation, and release workflows using ROBOT for OBO/OWL processing, Protege for editing, and Travis CI for continuous integration.

Builders gain reproducible builds, automated release tagging, and seamless OBO Foundry compliance without scripting glue code. The kit enforces best practices out of the box, reducing drift in collaborative ontology projects. The catch: It's early-stage and niche, assuming familiarity with OBO conventions and Docker workflows, limiting broader adoption.

Use Cases
  • Biomedical researchers standardizing phenotype ontologies
  • OBO Foundry contributors automating release validation
  • Ontology teams enforcing CI/CD for OWL/OWL2 reasoning

Source: INCATools/ontology-development-kit — based on the project README.

Quick Hits

RealReplicaBench RealReplicaBench tests long-horizon agents in high-fidelity, stateful, and reproducible replicas of real online services to evaluate robust, scalable AI performance. 622
Vanta Vanta is a two-faction battle royale where players own their earnings, launching in 2026 with blockchain-based reward mechanics. 344
super-simple-software-factory Super Simple Software Factory packages deterministic, repeatable agent-plus-code workflows as a single skill, embeddable into any repo with bounded coding agents. 382
Swiftlet Swiftlet is a lightweight Swift framework for building modular, type-safe iOS apps with minimal boilerplate and zero runtime overhead. 375
finger-frame-effect-ai Finger Frame Effect AI uses JavaScript to generate realistic hand-frame interactions in web apps, enabling intuitive gesture-based UI enhancements. 498
anydoc AnyDoc converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, and PDF to clean Markdown in Rust, with Node.js and Python bindings for seamless integration. 2.6k
Beyond GitHub

The AI Wire

What builders are reading today — the headlines, papers, and announcements that aren't trending repos.

From the labs & arXiv

Dify Adds Multi-Select Tool Inputs and Workflow Node Locator for Faster AI App Building 🔗

Latest release improves collaboration stability and adds agent DSL export for version-controlled configurations

langgenius/dify · TypeScript · ▲ 110 in 1d Est. 2023 · Latest: 1.16.1

The open-source LLM app platform Dify has rolled out workflow enhancements in version 1.16.

1 that streamline agent and RAG pipeline development. Tool nodes in visual workflows now support multi-select dropdown inputs, letting builders configure parameters with multiple predefined values without custom code. A new workflow node locator allows users to click a node ID in run logs or error messages to instantly highlight the corresponding block in the editor, cutting debugging time.

Collaboration reliability also sees fixes: unsaved edits are no longer dropped when real-time sync drops, and draft saves persist after reconnection. Agent apps can now be exported as DSL YAML directly from the sidebar, enabling version control of configurations. Observability gains include deeper tracing in RAG document processing for better visibility into indexing pipelines.

Built with TypeScript and Next.js, Dify supports self-hosting via Docker Compose with a minimum of 2 CPU cores and 4GB RAM. It integrates with hundreds of LLMs—including GPT, Llama3, and Mistral—through OpenAI-compatible APIs and offers a visual prompt IDE for testing and refining AI interactions.

The catch: Despite active development, the project carries 939 open issues, and its reliance on Docker Compose may complicate deployment for teams avoiding containerization or seeking lightweight, single-binary alternatives.

Previously in The Times “covered” — Aug 4

Use Cases
  • Dev teams building RAG chatbots with multi-source document retrieval
  • AI engineers exporting agent configs for Git-based version control
  • Product teams prototyping LLM workflows via visual drag-and-drop editor

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

More Stories

Superpowers v6.2.0 Fixes Subagent Review Loop with Plan-Scoped Workspaces 🔗

New release prevents cross-plan contamination and adds five-round circuit breaker for fix iterations

obra/superpowers · Shell · ▲ 843 in 1d 9mo old

The latest Superpowers release v6.2.

0 introduces plan-scoped workspaces to eliminate cross-plan ledger contamination in subagent-driven development. Workspaces now resolve per-plan directories under .superpowers/sdd/<plan-basename>/, ensuring task briefs and review packages operate in isolation. The review-fix loop now resumes the implementer instead of dispatching fresh agents, with a scoped re-review prompt and a five-round circuit breaker triggering controller adjudication. These changes stem from live eval campaigns addressing observed workflow failures. Superpowers remains Shell-based and agent-agnostic, supporting tools like Claude Code and Codex CLI.
The catch: Open issues exceed 300, signaling potential stability gaps despite recent structural improvements.

Previously in The Times “covered” — Jul 30

Use Cases
  • Enterprise teams using multiple coding agents in parallel workflows
  • Developers enforcing strict TDD and YAGNI principles via automation
  • Organizations seeking auditable, plan-isolated agent development trails

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

Hugging Face Diffusers Adds NVIDIA Cosmos 3 and Ideogram 4 Pipelines 🔗

Release v0.39.0 introduces unified world foundation model and flow-matching text-to-image generation

huggingface/diffusers · Python · 34.2k stars Est. 2022

The latest release adds Cosmos 3, NVIDIA’s unified world foundation model for Physical AI, enabling video-to-video and action-conditioned generation via a Mixture-of-Transformers architecture. It also integrates Ideogram 4, a flow-matching text-to-image model using asymmetric classifier-free guidance and structured prompt upsampling.

Both pipelines expand multimodal generation capabilities for developers building advanced media systems. The library now supports 30,000+ checkpoints on the Hub, with inference achievable in just a few lines of code using from_pretrained.
The catch: Despite rapid feature growth, 1,324 open issues indicate ongoing challenges in maintaining stability across diverse model integrations and hardware backends.

Use Cases
  • Researchers testing physical reasoning in generated video simulations
  • Artists creating controllable text-to-image outputs with LoRA adaptations
  • Engineers prototyping action-conditioned generation for robotics workflows

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

ComfyUI adds int8 convrot embedding lookup to speed diffusion workflows 🔗

Latest release includes memory optimizations and PrunaVAED support for faster LTX2.3 decoding

Comfy-Org/ComfyUI · Python · ▲ 204 in 1d Est. 2023

ComfyUI v0.30.

0 introduces int8 convrot embedding lookup, reducing memory pressure during model inference. The update also adds PrunaVAED support, accelerating LTX2.3 decoder performance. Memory management now uses an MRU policy with pinning infrastructure to load weights into process RAM more efficiently. These changes target users running complex diffusion pipelines on constrained hardware. The catch: Open issues exceed 4,400, indicating ongoing stability challenges despite rapid development.

Previously in The Times “covered” — Jul 30

Use Cases
  • Generating high-resolution images with Stable Diffusion XL
  • Creating animated video clips from text prompts
  • Building custom 3D model textures via node-based workflows

Source: Comfy-Org/ComfyUI — based on the README and release notes.

Quick Hits

transformers Hugging Face Transformers provides a unified framework for building and training state-of-the-art models across text, vision, audio, and multimodal tasks, enabling rapid prototyping and deployment. 163.4k
streamlit Streamlit lets builders create and share interactive data apps in pure Python with minimal code, turning scripts into shareable web interfaces in minutes. 45.5k
airflow Apache Airflow enables programmatic authoring, scheduling, and monitoring of complex data workflows, ensuring reliable orchestration of ETL and ML pipelines at scale. 46.4k
generative-ai Google Cloud’s Generative AI repo offers ready-to-run notebooks and sample code for leveraging Gemini and Enterprise Agent Platform to build production-grade AI applications. 17.6k
open-webui Open WebUI delivers a clean, extensible interface for interacting with LLMs via Ollama, OpenAI, and other backends, simplifying AI experimentation and integration. 147.9k

Rerun 0.35.0 Adds In-Viewer Catalog for Seamless Local Data Streaming 🔗

Experimental catalog eliminates need for separate server when querying multimodal robotics datasets

rerun-io/rerun · Rust · ▲ 13 in 1d Est. 2022 · Latest: 0.35.0

Rerun’s latest release introduces an experimental built-in catalog in its viewer, allowing developers to stream large .rrd files directly from disk without launching a separate metadata server.

This update streamlines workflows for robotics and computer vision teams working with multimodal data—such as synchronized video, point clouds, and joint states—by reducing setup overhead during local development and debugging.

The catalog, accessible via the viewer’s settings menu, implements the OSS redap server protocol and can be queried using the Python SDK, though connections are limited to localhost for security. Paired with an improved command palette (Cmd+k/Ctrl+k) that now supports entity selection and context-aware actions like refreshing datasets, the changes aim to make interactive data exploration faster and more intuitive.

Built in Rust with column-chunk storage optimized for multi-rate physical data, Rerun continues to offer SDKs in Python, Rust, and C++, enabling logging from sources like robot logs, UMI rigs, and simulation environments. Data remains queryable via dataframes or SQL and can stream directly into ML training pipelines, avoiding export jobs or stale copies.

The catch: The in-viewer catalog is still experimental and may have rough edges; users must manually enable it, and performance for video decoding may suffer if the nasm Cargo feature isn’t enabled due to missing dependencies.

Previously in The Times “covered” — Aug 3

Use Cases
  • Robotics engineers debugging multimodal sensor logs locally
  • ML teams streaming training data directly from Rerun viewer
  • CV researchers comparing pipeline outputs across synchronized modalities

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

More Stories

commaai/openpilot Adds Rivian and Acura Support in v0.11.1 🔗

Release improves driver monitoring and thermal policy for comma four hardware

commaai/openpilot · Python · ▲ 12 in 1d Est. 2016

The latest openpilot v0.11.

1 release adds support for Rivian R1S/R1T 2025 and Acura MDX 2022–24 models, expanding compatibility to over 300 vehicles. It features a new driver monitoring model, enhanced image processing for the driver-facing camera, and refined thermal management for the comma four device. Built primarily in Python with critical safety components in C, openpilot remains an open-source alternative to proprietary driver-assistance systems. The project enforces ISO26262 guidelines and runs continuous software-in-the-loop testing.
The catch: Real-world performance depends heavily on vehicle-specific harness installation and comma four hardware, limiting broader DIY adoption.

Previously in The Times “covered” — Aug 1

Use Cases
  • Developers testing ADAS on supported electric vehicles
  • Engineers customizing driver assistance via openpilot branches
  • Fleets evaluating open-source alternatives to OEM systems

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

TortoiseBot v2.0 adds ROS Noetic support and improved hardware integration 🔗

Release simplifies sensor mounting and SD card access while updating navigation parameters

rigbetellabs/tortoisebot · Makefile · 187 stars Est. 2021

TortoiseBot v2.0 introduces a custom power distribution board PCB, lasercut branding, and extra mounting holes for sensors and Jetson Nano.

Software updates include ROS Noetic compatibility, revised differential and camera nodes in Python 3, and refined DWA navigation settings. The project moved CAD files to a separate repo to reduce package size and now publishes /odom from Cartographer. Simulation runs in Ignition Gazebo with RViz2 integration for SLAM mapping and autonomous navigation via AMCL. Wi-Fi setup requires manual SD card editing.
The catch: Open issues persist, and the last commit was over a day ago, raising questions about active maintenance despite the recent v2.0 tag.

Use Cases
  • Educators teaching ROS 2 navigation with low-cost hardware
  • Hobbyists building autonomous mapping robots for home use
  • Researchers testing SLAM algorithms in simulation and real-world environments

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

Quick Hits

awesome-robot-descriptions A curated collection of robot descriptions in URDF and MJCF formats for simulation and modeling. 1.6k
IsaacLab A unified Python framework on NVIDIA Isaac Sim for streamlined robot learning and reinforcement learning workflows. 7.8k
robot_descriptions.py Provides programmatic access to 190+ robot descriptions across major Python robotics frameworks like ROS, PyBullet, and MuJoCo. 800
text-to-cad A JavaScript library offering agent-driven skills for automating CAD, CAE, and CAM tasks in engineering workflows. 12.9k
carla An open-source C++ simulator enabling realistic autonomous driving research with sensor-rich, traffic-aware environments. 14.3k
open_manipulator A Python-based AI-powered manipulator platform combining Open Manipulator hardware with intelligent control for dexterous tasks. 647

ImHex v1.38.1 ships signed Windows installers and fixes SSH memory crash 🔗

Release improves macOS auto-update flow and resolves selection gaps under display scaling

WerWolv/ImHex · C++ · ▲ 11 in 1d Est. 2020 · Latest: v1.38.1

WerWolv/ImHex released v1.38.

1, addressing long-standing usability friction for reverse engineers. Windows installers are now code-signed via SignPath, eliminating SmartScreen warnings on deployment. The macOS auto-updater unpacks new versions directly instead of leaving users to manually extract DMGs, streamlining updates. Selection rendering gaps under non-standard display scaling have been closed, and opening an already-loaded file now switches tabs rather than throwing an error. A critical bug where clicking Cancel during remote SSH file opens triggered a crash has been patched, alongside fixes to the pattern editor and data inspector value editing. Extended ASCII display, previously enabled by default in v1.38.0, is now opt-in via the paragraph icon in the footer.
The catch: Despite its rich feature set, ImHex remains a single-maintainer project with 378 open issues, raising concerns about long-term triage latency and dependency sustainability for enterprise workflows.

Previously in The Times “covered” — Jul 30

Use Cases
  • Reverse engineers analyzing firmware binaries
  • Developers debugging endianness in network packets
  • Security researchers patching IPS-format game ROMs

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

More Stories

DVWA adds OpenAPI module for vulnerable API security training 🔗

Release 2.5 introduces hands-on practice with insecure API endpoints

digininja/DVWA · PHP · 13.5k stars Est. 2013

The Damn Vulnerable Web Application (DVWA) project released version 2.5, adding a new module focused on exploiting and defending against vulnerable APIs using OpenAPI specifications.

Built with PHP and MariaDB, DVWA remains a go-to tool for security professionals and educators to legally practice common web flaws like SQL injection and cross-site scripting in isolated environments. The update expands its scope beyond traditional web vulnerabilities to include API-specific risks, reflecting modern attack surfaces. The catch: DVWA intentionally contains undocumented flaws and must never be deployed on public-facing servers due to high compromise risk.

Use Cases
  • Security teams testing API vulnerability scanners
  • Students learning secure API design principles
  • Developers practicing input validation in REST endpoints

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

NGINX Patches Critical Buffer Overflow in Map Regex Handling 🔗

Release 1.31.3 fixes three CVEs including memory disclosure and use-after-free flaws

nginx/nginx · C · 31.3k stars Est. 2015

NGINX 1.31.

3 patches a buffer overflow (CVE-2026-42533) triggered by map directives using regex, alongside memory disclosure in ngx_http_slice_module (CVE-2026-60005) and a use-after-free in ngx_http_ssi_module (CVE-2026-56434). The release also includes architectural tweaks: cache line size configuration for loongarch64, HTTP/2 frame handling fixes, and stricter bounds checks in complex value compilation. Despite its ubiquity as a reverse proxy and load balancer, the project shows signs of maintenance strain with 458 open issues and reliance on incremental fixes rather than systemic overhauls.
The catch: Rapid patching masks deeper architectural complexity that complicates auditing at scale.

Previously in The Times “covered” — Jul 29

Use Cases
  • DevOps teams terminating TLS termination and HTTP/2 multiplexing
  • [8-12 word use case]
  • [8-12 word use case]

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

BBOT 3.0.1 fixes DNS tag loss and WAF bypass logic for accurate recon 🔗

Parallelized bypass checks and cloud tag propagation restore scan integrity after breaking CLI changes

blacklanternsecurity/bbot · Python · ▲ 7 in 1d Est. 2022

BlackLanternSecurity’s bbot v3.0.

1 patches critical flaws in its recursive scanner: cloudcheck tag propagation now survives DNS resolution, preventing loss of asset context in subdomain trees. The waf_bypass module parallelizes checks up to 100 threads and distinguishes IP-based from DNS-based bypass candidates in logs. Control characters in output are escaped to avoid terminal garble. Despite v3.x’s breaking CLI and API changes, this point release targets stability for bug bounty and ASM workflows. The catch: Heavy reliance on Neo4j and custom YAML presets creates a steep onboarding curve for teams preferring lightweight, JSON-native tools.

Previously in The Times “covered” — Jul 30

Use Cases
  • Security teams mapping cloud asset relationships via DNS and open ports
  • Bug hunters enumerating subdomains with mutation-based brute force
  • Red teams testing WAF evasion using parallelized bypass checks

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

Quick Hits

web-check A comprehensive OSINT toolkit that uncovers hidden website details — from tech stack to exposed assets — for rapid threat intelligence gathering. 34.4k
mitmproxy An interactive proxy that lets you inspect, modify, and replay HTTPS traffic in real time to debug apps and test security flaws. 44.6k
opencti A centralized platform to ingest, enrich, and correlate threat data from multiple sources into actionable cyber threat intelligence. 9.8k
juice-shop A deliberately vulnerable web app designed to teach and practice modern web security exploits in a realistic, up-to-date environment. 13.6k
vuls An agentless scanner that automatically detects vulnerabilities across servers, containers, and dependencies without installing anything on targets. 12.2k

OpenAI’s Codex CLI Adds Session Naming and Forking for Local Agents 🔗

Latest release enables pinned threads, plugin manifests, and WebSocket connections to remote code hosts

openai/codex · Rust · ▲ 270 in 1d Est. 2025 · Latest: rust-v0.146.0

The open-source Codex CLI, a Rust-based coding agent from OpenAI, now lets users name sessions with /new or /clear, pin important threads, and fork conversation history—including temporary forks that stay hidden from listings. The v0.

146.0 release adds support for Agent Plugins manifests, enabling workspace plugin publishing and access to marketplaces for Amazon Bedrock and Claude Code. Users can connect the app-server to remote Code Mode hosts via WebSocket and trigger standalone web search for compatible custom models. Executor-provided skills are now discoverable, with secure resource access for explicitly selected skills. Installation remains platform-specific: macOS ARM64 users pull codex-aarch64-apple-darwin.tar.gz, Linux x86_64 grabs codex-x86_64-unknown-linux-musl.tar.gz, and Windows relies on standalone installers from OpenAI’s releases. After setup, running codex prompts sign-in via ChatGPT to unlock Plus, Pro, or Enterprise features, or API key use with extra configuration.

The catch: Despite active development—11,801 open issues and commits just days ago—the CLI still requires manual binary renaming after extraction and lacks native IDE integration beyond third-party editor plugins, creating friction for seamless workflow adoption.

Previously in The Times “covered” — Aug 2

Use Cases
  • Developers debugging locally with persistent, named agent sessions
  • Teams sharing plugin workflows via Bedrock and Claude Code marketplaces
  • Engineers linking terminals to remote code hosts over secure WebSocket connections

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

More Stories

Kubernetes 1.36.3 patches security and scheduling bugs in core control plane 🔗

Release fixes etcd watch leaks and improves node affinity for stateful workloads

kubernetes/kubernetes · Go · ▲ 25 in 1d Est. 2014

The Kubernetes project released v1.36.

3, addressing critical etcd watch resource leaks that could cause API server memory exhaustion under heavy load. Scheduling enhancements now better handle pod disruption budgets during voluntary node evictions, reducing unintended restarts for stateful applications. Built in Go, the platform manages containerized workloads across hybrid clusters using declarative APIs and automated self-healing. Despite its maturity, the project maintains 2,757 open issues, reflecting ongoing complexity in scaling and multi-tenancy.
The catch: Operators still report steep operational overhead when debugging network policies and storage class interactions at scale.

Use Cases
  • DevOps teams deploy microservices across on-prem and cloud
  • SaaS providers scale stateful workloads with rolling updates
  • Platform engineers automate hybrid cluster lifecycle management

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

RAGFlow v0.26.4 adds Snowball stemmer for 16 languages, fixes parsing bugs 🔗

Release improves tokenization and resolves MCP server, Docling formula extraction issues

infiniflow/ragflow · Go · ▲ 93 in 1d Est. 2023

RAGFlow’s latest release integrates a language-aware Snowball stemmer supporting 16 languages, enhancing multilingual tokenization across its pipeline. It also patches critical bugs: the MCP server now correctly handles paginated chat lists, Docling parser preserves mathematical formulas, and metadata edits persist to backend.

Built in Go, the engine fuses RAG with agent capabilities for LLM context layers.
The catch: Nearly 1,900 open issues suggest ongoing stability challenges despite rapid feature expansion.

Previously in The Times “covered” — Jul 30

Use Cases
  • Enterprises building multilingual AI agents with document retrieval
  • Developers syncing knowledge from Confluence, Notion, and S3
  • Teams deploying self-hosted RAG pipelines with GPT-5 and DeepSeek v4 support

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

Microsoft Terminal gains stability fixes in v1.26 after long development 🔗

Open-source project with 9,485 forks addresses memory corruption and driver deadlock issues

microsoft/terminal · C++ · ▲ 8 in 1d Est. 2017

Microsoft's Windows Terminal repository released v1.26 with bug fixes resolving tab-drag crashes, screen reader announcements, and URL detection during scrolling.

The update includes a speculative initialization shift to prevent NVIDIA driver deadlocks and a memory corruption fix in DRCS font handling. Despite 1,748 open issues and near-decade maturity, the project maintains active contributions via good-first-issues and Hacktoberfest tags.
The catch: Manual installation requires VC++ v14 Desktop Framework on older Windows 10 builds, creating a barrier for enterprise users avoiding Store dependencies.

Use Cases
  • Developers customizing command-line interfaces with Windows Terminal
  • System administrators managing WSL2 environments through integrated console
  • Power users configuring split-pane workflows for simultaneous tool monitoring

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

Quick Hits

uv astral-sh/uv: A Rust-powered Python package manager that installs and resolves dependencies in seconds, drastically speeding up development workflows. 88.4k
starship starship/starship: A lightning-fast, fully customizable shell prompt that enhances productivity-aware terminal 59.3k
obs-studio obsproject/obs-studio: Free, open-source tool for high-quality video game-like feedback in any terminal, written in Rust. 74.6k
linux torvalds/linux: The core of modern computing — a modular, high-performance kernel enabling everything from smartphones to supercomputers. 241.8k
json nlohmann/json: A single-header, standards-compliant C++ library that makes parsing, generating, and manipulating JSON intuitive and efficient. 50.3k

ESP32 Pinout Tool Eliminates Guesswork With Real-Time Conflict Detection 🔗

Built from Espressif’s official KiCad libraries, it prevents hardware failures by mapping safe GPIOs across 23 modules.

FelixKunzJr/ESPPinoutWebsite · TypeScript · ▲ 1 in 1d 2mo old

FelixKunzJr/ESPPinoutWebsite delivers an interactive pinout reference for the entire ESP32 family, hosted live at esp32pin.com.

Unlike static diagrams or datasheet snippets, the tool renders pin constraints directly on schematic and module views, flagging strapping pins, ADC2/Wi-Fi conflicts, flash-reserved GPIOs, and input-only pins in real time. Its pin mapping builder includes live conflict detection, allowing users to select GPIOs for peripherals while instantly seeing which choices risk boot failures, flash corruption, or Wi-Fi instability. Selections export as Arduino #define statements or shareable URLs, with PNG output for documentation.
Technically, the site avoids manual data entry by generating pin names, pad layouts, and schematic symbols from Espressif’s official KiCad libraries via src/data/chips/generated.ts. Family-specific rules—like strapping roles and boot-mode dependencies—are maintained in a curated FAM table and validated against Espressif’s esp-gpio-tool dataset through automated cross-checking in tests/espressif-crosscheck.test.ts. This ensures the data stays synchronized with silicon-level specifications, reducing the risk of silent drift. The project covers 23 modules across ESP32, S2, S3, C3, C5, C6, and H2 variants, each rendered with accurate castellated pad layouts and EDA-style schematics.
The catch: While the tool excels at preventing hardware misconfiguration, it does not simulate runtime behavior or validate electrical characteristics like drive strength, pull-up/down requirements, or analog signal integrity—builders must still consult datasheets for final validation.

Previously in The Times “covered” — Jul 31

Use Cases
  • Embedded developers selecting safe GPIOs for Wi-Fi and ADC
  • Hardware designers verifying pin compatibility before PCB layout
  • Makers avoiding boot-mode strapping errors on dev boards

Source: FelixKunzJr/ESPPinoutWebsite — based on the project README.

More Stories

PiKVM Enables Remote Server Control via Raspberry Pi-Based IP-KVM 🔗

Supports HDMI capture, virtual media, and power management across Pi 2/3/4/Zero2W platforms

pikvm/pikvm · Unknown · 10.2k stars Est. 2019

The pikvm/pikvm project delivers an open-source IP-KVM solution built on Raspberry Pi hardware, allowing remote keyboard, video, and mouse control of servers even when the OS is unresponsive. It supports HDMI-to-CSI or USB video capture, virtual CD/DVD and flash drive emulation, USB and Bluetooth HID, PS/2 compatibility, and server power control via ATX functions.

Access is available through a Web UI or VNC, with H.264 video streaming at 35-50ms latency and two-way audio. The system includes IPMI, Redfish, and Wake-on-LAN support, GPIO control, and HTTPS-secured extensible authorization. The catch: Raspberry Pi 5 is not supported due to lack of compatible GPU video encoders, limiting use of newer Pi hardware despite its availability.

Use Cases
  • IT admins reboot unresponsive servers remotely
  • Developers test BIOS-level firmware changes
  • Technicians reinstall OS via virtual media over IP

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

vdbrink's Docs Streamline Home Automation Setup for Builders 🔗

Recent commit adds Node-RED flow examples for ESP32-based Zigbee networks

vdbrink/vdbrink.github.io · CSS · 48 stars Est. 2021

The vdbrink/vdbrink.github.

io repository offers practical CSS-styled documentation for integrating Node-RED, Home Assistant, and ESPHome. It provides step-by-step guides, YAML snippets, and wiring tips focused on DIY electronics projects using ESP32, ESP8266, and Zigbee devices. The latest commit from one day ago includes updated flow diagrams for Node-RED, improving clarity on MQTT topic mapping and state persistence. Builders use it to reduce setup time when configuring home automation systems from scratch. The catch: Documentation assumes familiarity with YAML and basic networking, offering limited hand-holding for absolute beginners in home automation.

Use Cases
  • Configure ESP32 with ESPHome for local sensor control
  • Wire Zigbee devices into Home Assistant via deCONZ
  • Design Node-RED flows for automated lighting schedules

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

Quick Hits

mural A low-cost, open-source wall plotter that turns any surface into a canvas for precise, scalable drawings — ideal for makers seeking affordable large-format output. 284
axi Synthesizable SystemVerilog AXI IP modules with full verification infrastructure, enabling high-performance, reusable on-chip communication for FPGA/ASIC designers. 1.6k
librealsense Cross-platform C++ SDK for Intel RealSense depth cameras, providing real-time 3D perception, tracking, and facial recognition for robotics and AR applications. 8.9k
iiab Turn a Raspberry Pi into an offline educational hotspot delivering curated global knowledge — Wikipedia, Khan Academy, and more — without internet. 1.9k
TuyaOpen Unified C framework for rapid AIoT development across Tuya chips, enabling seamless hardware integration of AI agents and IoT sensors with minimal code. 1.8k
minibolt A step-by-step guide to building a Bitcoin & Lightning node, and other stuff, on a personal computer. 91

GDevelop Integrates AI Assistance to Streamline No-Code Game Creation 🔗

Latest release fixes panel sprite crashes and improves performance for HTML5 exports

4ian/GDevelop · JavaScript · ▲ 29 in 1d Est. 2014 · Latest: v5.6.276

GDevelop’s v5.6.

276 update introduces AI-assisted event generation, letting creators describe game logic in plain text for the engine to translate into visual scripts. This lowers the barrier for beginners while speeding iteration for veterans building 2D, 3D, or multiplayer titles across web, mobile, and desktop. The release also patches a side-panel crash and optimizes sprite rendering, addressing stability issues reported in the 618 open issues. Built on JavaScript with MIT-licensed core and HTML5 engines, GDevelop exports to gd.games, Steam, and app stores without requiring code. Its event system and modular behaviors remain central, now augmented by AI that suggests or builds logic alongside users.
The catch: AI features are experimental and may not handle complex multiplayer synchronization or 3D logic reliably, requiring manual refinement for networked games.

Previously in The Times “covered” — Aug 4

Use Cases
  • Hobbyists create mobile puzzle games without writing code
  • Educators teach game design using visual event systems
  • Indie devs prototype multiplayer HTML5 games for web distribution

Source: 4ian/GDevelop — based on the README and release notes.

More Stories

OpenSpeedy Hooks Windows Wait Functions to Boost Game Speed 🔗

New release adds WaitForSingleObject/Ex and WaitForMultipleObjects/Ex support for frame-rate scaling

game1024/OpenSpeedy · TypeScript · 16.9k stars Est. 2025

OpenSpeedy 3.3.

8 enhances game speed modification by hooking Windows wait primitives commonly used for frame limiting.used for frame pacing. The tool now scales timeouts in WaitForSingleObject, WaitForSingleObjectEx, WaitForMultipleObjects, and WaitForMultipleObjectsEx, preserving behavior for zero and INFINITE timeouts. It removed the redundant QueryPerformanceFrequency hook to avoid scale cancellation. Built with Tauri and TypeScript, it adjusts speed via Ring-3 hooking without kernel changes. Users select a process and apply a multiplier for immediate effect. The catch: Hooking wait functions may still trigger anti-cheat detection in online games, risking bans despite Ring-3 operation.

Previously in The Times “covered” — Jul 30

Use Cases
  • Single-player gamers adjust pacing in legacy titles
  • Modders test game mechanics at variable speeds
  • Developers debug time-sensitive code in Windows builds

Source: game1024/OpenSpeedy — based on the README and release notes.

Pixelorama adds keyframe animation and 3D layer editing in v1.2 update 🔗

Built on Godot 4.6.3, the release introduces autotiling and brush-enabled shape tools

Orama-Interactive/Pixelorama · GDScript · ▲ 6 in 1d Est. 2019

Pixelorama’s v1.2 update brings keyframe-based timeline editing for animating layer opacity and effects with interpolation.

The 3D layer system was rewritten to share objects across frames, enabling property animation, GLTF import, and direct drawing on 3D models — though this breaks pxo file compatibility with older versions. Autotiling support now lets users configure tilemap bits per tile, and shape tools accept custom brushes. Built with GDScript, it runs on Windows, Linux, macOS, and web. The catch: 3D layer data in pxo files is incompatible with versions prior to v1.2, requiring conversion or recreation of older projects.

Use Cases
  • Indie developers creating animated sprites for 2D games
  • Artists designing tilesets with autotiling for procedural level generation
  • Hobbyists experimenting with 3D pixel art and model painting

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

Quick Hits

raylib raylib lets C developers build games quickly with minimal setup, offering cross-platform 2D/3D rendering, input, and audio in a single-header-friendly library. 34.2k
quest-system quest-system provides Godot 4 developers a modular, extensible framework to design and manage complex quests with visual scripting and clean GDScript integration. 459
Reia Reia combines Godot’s flexibility with Rust’s performance to deliver a scalable, action-adventure MMO framework for persistent online RPG experiences. 898
assimp Assimp enables seamless import of over 40 3D model formats into a unified data structure, simplifying asset pipelines for game and simulation developers. 13.1k
godot-ai hi-godot/godot-ai delivers production-ready AI tools and an MCP server for Godot, offering easy installation and powerful LLM integration without licensing hassles. 1.4k
Super-Mario-Bros.-Remastered-Public A Remake / Celebration of the original 'Super Mario Bros.' games. Features new levels, custom modes, new characters, alongside a full level editor / custom level system! 3k
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