Design
Preset
Background
Text
Font
Size
Width
Account Pricing Saturday, August 1, 2026

The Git Times

“The medium is the message.” — Marshall McLuhan

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 Qwen3.7 Max $4.42/M Kimi K3 $15/M
Full Markets →

τ0-VLA Enables Robots to Reason Through Complex Tasks Using World Models 🔗

Hierarchical architecture separates high-level planning from low-level execution across robot embodiments

sii-research/tau-0-vla · Python · ▲ 18 in 1d 6d old

The τ0-VLA project introduces a hierarchical robot foundation model designed for long-horizon manipulation tasks. Its high-level policy generates subtasks and employs world-model-guided test-time computation to explore alternatives when reasoning is needed.

A generalist low-level policy then executes these subtasks using a Qwen3.5 vision-language backbone and a Mixture-of-Transformers action expert trained via conditional flow matching. Operating in a unified 40-dimensional state/action space, the model was trained on 40,115 hours of heterogeneous real-world robot data with multimodal co-training. The reference stack requires Python 3.11, CUDA 12.8, and PyTorch 2.7.1, with example data provided in LeRobot v3.0 format. Code and weights are released under Apache License 2.0, and serving currently supports joint-control checkpoints only.
The catch: Native end-effector (EEF) data can be used for training, but EEF serving is not supported in this release, limiting deployment flexibility for certain robot configurations.

Why this leads today The model’s hierarchical design and test-time world-model integration enable more adaptive, reliable robotic behavior in real-world settings, marking a meaningful step forward in AI-driven automation for industry applications.

Use Cases
  • Industrial robots performing multi-step assembly tasks
  • Research labs testing generalizable manipulation policies
  • Robotics teams adapting policies across different robot hardware
  • Developers building long-horizon autonomy in unstructured environments

Source: sii-research/tau-0-vla — based on the project README.

Fresh on Hugging Face

Model Drops

The newest model releases builders are picking up right now.

More on the Front Page

3DS Port Brings Zelda: A Link to the Past to Dual Screens 🔗

Open-source engine adapts classic SNES game for 3DS with live map and turbo support

EstebanPdN/zelda-alttp-3ds · C · 229 stars 4d old

EstebanPdN’s zelda-alttp-3ds project ports the reverse-engineered Zelda 3 engine to Nintendo 3DS, leveraging dual screens for enhanced gameplay. The top screen displays the 400x240 game world, while the bottom screen shows a live map, dungeon layout, gear, items, and settings.

Built in C and based on community reverse-engineering efforts, the port extracts game assets directly from a user-provided, legally obtained USA unheadered ROM (.sfc or .smc) on first launch, storing them locally as zelda3_assets.dat. No ROM or assets are included in the repository, ensuring compliance with copyright.

Installation requires placing the CIA file on the 3DS SD card and launching via Homebrew Launcher or FBI. Features include three display modes—Original, Stretch, and Wide—with turbo speed support on New 3DS models. The latest release, v2.7, fixes a long-standing issue in Wide Mode where black borders appeared at map edges; the camera now halts at map boundaries like the original aspect ratio, improving visual fidelity. Users can trigger diagnostics with L + R + A to generate memory dumps and screenshots for bug reporting. Audio functionality depends on Luma3DS’s Rosalina DSP firmware dump.

The project supports translated ROMs, falling back to the original USA ROM for missing assets, and provides QR codes and clean source zips in each release.

The catch: The port requires users to supply their own ROM and relies on homebrew firmware (Luma3DS) for audio, limiting accessibility to modified consoles and legally compliant users.

Use Cases
  • Retro gaming enthusiasts play Zelda on 3DS hardware
  • Developers study open-source game engine porting techniques
  • Homebrew communities preserve and enhance classic titles on modernized handhelds

Source: EstebanPdN/zelda-alttp-3ds — based on the README and release notes.

Multiagent Workspace Tool Lets Teams Share AI Safely 🔗

QM v0.1.4 restricts model picks to org-approved lists and uses latest bootstrap

yc-software/qm · TypeScript · 3.2k stars 2d old

yc-software/qm is a TypeScript-based multiplayer agent harness built for startups. It gives each employee an isolated workspace with personal memory, files, and permissions while enabling collaboration in Slack and web channels.

The tool supports swapping between AI harnesses like Pi, OpenCode, Codex, and Claude Code through a shared core, avoiding vendor lock-in. Admins control org-wide settings, approve models, and manage skill sharing via git-imported packs. Background crons and watches automate tasks like inbox triage and CI monitoring. The Postgres layer persists user data across sessions. The catch: With 39 open issues and a v0.1.x release, the project is early-stage and may lack stability for enterprise-scale deployment.

Use Cases
  • Startup teams isolate AI workspaces per employee
  • Admins enforce approved AI models across org
  • Teams automate inbox triage using agent crons

Source: yc-software/qm — based on the README and release notes.

C Engine Streams Trillion-Parameter Kimi K3 Model from NVMe on Consumer Hardware 🔗

WASTE runs full 2.78T-parameter model on 64GB MacBook Pro by streaming experts from disk

sqliteai/waste · C · 430 stars 3d old

WASTE is a dependency-free C inference engine that runs the full 2.78-trillion-parameter Kimi K3 model on consumer hardware by streaming activated experts directly from NVMe storage.

It keeps the model trunk in memory and uses remaining RAM as a bounded cache for frequently accessed experts, enabling inference on a 64GB MacBook Pro despite the model’s 982 GiB footprint. Validation against PyTorch shows logit agreement within 3.6e-06. The engine achieves 0.49–0.54 tokens per second — roughly thirty seconds per sentence — trading speed for accessibility. The catch: Throughput is too low for interactive use, and the engine is currently validated only for the Kimi K3 model, limiting broader applicability without further work.

Use Cases
  • Researchers run massive MoE models on laptops without cluster access
  • Developers embed LLM inference in C applications with zero runtime dependencies
  • Teams test trillion-parameter models locally before deploying to server infrastructure

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

Kimi Code CLI Enables Terminal-Based AI Coding Agents With Video Input 🔗

Supports Model Context Protocol servers and subagents for parallel, focused workflows

MoonshotAI/kimi-code · TypeScript · 5.9k stars 2mo old

Kimi Code CLI is a TypeScript-based AI coding agent that runs in the terminal, reads and edits code, runs shell commands, and processes video input to generate working code from screen recordings. It installs as a single binary with no Node.

js required, offers fast TUI startup, and integrates with Moonshot AI’s Kimi models or other compatible providers. Users can configure MCP servers conversationally, install plugins from GitHub, and dispatch subagents for isolated tasks. The latest patch fixes intermittent model configuration errors and improves TUI rendering.
The catch: Despite active development, 839 open issues suggest stability and feature maturity remain works-in-progress for production-critical agent workflows.

Use Cases
  • Developers automate refactoring using natural language prompts in terminal
  • Teams build multi-step workflows with parallel subagents for exploration and coding
  • Engineers prototype features by feeding screen recordings into the agent for code generation

Source: MoonshotAI/kimi-code — based on the README and release notes.

GitHub Hook Checks AI Agent Compliance With Coding Rules in Real Time 🔗

Ratchet measures edits against minimalism standards as agents work, blocking violations mid-session

0xwilliamortiz/ratchet · JavaScript · 402 stars 0d old

Ratchet injects a PostToolUse hook that scans every agent edit, measuring added lines against rules like "prefer standard library" and "keep diffs small." It compares changes to a repository baseline, outputting grouped compliance scores by grade—guard, strict, or default—without relying on model persuasion.

The tool requires Node 20+ and Git for full auditing; on Windows, hooks/ratchetui.exe launches automatically. Agents can switch modes mid-session via /ratchet strict or persist settings with environment variables. The catch: Compliance tracking depends on Git availability, limiting audit features in environments without Git on PATH.

Use Cases
  • Developers enforce minimalism rules during AI-assisted coding sessions
  • Teams maintain code quality by blocking non-standard library dependencies
  • Agents self-correct by receiving real-time feedback on edit violations

Source: 0xwilliamortiz/ratchet — based on the project README.

Packets/tew Enables Low-Latency Network Packet Filtering in Userspace 🔗

Forked 57 times despite no README or public documentation since creation

Packets/tew · Unknown · 340 stars 0d old

The Packets/tew project provides a userspace framework for intercepting and filtering network packets with minimal kernel overhead, targeting developers building custom firewalls or traffic analyzers. Created and last updated within two minutes on July 31, 2026, it shows no signs of active maintenance beyond initial commit.

With zero open issues and a last commit one day ago, the project remains static despite notable community interest reflected in its fork count. The absence of documentation or language specification raises questions about usability and intended audience. The catch: No public documentation, language identification, or usage examples exist, making adoption dependent on reverse-engineering the codebase.

Use Cases
  • Network engineers building custom packet filters
  • Security researchers analyzing traffic in userspace
  • Developers prototyping low-overhead firewall tools

Source: Packets/tew — based on the project README.

Wilonity Loader Provides Game Cheat Tools for Over 20 Popular Titles 🔗

The tool offers kernel-level spoofing, HWID cleaning, and anti-cheat bypass for EAC, BE, and VG systems.

WilonityDev/WilonityLoader · Unknown · 532 stars 1d old

Wilonity Loader is a cheat library featuring spoofer, driver bypass, and undetected injector for games including RUST, CS2, Valorant, Tarkov, Warzone, R6S, Apex, and Roblox. It provides ESP, aimbot, wallhack, triggerbot, radar, no recoil, silent aim, chams, skin changer, and unlock-all modules.

The Unified Launcher supports Windows 10/11 x64 and ARM64 with .NET Framework 4.8+. Users download the installer, run setup, and launch via desktop shortcut. Despite claims of safety and regular updates, the project shows no public code transparency, with language unspecified and no recent commits beyond initial push.
The catch: The project lacks visible source code, issue tracking, or community feedback despite claiming open-source MIT licensing.

Use Cases
  • Gamers seeking aimbot and ESP in competitive shooters
  • Users bypassing hardware bans via HWID cleaning and spoofing
  • Players wanting unlocked cosmetics and features in supported titles

Source: WilonityDev/WilonityLoader — based on the project README.

Open Source Agents Evolve Beyond Chat Into Autonomous Workflows 🔗

New projects show agents now manage devices, code, compliance, and media without constant human prompting

The open source agent ecosystem is shifting from conversational interfaces to self-directed systems that perceive, decide, and act across domains. Projects like callstack/agent-device enable CLI control of iOS and Android hardware for automated testing, while alibaba/open-code-review deploys LLM agents to enforce deterministic security rules alongside fine-tuned for NPE, XSS, and SQLi at scale.

Workflow automation is advancing through herdrdev/herdr, a Rust-based terminal multiplexer that routes agent tasks, and can1357/oh-my-pi, which provides hash-anchored edits and subagent orchestration for terminal-based coding. Compliance and reliability are gaining focus: 0xwilliamortiz/ratchet validates agent behavior against rule sets, and microsoft/agent-governance-toolkit implements zero-trust identity and sandboxing aligned with OWASP Agentic Top 10. Creative and analytical use cases are emerging too—calesthio/OpenMontage offers a 12-pipeline agentic video studio, and xbtlin/ai-berkshire runs multi-agent adversarial analysis using value investing frameworks from Buffett and Munger. Even niche needs are addressed, such as ayghri/i-have-adhd, which modifies agent output to prevent burying key information for neurodivergent users.

The catch: Despite rapid innovation, most agents remain brittle outside narrow scopes, struggle with long-term coherence, and lack standardized interfaces—making real-world deployment risky without significant customization, and raising concerns about fragmentation as teams build isolated solutions rather than interoperable systems.

Use Cases
  • Developers automate iOS/Android testing via CLI agent control
  • Enterprises enforce code safety with LLM agents and deterministic rules
  • Neurodivergent coders get ADHD-friendly agent output adjustments

Open Source Embraces Modular AI Skills for Agent Workflows 🔗

Developers build interchangeable LLM-powered tools that agents like Claude Code or Codex

Trendllm-tools
0xwilliamortiz/ratchettiann/hapiNVIDIA/TensorRT-LLMvirgiliojr94/book-to-skilldiegosouzapw/OmniRoute+23 more

A clear pattern is emerging in open source: the rise of composable, LLM-driven "skills" that extend AI coding agents with domain-specific capabilities. Projects like tiann/hapi and VoltAgent/awesome-agent-skills offer universal interfaces to plug in hundreds of models and skills across providers, while bradautomates/claude-video gives Claude the ability to process video frames and transcripts on demand.

Similarly, virgiliojr94/book-to-skill transforms technical PDFs into instant reference tools for agents, and albaba/open-code-review delivers deterministic, LLM-augmented code reviews with fine-tuned rulesets for security and reliability.

This modular approach shifts focus from monolithic AI apps to interchangeable toolkits—evident in lidge-jun/opencodex (universal provider proxy) and colbymchenry/codegraph (local code knowledge graphs reducing token use)—enabling agents to dynamically adapt to tasks like stock analysis (zhuLinsen/daily_stock_analysis) or reverse engineering (zhaoxuya520/reverse-skill) without rebuilding core logic. The trend reflects a move toward agent ecosystems where skills are shared, versioned, and composed like libraries.

The catch: Much of this remains experimental, with inconsistent skill interfaces, overlapping functionality, and uncertain long-term maintenance—raising concerns about fragmentation and reliability in production agent workflows.

Use Cases
  • Developers add video understanding to Claude Code agents
  • Security teams automate code reviews with LLM-augmented rules
  • Analysts turn technical books into instant AI agent references

Open Source Data Infrastructure Embraces Embedded, AI-Accelerated Local Processing 🔗

Projects shift from centralized clusters to lightweight, dependency-free tools running on-device with streaming and AI augmentation

Trenddata-infra
sqliteai/wastedeedy/qr-data-transferChatLab/ChatLabytsaurus/ytsaurusCoreBunch/Instatic+9 more

A clear pattern emerges in open-source data infrastructure: projects are prioritizing embedded, local-first execution over heavyweight cluster dependencies. sqliteai/waste exemplifies this by streaming activated weights of the Kimi K3 model directly from NVMe, enabling trillion-parameter LLM inference without loading full models into RAM—critical for edge deployment.

Similarly, ChatLab/ChatLab analyzes chat histories locally using AI, keeping sensitive data on-device while offering retrieval-augmented insights. ytsaurus/ytsaurus and ClickHouse/ClickHouse continue to power scalable analytics, but newer entrants like opengeos/GeoLibre bring GIS processing to browsers and notebooks, reducing reliance on backend servers. Even static site tools like CoreBunch/Instatic now integrate content databases and role management, blurring the line between CMS and lightweight data platforms. This shift reflects a broader move toward self-contained, AI-enhanced tools that minimize latency, maximize privacy, and reduce operational overhead—ideal for developers seeking autonomy without sacrificing capability.
The catch: Many of these tools trade maturity for novelty; streaming inference remains experimental, local AI analysis often lacks robust evaluation, and integrating AI into traditionally rigid systems (like CMS or GIS) risks creating fragile, hard-to-maintain pipelines that work in demos but struggle under real-world variability.

Use Cases
  • Developers run local LLMs on laptops without GPU memory limits
  • Teams analyze private chat logs with AI without exposing data
  • Field workers visualize geospatial data offline in disconnected environments

Quick Hits

worldmonitor A.Its AI aggregates news, tracks geopolitics and infrastructure, offering builders a unified real-time dashboard for proactive global situational awareness. 77.6k
qr-data-transfer Enables secure, offline data transfer via QR codes — ideal for builders needing lightweight, cross-platform file sharing without internet or complex setup. 278
interior Adds delightful, purposeful micro-interactions to React apps — perfect for builders refining UX with subtle, half-second feedback that feels intuitive and polished. 354
sakura-crossing Lets builders craft immersive, anime-style 3D neighborhoods in Three.js — zero textures, pure code-driven cel-shading for stylized, performative web scenes. 238
three.js The foundational JavaScript 3D library — empowers builders to create anything from simple models to complex interactive worlds with WebGL, no engine lock-in. 114.1k
Beyond GitHub

The AI Wire

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

From the labs & arXiv

OpenClaw v2026.7.1 Overhauls Onboarding and Control UI for Personal AI 🔗

Release adds GPT-5.6 support, improves mobile apps, and refines multi-channel messaging across 25 platforms

openclaw/openclaw · TypeScript · ▲ 111 in 1d 8mo old · Latest: v2026.7.1

OpenClaw’s latest release streamlines setup with a guided onboarding flow that validates connections before saving and preserves progress if interrupted. The Control UI now organizes conversations side by side, surfacing live Tasks, cost tracking, file transfers, and Gateway health alongside chat.

Official iOS, Android, and macOS apps received updates to voice input, permissions, offline reading, and connection recovery. Model support expands to include GPT-5.6, Tencent Hy3, and Meta Muse Spark 1.1, alongside existing compatibility with Anthropic, OpenAI, Google Gemini, xAI Grok, and OpenRouter. Users can sign in via OAuth for subscribed services like Claude Pro or ChatGPT Codex instead of managing raw API keys. The Gateway runs as a daemon via launchd, systemd, or Windows Scheduled Task, with a native Windows Hub companion app for tray control and local MCP mode. Over 3,000 contributions from 532 developers shaped this release, targeting refinements in Telegram, Slack, Discord, and Apple Messages handling, plus fixes for crash loops, scheduled work, and workspace terminals.
The catch: Despite broad channel support, OpenClaw remains a single-user assistant by design—no multi-user or team collaboration features exist, limiting its utility in shared or organizational workflows.

Previously in The Times “covered” — Jul 31

Use Cases
  • Developers running a private AI assistant on personal devices
  • Power users managing cross-platform chats via WhatsApp, Signal, or Telegram
  • Individuals seeking local, always-on AI with full data ownership and OAuth login

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

More Stories

GitHub repo centralizes AI system prompts for security research 🔗

Aggregates prompts from 25+ AI tools despite 157 open issues and aging codebase

x1xhlol/system-prompts-and-models-of-ai-tools · Unknown · 142.5k stars Est. 2025

The x1xhlol/system-prompts-and-models-of-ai-tools repository compiles system prompts, internal tools, and model details from AI coding assistants like Cursor, Devin, and Replit. It serves as a reference for understanding how these tools operate and identifying potential prompt injection risks.

Maintained with near-daily commits, the project includes warnings about securing AI data and promotes ZeroLeaks for vulnerability scanning. The catch: With 157 open issues and no clear maintenance structure, the accuracy and completeness of prompts for newer tool versions remain uncertain.

Use Cases
  • Security researchers analyzing prompt injection vulnerabilities
  • Developers comparing AI coding assistant behaviors
  • Startups auditing third-party AI tool configurations

Source: x1xhlol/system-prompts-and-models-of-ai-tools — based on the project README.

Open WebUI v0.11.0 Redesigns Interface for Focused AI Chat 🔗

Narrower chat column and lighter typography aim to reduce visual clutter during extended sessions

open-webui/open-webui · Python · 147.5k stars Est. 2023

The latest release of Open WebUI overhauls its user interface with a narrower conversation pane, refined spacing, and consistent menus across chat and admin panels. Built in Python, the self-hosted platform continues to support Ollama, OpenAI-compatible APIs, and RAG-based inference for local AI workflows.

Administrators retain granular RBAC controls, while users gain access to Notes and Channels for contextual collaboration. Despite active development—396 open issues and near-daily commits—the redesign prioritizes aesthetics over new backend capabilities. The catch: The interface changes may disrupt muscle memory for users accustomed to the prior layout, with no option to revert to the classic view.

Previously in The Times “covered” — Jul 25

Use Cases
  • Developers testing local LLMs via Ollama
  • Teams sharing AI-assisted notes in real time
  • Admins enforcing role-based access to model endpoints

Source: open-webui/open-webui — based on the README and release notes.

GitHub repo offers 28 production-grade GenAI agent tutorials for enterprise deployment 🔗

Recent commits show active maintenance with new Docker and GPU scaling labs added

NirDiamant/agents-towards-production · Jupyter Notebook · 21.2k stars Est. 2025

NirDiamant/agents-towards-production delivers hands-on Jupyter notebook tutorials for building production-ready GenAI agents, covering stateful workflows, vector memory, real-time web search, Docker deployment, FastAPI endpoints, security guardrails, GPU scaling, and multi-agent coordination. The repo includes 17 video-lecture-and-lab modules, each with an AI assistant in Claude Code for guided learning.

Recent activity indicates ongoing updates, including enhanced deployment and scaling content. The catch: Despite its breadth, the project’s narrow focus on Python and specific tools like LangGraph may limit adoption in teams using alternative frameworks or polyglot stacks.

Use Cases
  • Developers building stateful GenAI agents with vector memory
  • Teams deploying AI agents via Docker and FastAPI endpoints
  • Engineers implementing multi-agent coordination and GPU-accelerated workflows

Source: NirDiamant/agents-towards-production — based on the project README.

Quick Hits

yolov5 Ultralytics YOLOv5 delivers fast, accurate object detection, segmentation, classification, training, and export in PyTorch — ideal for builders needing production-ready vision models. 57.8k
RAG_Techniques NirDiamant/RAG_Techniques provides hands-on Jupyter tutorials for advanced Retrieval-Augmented Generation methods, empowering developers to enhance LLM accuracy with external knowledge. 28.9k
GenAI_Agents NirDiamant/GenAI_Agents offers 50+ practical notebooks for building Generative AI agents — from simple chatbots to sophisticated multi-agent systems — accelerating AI application development. 23.6k
paperless-ngx Paperless-ngx transforms document chaos into order: scan, index, and archive files with powerful search and automation — a self-hosted, developer-friendly DMS for builders. 43.8k
supervision Roboflow Supervision supplies reusable, high-quality computer vision utilities — annotation, tracking, counting, and more — so you spend less time coding and more time deploying. 48.5k

Scikit-Robot Unifies Robot Visualization and Motion Planning in Pure Python 🔗

Recent update adds Viser-based web interface with real-time IK and collision-aware trajectory planning

iory/scikit-robot · Python · 163 stars Est. 2019

The iory/scikit-robot project has released a significant update integrating the Viser web viewer for interactive robot manipulation. Users can now launch a browser-based interface via skr viser that provides real-time joint sliders, drag-to-solve inverse kinematics, and obstacle placement for motion planning.

The framework supports URDF models and enables collision-free trajectory planning using augmented Lagrangian optimization, with visual feedback directly in the browser.

Built as a pure-Python library, scikit-robot avoids ROS dependencies while offering kinematics, path planning, and visualization tools through a unified skr CLI. It supports optional backends like PyBullet, Open3D, and Fast-Simplification for advanced simulation and mesh processing. The update streamlines workflows for developers prototyping robot behaviors without heavy middleware.

The catch: The project relies on community maintenance, with only two open issues but no recent commits beyond routine updates, raising questions about long-term support for evolving robotics standards like ROS 2 Iron or Gazebo Harmonic.

Previously in The Times “covered” — Jul 26

Use Cases
  • Robotics researchers testing IK algorithms interactively
  • Educators demonstrating motion planning in classrooms
  • Developers validating URDF models before hardware deployment

Source: iory/scikit-robot — based on the project README.

More Stories

UBC-Thunderbots Refine Robot Soccer AI with Updated CODEOWNERS 🔗

v1.4.0 release assigns maintainership for core RoboCup SSL software stack

UBC-Thunderbots/Software · C++ · 66 stars Est. 2018

The UBC-Thunderbots team pushed a minor update to their C++ robot soccer AI, updating the CODEOWNERS file to clarify maintenance responsibility for their RoboCup Small Size League software. The project, active since 2018, enables autonomous coordination of SSL robots through vision, strategy, and low-level control systems.

Despite steady development, 204 open issues highlight ongoing challenges in perception reliability and real-time decision-making under match conditions.
The catch: Scaling this research-grade system beyond controlled SSL environments remains unproven due to hardware-specific dependencies and limited documentation for external adaptation.

Use Cases
  • University researchers testing multi-robot coordination algorithms
  • Student teams preparing for RoboCup SSL competition
  • Engineers studying real-time embedded control in dynamic environments

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

Open-source toolkit simplifies CAD workflows with modular CAD automation 🔗

Open-source CAD toolkit streamlines design automation with modular, reusable components

earthtojake/text-to-cad · JavaScript · 12.3k stars 3mo old

The CAD Toolkit simplifies computer-aided design workflows through modular, reusable components for geometry generation, assembly, and export. Built in TypeScript, it supports parametric modeling and integrates with formats like STEP and STL.

Developers chain operations via a fluent API to automate repetitive design tasks. The toolkit emphasizes composability, letting users build complex assemblies from primitive shapes and transformations. It targets engineers and developers seeking to automate repetitive CAD tasks without relying on heavy GUI tools. The catch: Advanced parametric constraints and assembly conflict resolution remain underdeveloped compared to mature commercial CAD systems.

Previously in The Times “covered” — Jul 29

Use Cases
  • Mechanical engineers automate bracket and enclosure design
  • Developers generate STL files for 3D printing pipelines
  • Educators teach parametric design principles via code-driven modeling

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

Comma AI adds Rivian and Acura support to openpilot v0.11.1 🔗

New driver monitoring model and thermal policy improve comma four reliability

commaai/openpilot · Python · ▲ 10 in 1d Est. 2016

The latest openpilot release v0.11.

1 adds support for Rivian R1S/R1T 2025 and Acura MDX 2022-24 models, expanding compatibility to over 300 vehicles. It features an updated driver monitoring model, enhanced image processing for the driver-facing camera, and refined thermal management for the comma four hardware. Developers can install via openpilot.comma.ai on supported devices, with prebuilt branches recommended for stability. The project maintains ISO26262-aligned safety protocols, including software-in-the-loop testing. The catch: 149 open issues indicate ongoing challenges in edge-case handling and hardware integration across diverse vehicle platforms.

Previously in The Times “covered” — Jul 28

Use Cases
  • Engineers retrofit ADAS in compatible hybrid SUVs
  • Developers test custom perception models on real vehicles
  • Fleets upgrade older cars with open-source driver assistance

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

Quick Hits

kornia Kornia provides GPU-accelerated geometric computer vision tools for spatial AI, enabling robust image transformations and pose estimation in robotics pipelines. 11.3k
webots Webots offers a professional-grade robot simulator with physics, sensors, and actuators for testing and developing robotic systems in virtual environments. 4.5k
IsaacLab IsaacLab delivers a unified, modular framework for robot learning on NVIDIA Isaac Sim, streamlining reinforcement learning and policy training across diverse robotic platforms. 7.8k
newton Newton is a GPU-accelerated physics engine built on NVIDIA Warp, offering high-fidelity, real-time simulation tailored for robotic dynamics and contact-rich interactions. 5.3k
URDF-Studio URDF-Studio is a web-based visual tool for designing, editing, and validating robot URDF models with integrated motor libraries, AI assistance, and MuJoCo export. 447

Infisical Adds AI Agent Vault to Secure Credential Access in Automated Workflows 🔗

New feature brokers API access for AI agents without exposing real secrets to prompt injection risks

Infisical/infisical · TypeScript · ▲ 19 in 1d Est. 2022 · Latest: v0.162.15

Infisical’s latest release introduces Agent Vault, a proxy-based system that injects secrets into outbound AI agent requests just before forwarding, preventing credential leakage from prompt injection attacks. The feature addresses a growing concern as teams integrate LLMs into automation pipelines where agents need API access but must not retain long-lived keys.

Built on the platform’s existing secret sync and rotation capabilities, Agent Vault works with services like AWS, PostgreSQL, and RabbitMQ, aligning with Infisical’s broader goal of unifying secrets, certificates, and privileged access management. The update also includes improved machine identity templates in the UI, full secret path visibility in approval tooltips, and a new Helm chart for the Infisical NKP partner catalog. Underlying infrastructure sees refinements: PostgreSQL connection pooling is now configurable, and FIPS test image rebuilding has been scaled back to reduce CI overhead. These changes reflect a focus on operational stability and developer experience as the project nears v0.162.16.
The catch: Agent Vault adds a network hop and potential latency for agent-driven workflows, which may not suit high-frequency trading or real-time inference systems requiring sub-millisecond API calls.

Previously in The Times “covered” — Jul 29

Use Cases
  • Secure API keys for AI agents in CI/CD pipelines
  • Rotate database credentials for microservices automatically
  • Sync secrets across GitHub, Vercel, and AWS environments

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

More Stories

OpenZeppelin Contracts v5.7.0 tightens EIP712 string limits for proxy safety 🔗

Breaking changes enforce 31-byte name/version fields to prevent storage clashes in upgradeable systems

OpenZeppelin/openzeppelin-contracts · Solidity · 27.2k stars Est. 2016

OpenZeppelin Contracts v5.7.

0 drops storage fallbacks for long EIP712 name/version values, requiring both to fit in a 31-byte ShortString or revert. This ensures domain consistency behind proxies without initializers. The release also renames ERC2771ForwarderFailureInAtomicBatch and updates Governor queueing errors. Deprecations replace at() with pos() in Checkpoints and enumerable collections. New BlockHeader and Create3 libraries aid header verification and deterministic deployment. Despite 360 open issues and a decade of use, the library remains foundational for ERC20/ERC721 and role-based access.
The catch: Developers must audit contracts manually when upgrading past v4.x due to assumed storage layout incompatibility in major versions.

Use Cases
  • Devs deploy ERC20 tokens with audited, gas-optimized implementations
  • Teams build upgradeable DAOs using Governor and role-based access control
  • Projects generate standard-compliant NFTs via ERC721 and ERC1155 templates

Source: OpenZeppelin/openzeppelin-contracts — based on the README and release notes.

Trickest/cve aggregates public exploit code for known vulnerabilities 🔗

Automated workflow pulls PoCs from GitHub, HackerOne, and CVE references

trickest/cve · HTML · ▲ 2 in 1d Est. 2022

The trickest/cve project maintains a continuously updated repository of proof-of-concept exploits for nearly all publicly disclosed CVEs. It uses automated scripts to gather PoCs via reference analysis, HackerOne reports, and GitHub searches, then merges results while filtering false positives.

Output includes organized markdown files, version-specific badges via shields.io, and a searchable HTML table for quick lookup. The project runs on HTML templates and relies on tools like ffuf and custom regex to validate findings. The catch: Automation may miss niche or privately shared exploits, and false positives require manual blacklist tuning despite ongoing refinement.

Use Cases
  • Security teams validate patch urgency using actual exploit code
  • Penetration testers build exploit libraries for red-team operations
  • Developers check dependencies against known vulnerable versions with PoCs

Source: trickest/cve — based on the project README.

Quick Hits

opennhp OpenNHP/opennhp delivers a lightweight, cryptography-driven Zero Trust toolkit to secure infrastructure, apps, and data in AI-native environments. 13.8k
juice-shop juice-shop/juice-shop provides a modern, intentionally insecure web app for hands-on learning and testing of real-world OWASP vulnerabilities. 13.6k
web-check lissy93/web-check is an all-in-one OSINT powerhouse that automatically gathers and analyzes website metadata, tech stacks, and security exposures. 34.3k
setup-ipsec-vpn hwdsl2/setup-ipsec-vpn lets you deploy a secure, multi-protocol IPsec VPN server in minutes across Linux and Raspberry Pi with full client configs. 28.3k
Reverse-Engineering mytechnotalent/Reverse-Engineering offers a free, deep-dive tutorial covering reverse engineering across x86, x64, ARM, AVR, and RISC-V architectures. 14k
hosts StevenBlack/hosts aggregates and extends trusted hosts files to block ads, malware, porn, and social media with optional category extensions. 30.8k

Bun v1.3.14 sharpens JavaScript toolchain with faster installs and HMR gains 🔗

Release improves dependency resolution and hot module replacement for React and Vue apps

oven-sh/bun · Rust · ▲ 24 in 1d Est. 2021 · Latest: bun-v1.3.14

Bun v1.3.

14 refines its all-in-one JavaScript toolkit with targeted upgrades to the package manager and bundler. The bun install command now resolves peer dependencies more aggressively, reducing duplicate packages in monorepos and cutting install times by up to 30% in benchmarked workflows. Hot Module Replacement (HMR) received a stability pass for React 19 and Vue 3.4, addressing edge cases where component state incorrectly persisted during rapid edits.

The release also expands Bun’s plugin system, allowing developers to intercept file transformations via a new onLoad hook — useful for custom loaders like Markdown-to-JSX or SVG-to-react-component pipelines. Windows arm64 support exited beta, with prebuilt binaries now matching x64 performance on Surface Pro X and similar devices.

Bun remains a single executable (bun) combining runtime, bundler, test runner, and package manager, powered by JavaScriptCore and written in Rust. It continues to promise sub-10ms startup times and lower memory usage than Node.js, though real-world gains vary by app size and dependency tree depth.

The catch: Bun’s Node.js compatibility layer still lags on native addons and certain worker_threads behaviors, requiring polyfills or avoided in performance-sensitive backend services.

Previously in The Times “covered” — Jul 31

Use Cases
  • Frontend teams accelerating React dev cycles with instant HMR
  • CLI tool creators bundling TypeScript scripts into single binaries
  • Full-stack developers replacing npm, Jest, and esbuild with one tool

Source: oven-sh/bun — based on the README and release notes.

More Stories

Rust Kernel Powers Near-Instant Code Graph Sync for AI Agents 🔗

v1.5.0 cuts graph updates to under a second, even on massive repos

colbymchenry/codegraph · C · 63.9k stars 6mo old

CodeGraph's latest release replaces its parser with a native Rust kernel, enabling live code graph updates in under a second for repositories up to 27,000 files. The engine adapts to available cores and memory, indexing the Linux kernel on a 2-core VPS in 12 minutes — down from 26.

It auto-syncs with Claude Code, Cursor, Gemini, and other agents via MCP, delivering surgical context with fewer tokens and zero external calls. All parsing is 100% local, with fallbacks for unsupported platforms or syntax errors.
The catch: The project has 378 open issues and relies on a complex Rust binarily on prebuilt kernels, leaving custom platforms unsupported.

Use Cases
  • AI agents get real-time codebase context during editing
  • Dev teams reduce tool calls with pre-indexed dependency graphs
  • Large monorepos maintain fast incremental updates on modest hardware

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

Quick Hits

lazygit LazyGit simplifies Git workflows with an intuitive terminal UI, letting builders manage repositories faster without memorizing complex commands. 80.9k
lede LEDE provides a lightweight, customizable Linux distribution for embedded devices, empowering builders to create efficient network routers and IoT gateways. 31.6k
rustlings Rustlings offers bite-sized, hands-on exercises that rapidly build Rust fluency through guided practice in reading and writing idiomatic code. 63.7k
traefik Traefik acts as a dynamic, cloud-native reverse proxy and load balancer that auto-discovers services and routes traffic with zero config overhead. 64.2k
cc-switch CC-Switch unifies access to multiple AI coding agents in one cross-platform desktop app, streamlining workflows for builders using Claude, Codex, Grok, and more. 123.1k
act Act lets developers run GitHub Actions locally in a real containerized environment, enabling fast debugging and testing without pushing to CI. 71.3k
ClickHouse ClickHouse delivers blazing-fast real-time analytics on massive datasets using columnar storage and vectorized query execution, ideal for high-performance data apps. 49k

Photobooth App Adds Rclone Sync for Seamless Multi-Device Backup 🔗

New plugin replaces manual setup with cross-platform automatic file transfers via rclone-bin-api

photobooth-app/photobooth-app · Python · 303 stars Est. 2022 · Latest: v8.7.0

The photobooth-app now includes an Rclone Synchronizer plugin that automates backup of photos, GIFs, and collages across devices. Built on the new rclone-bin-api PyPI package, it delivers a recent version of Rclone for Windows, Linux, and macOS—including ARM—without manual installation.

This replaces fragmented sync methods like QR share, FTP/Nextcloud, and USB filetransfer with a unified tool.

Backend improvements also enhance reliability: services can now signal permanent crashes (e.g., bad config) to prevent futile recovery loops, and TCP keepalive is enabled in multicamera setups via pynng to avoid dropped frames during long sessions. The frontend saw minor updates, including renaming the download portal to "sharepage" and visual refinements.

The app remains a full-featured DIY photobooth solution, supporting DSLR, Raspberry Pi (via Picamera2), and webcams for stills, boomerangs, collages, and 3D wigglegrams. It integrates with WLED for countdown LED rings and runs on Linux, Raspberry Pi, and Windows under an MIT license.

The catch: The Rclone sync plugin does not yet replicate all features of legacy sync options, so users relying on niche workflows (e.g., custom FTP scripts or USB automation) may need to retain older methods during transition.

Previously in The Times “covered” — Jul 28

Use Cases
  • Wedding planners automate photo backup to cloud storage
  • Raspberry Pi builders create synchronized multi-camera rigs
  • Event organizers share instant galleries via local network sharepage

Source: photobooth-app/photobooth-app — based on the README and release notes.

More Stories

Wearable terminal dispatches AI agent tasks without phone or app 🔗

PIN hardware now ships at $129, Ring enters private beta for invisible control

Nexting-ai/nexting · C · 424 stars 4mo old

Nexting lets users tap or speak to dispatch tasks to Claude Code, Codex, or OpenClaw agents directly from wearable hardware. The PIN version clips to collars and ships now; the Ring form factor is in private beta.

Firmware v2.2.4 adds mandatory hardware_id BLE characteristics for app binding, requiring flash on pre-June 2026 devices. Setup takes under a minute per agent via background daemons on host machines.
The catch: Agents must run locally or via managed Nexting Pro — no direct cloud model access.

Use Cases
  • Developers dispatch coding tasks mid-conversation
  • Engineers trigger infrastructure checks hands-free
  • Researchers query local models while prototyping hardware

Source: Nexting-ai/nexting — based on the README and release notes.

Kactus2 Adds Dark Theme and Memory Visualization in v3.14.1 🔗

Updated to Qt 6.6.2 with improved Linux builds and address block editing

kactus2/kactus2dev · C++ · 258 stars Est. 2016

Kactus2 v3.14.

1 upgrades to Qt 6.6.2, adding dark theme support that adapts to system settings on Linux and forces Fusion style on Windows. Users can now edit and visualize stride in address blocks and registers, with multi-dimensional memory items replicated in the visualizer. Linux build processes were refined, and bugs in portmap and address block validation were fixed. The tool remains focused on IP-XACT-based SoC design, enabling hardware reuse and HDL generation from component configurations.
The catch: Active development is sparse, with 34 open issues and commits slowing — last push was over a year ago despite the 2026 timestamp suggesting possible clock skew.

Use Cases
  • Hardware engineers reuse IP-XACT components in SoC designs
  • Developers generate HDL headers from configured IP instances
  • Teams integrate hardware and software via memory map visualization

Source: kactus2/kactus2dev — based on the README and release notes.

OpenC3 Cosmos Adds Script Runner History and Lifecycle Controls 🔗

New per-plugin UV environments and API enhancements improve test automation for embedded systems

OpenC3/cosmos · Ruby · 234 stars Est. 2022

OpenC3 Cosmos 7.2.

1 introduces Script Runner script history and lifecycle controls, enabling developers to track and manage test procedure execution within a graphical interface. The release adds per-plugin UV environments for isolated dependency management and new APIs like set_state_color and delete_limits_set for enhanced telemetry and command handling. Container support now includes Podman and registry login, streamlining deployment workflows. These updates strengthen Cosmos’s role in automating embedded systems testing across TCP/IP, UDP, and serial interfaces.
The catch: With 313 open issues and a Ruby-only core, teams seeking multi-language extensibility or lower maintenance overhead may find the ecosystem limiting.

Previously in The Times “covered” — Jul 26

Use Cases
  • Engineers automate spacecraft subsystem test sequences
  • Technicians validate IoT device telemetry limits in real time
  • Developers script hardware-in-the-loop validation for edge devices

Source: OpenC3/cosmos — based on the README and release notes.

Quick Hits

librealsense librealsense provides a cross-platform SDK for Intel RealSense depth cameras, enabling real-time 3D perception, object tracking, and spatial mapping for robotics and AR/VR applications. 8.9k
gdsfactory gdsfactory streamlines hardware design with Python, letting builders rapidly prototype photonic chips, PCBs, and 3D-printable components through intuitive, reusable code — turning complex layouts into manufacturable designs fast. 999
stack-chan stack-chan delivers an adorable, programmable M5Stack robot with expressive animations and sensor interactions, letting builders create engaging, kawaii-powered embedded projects with minimal code. 1.6k
FlightTracker FlightTracker turns a Raspberry Pi into a live aviation dashboard, displaying real-time aircraft positions, weather, and satellite data on a dot matrix display for immersive, desk-friendly flight monitoring. 183
PipelineC PipelineC extends Python with automatic pipelining constructs, letting hardware designers write high-level code that compiles to optimized, pipelined RTL — accelerating FPGA development without manual timing tweaks. 740

Egui 0.35.0 Adds Inspection Protocol for Agent-Driven GUI Testing 🔗

New egui_mcp crate enables AI agents to interact with and debug running egui applications

emilk/egui · Rust · ▲ 14 in 1d Est. 2019 · Latest: 0.35.0

The latest release of egui, version 0.35.

0, introduces an inspection protocol that allows external tools to read and influence the GUI state of running applications. Built into the egui_inspection crate, this feature exposes the accessibility tree via a network interface when launched with EGUI_INSPECTION=1, listening on port 5719. The protocol lets agents observe widget hierarchies and inject events, enabling automated testing and debugging without modifying app code.

A first consumer of this protocol is egui_mcp, a Model Context Protocol server developed by Rerun. It allows AI agents to "see" and manipulate egui interfaces — useful for reproducing bugs, verifying fixes, or exploring UI behavior in real time. The integration requires minimal setup: enable the inspection feature in eframe and run the app with the environment flag.

This shift toward observability and agent interoperability reflects egui’s evolution beyond a simple immediate-mode GUI into a platform for intelligent tooling. While still lightweight and dependency-free at its core, egui now supports advanced workflows for developers building complex, interactive Rust applications across web and native targets.

The catch: The inspection protocol adds network surface area and is opt-in, meaning agents cannot inspect apps by default — a limitation for zero-config debugging but a necessary trade-off for security and performance in production builds.

Previously in The Times “covered” — Jul 28

Use Cases
  • Debug game UIs with AI agents in real time
  • Automate accessibility testing of Rust GUIs
  • Inspect widget state during live app development

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

More Stories

GDQuest updates GDScript learning app with syntax highlighting fixes 🔗

Latest release resolves BBCode rendering issues in Godot 4 tutorial lessons

GDQuest/learn-gdscript · GDScript · 2.7k stars Est. 2021

GDQuest’s learn-gdscript project released version 1.5.

2, fixing symbol loss in code examples and improving BBCode string and number highlighting. The browser-based tool teaches GDScript fundamentals for Godot 4 game development through interactive lessons. It remains free and open source, with desktop builds available on Itch.io for better performance. The catch: The app intentionally avoids advanced programming depth, serving only as a beginner’s stepping stone before tackling full Godot 4 curricula.

Use Cases
  • New coders learning GDScript basics in-browser
  • Game dev beginners practicing Godot 4 scripting concepts
  • Educators assigning introductory programming exercises

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

Minecraft Shaderpack Revelation Balances Realism With Performance Controls 🔗

Built for Iris 1.7.0+, it uses GLSL shaders to enhance lighting without OptiFine support

HaringPro/Revelation · GLSL · 604 stars Est. 2024

Revelation is a physically based shaderpack for Minecraft: Java Edition that delivers atmospheric visuals through OpenGL 4.0+ rendering.

It supports Voxy Distant Horizons, Super Resolution, and HDR mods while integrating NASA star maps for celestial realism. Recent commits show ongoing debugging and AMD GPU testing, indicating active maintenance despite its 2.2-year age. The project avoids OptiFine, relying instead on Iris for compatibility, which may limit users on older mod setups. The catch: Its dependence on recent Iris versions and lack of OptiFine support could exclude players on legacy or lightweight modpacks.

Use Cases
  • Enhance Minecraft visuals with realistic lighting and atmospheric effects
  • Combine with distant horizon mods for extended render depth
  • Utilize NASA-derived star maps for accurate night sky rendering

Source: HaringPro/Revelation — based on the project README.

Quick Hits

SPIRV-Cross SPIRV-Cross enables developers to reflect on and decompile SPIR-V shaders into readable GLSL, simplifying cross-platform graphics debugging and integration. 2.5k
comedot Comedot provides a Godot-based template and component framework to accelerate 2D game development with clean, reusable architecture. 505
MeltyTool MeltyTool lets builders batch-view and extract assets from N64, GameCube, 3DS, and PC games, streamlining reverse-engineering and modding workflows. 198
luanti Luanti offers an open-source voxel engine with intuitive modding tools, empowering creators to build and share custom games easily. 13.4k
boardgame.io Boardgame.io handles state synchronization and networking for turn-based games, letting developers focus on gameplay rather than multiplayer infrastructure. 12.4k
imgui Dear ImGui: Bloat-free Graphical User interface for C++ with minimal dependencies 75.4k
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