Design
Preset
Background
Text
Font
Size
Width
Account Pricing

The Git Times

AI Models
Claude Opus 5 $25/M GPT-5.6 Luna $1.20/M Gemini 3.1 Pro Preview $12/M Grok 4.6 $6/M DeepSeek V4 Pro 0813 $1.98/M Qwen3.8 2.4T A95B $6/M Kimi K3 $15/M
Full Markets →

Orc: Orca Enables Parallel Agent Orchestration Across Worktrees for Developer Teams 🔗

Combines multiple AI coding agents in isolated git worktrees with mobile monitoring and remote execution capabilities

stablyai/orca · TypeScript · ▲ 58.1k in 1d 5mo old · Latest: v1.4.193

Orca lets developers run several AI coding agents—like Codex, Claude Code, or OpenCode—side by side, each in its own git worktree, so teams can fan out a single prompt, compare results, and merge the best output. Built in TypeScript, it provides a unified interface with Ghostty-class terminals featuring WebGL rendering, infinite splits, and persistent scrollback.

Users can interact with live Chromium windows to inject HTML, CSS, and screenshots into agent prompts, review PRs and issues in-app, and drop comments on diffs that get shipped back to agents. The Orca Monitor mobile app (iOS/Android) delivers real-time notifications when agents finish or need input, enabling follow-ups from anywhere. Remote agent execution is supported via VPS with auto-reconnect, port forwarding, and full file/terminal access. Agents can also drive Orca through CLI commands like orca worktree create and snapshot, enabling workflow automation. Despite rapid daily updates, the project carries 4,894 open issues, indicating ongoing stability and scalability challenges under heavy parallel agent loads.

The catch: High agent concurrency increases token usage and infrastructure costs, and the mobile app’s functionality remains dependent on stable desktop/VPS backends, limiting true standalone mobility.

Why this leads today Orca enables developers to run any coding agent with their own subscription across desktop, mobile, and VPS, unifying AI agent orchestration to scale individual productivity to fleet-level automation.

Use Cases
  • Compare agent outputs on complex refactoring tasks
  • Monitor and steer coding agents from mobile devices
  • Automate agent-driven workflows via CLI scripting

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

More on the Front Page

Rust-based DBX adds SQLite SSH access for remote file querying 🔗

Enables direct editing of remote SQLite files via temporary worker uploads without local copying

t8y2/dbx · Rust · ▲ 17.6k in 1d 4mo old

The DBX database client now supports SSH tunneling for SQLite connections, letting users query and modify remote .db files on Linux hosts through ephemeral workers that handle read/write operations in place.

This eliminates the need to download entire database files for local editing, improving workflow efficiency for distributed environments. Built in Rust, the 20 MB tool remains dependency-free, offering CLI, desktop, Docker, and MCP Server modes with AI-assisted SQL generation across 90+ database systems. Recent updates also enhance Elasticsearch deep pagination and add Xugu task management integration.
The catch: Open issues exceed 1,100, indicating ongoing stability and feature maturity challenges despite rapid development.

Use Cases
  • Developers querying remote SQLite via SSH without file transfer
  • Teams standardizing database access across CLI, desktop, and Docker
  • AI-assisted SQL generation for MySQL, PostgreSQL, and MongoDB workflows

Source: t8y2/dbx — based on the README and release notes.

OpenCodex Lets Any LLM Power Codex and Claude Code Tools 🔗

Local proxy routes requests to 40+ LLMs while managing ChatGPT account pools for auth

lidge-jun/opencodex · TypeScript · 12.7k stars 2mo old

OpenCodex acts as a lightweight TypeScript proxy that translates OpenAI Codex's Responses API to work with any LLM—Claude, Gemini, Grok, DeepSeek, Ollama, or custom endpoints. It enables Claude Code, Codex CLI, Claude Desktop, and Grok Build to run models beyond their native providers, preserving workflows while swapping the underlying "brain.

" The tool includes a web dashboard at localhost:10100 for configuring providers, managing model picks, and routing Codex sessions across a pooled set of ChatGPT accounts to avoid quota limits. Existing threads stay pinned to their originating account for continuity.
The catch: With 121 open issues and rapid development, stability for enterprise-scale or long-running agent workflows remains unproven despite active maintenance.

Use Cases
  • Developers run Codex CLI with local Llama 3 via Ollama
  • Teams balance ChatGPT API usage across multiple accounts automatically
  • AI agents use Claude Code interface powered by Gemini Pro 1.5

Source: lidge-jun/opencodex — based on the README and release notes.

DLSS5-Feeder Enables Neural Upscaling in DLSS-Less 32-Bit Games 🔗

It feeds ReShade depth and motion vectors to DLSS 5 via a private D3D12 device for temporal rendering.

jlrouzies-fr/DLSS5-Feeder · C++ · 370 stars 1d old

DLSS5-Feeder lets builders run NVIDIA's DLSS 5 neural rendering in 32-bit games lacking native DLSS support by synthesizing a DLAA contract from ReShade-provided depth and motion vectors. It works across D3D9, D3D11, D3D12, and Vulkan, using a cross-process host to bypass the 64-bit-only renodx-dlss add-on.

The tool injects into ReShade’s effect chain, runs a real DLSS evaluate, and returns the neural-upscaled frame. OpenGL support was added in v0.7.0, though only tested in one 32-bit title so far.
The catch: 64-bit OpenGL interop remains unproven in practice, with no confirmed game tests and unresolved NGX hook behavior in ReShade’s opengl32.dll context.

Use Cases
  • Upscale 32-bit DirectX 9 games with DLSS 5 quality
  • Add temporal anti-aliasing to Vulkan titles without native DLSS
  • Enable neural rendering in OpenGL games via ReShade layer injection

Source: jlrouzies-fr/DLSS5-Feeder — based on the README and release notes.

Rust-Powered Microduck Robot Walks, Learns, and Self-Recovers Using Neural Policies 🔗

Fifteen servos run on Rockchip RK3566 via daemons communicating over JSON-RPC Unix sockets

pollen-robotics/microduck · Rust · 4.3k stars 1mo old

The pollen-robotics/microduck project delivers a 25 cm, 800 g biped duck robot driven by reinforcement learning policies in Rust. Its brain consists of seven daemons—robotd for control, updaterd for signed releases, configd for Wi-Fi, btd for Bluetooth, padd for gamepad input, mediad for WebRTC camera streaming, and tofd for depth sensing—all communicating via a shared JSON-RPC contract over Unix sockets.

Neural policies trained in MuJoCo and PPO are exported to ONNX and executed at 50 Hz, enabling walking, object pickup, self-righting, ball-kicking, and voice quacking. Users can switch drive modes via D-pad or attach wheels for alternative locomotion.
The catch: The robot relies on a specific Rockchip RK3566 board, limiting portability to other hardware without significant rework.

Use Cases
  • Researchers testing sim2real reinforcement learning transfer
  • Hobbyists building programmable bipedal robots
  • Educators demonstrating embodied AI and robot control systems

Source: pollen-robotics/microduck — based on the README and release notes.

Hindsight Agent Memory System Boosts Long-Term Long-Term Recall Accuracy 🔗

Benchmarks show state-of-the-art performance on LongMemEval with independent validation

vectorize-io/hindsight · Python · 22k stars 10mo old

Hindsight, a Python-based agent memory system, focuses on enabling agents to learn over time rather than just recall conversation history. It eliminates shortcomings of RAG and knowledge graphs, delivering top-tier performance on long-term memory tasks as validated by Virginia Tech and The Washington Post.

Used in production by Fortune 500 enterprises and AI startups, Hindsight integrates with 25+ LLM providers and offers coding agent support via Claude Code and Cursor. The system provides a local API at http://localhost:8888 and UI at http://localhost:9999, with continuous benchmark updates at benchmarks.hindsight.vectorize.io.
The catch: Despite strong benchmark results, real-world scalability and cost efficiency at enterprise scale remain unverified by third parties.

Use Cases
  • Enterprises deploying long-context AI agents
  • Startups building persistent LLM-powered applications
  • Developers enhancing coding agents with codebase memory

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

AI Agents Forge Modular Skills and Orchestration Layers 🔗

Open source shifts from monolithic models to composable agent ecosystems with shared standards

Trendai-agents
KKKKhazix/sun-style-writingsaurabhkumar8112/cyclomatic-complexity-skillstickerdaniel/linkedin-mcp-serverstablyai/orcavectorize-io/hindsighttt-a1i/archifyTHU-MAIC/OpenMAICK-Dense-AI/scientific-agent-skillsopenJiuwen-ai/jiuwenswarmcalesthio/OpenMontageworkweave/routervolcengine/OpenVikingbojieli/ai-agent-bookdiffusionstudio/editorcan1357/oh-my-piagentconnect-md/agentconnectayghri/i-have-adhdVoltAgent/awesome-agent-skillswang2122/sprix-sage-routerblader/humanizertinyhumansai/openhumanchuspeeism/dashi-ppt-skillCopilotKit/OpenBotUntrivial-ai/agent-orchestratorfuxicodex/Fuxichaitanyagiri/munder-difflinSpaceZephyr/creator-buddybacknotprop/plannotatorNVIDIA/cosmosrokbenko/quackdMrDavidAlv/Axioma_robotGlaube-TY/siyuan-homepageautowarefoundation/autowarecurl/curl

The open source AI agent landscape is rapidly coalescing around modular skill libraries and orchestration frameworks. Projects like K-Dense-AI/scientific-agent-skills offer 161 validated scientific capabilities—from literature review to experimental design—plugging into agents via Cursor or Claude Code.

Similarly, VoltAgent/awesome-agent-skills curates 1,000+ community-vetted skills for coding, writing, and data tasks, promoting interoperability across agent runtimes. Orchestration is maturing too: stablyai/orca provides a desktop and mobile ADE for managing fleets of parallel agents with cost-aware routing, while workweave/router (Go) directs prompts to optimal models in under 50ms, cutting LLM costs by 40-70%. Memory and context are being systematized—volcengine/OpenViking unifies agent memory, RAG, and skills in a self-evolving context database, and vectorize-io/hindsight focuses on agent memory that learns from interaction history. Even niche domains are agentized: calesthio/OpenMontage turns agents into video production studios with 12 pipelines and 700+ skills, and diffusionstudio/editor enables professional video editing through agentic workflows. The pattern is clear: agents are becoming composable, specialized, and orchestrated—less like chatbots, more like programmable workers.
The catch: Despite rapid growth, the ecosystem remains fragmented by competing standards (MCP, Agent Skills, A2A), with many skills poorly documented or untested at scale; orchestration tools often lock users into specific runtimes, and real-world agent reliability in complex, long-horizon tasks remains unproven outside controlled demos.

Use Cases
  • Scientists automate literature reviews using validated skills
  • Developers route prompts to cut LLM costs by half
  • Teams orchestrate fleets of coding agents for CI fixes

Rust Powers a Quiet Revolution in Practical Open Source Tools 🔗

Developers are choosing Rust for performance, safety, and portability in everyday utilities from AI agents to database clients

Trendrust-systems
jub0t/Concatsoirihiroka/shrimplypollen-robotics/microduckpnpm/pnpmJetBrains/kotlint8y2/dbxGaoshu705/QzoneArchiveAprilNEA/OpenLogiherdrdev/herdrarcboxlabs/arcboxfirecrawl/anydoctinyhumansai/openhumandeeplethe/utopiablock/buzzBigPizzaV3/CodexPlusPluselabit/robotmkrustdesk/rustdeskdenoland/denoaa-proxy/aa-proxy-rs

A clear pattern is emerging in open source: Rust is becoming the language of choice for building reliable, high-performance tools that solve real problems. Projects like pnpm demonstrate how Rust enables fast, space-efficient package management at scale.

t8y2/dbx delivers a 20MB cross-platform database client with AI assistance and MCP Server support across 90+ systems. firecrawl/anydoc converts complex document formats to clean Markdown using Rust core with language bindings. In AI infrastructure, arcboxlabs/arcbox provides isolated, OCI-compatible agent runtimes with sub-100ms boot, while herdrdev/herdr offers a runtime for coding agents. On the desktop, AprilNEA/OpenLogi remaps Logitech devices locally via HID++ without telemetry, and rustdesk/rustdesk enables self-hosted remote desktop access. Even niche tools like soirihiroka/shrimply (a shrimp-themed video editor) and pollen-robotics/microduck (a biped duck robot) show Rust’s accessibility for creative, embedded projects. This shift reflects a maturing ecosystem where safety, speed, and portability aren’t trade-offs but foundations.
The catch: Despite momentum, many Rust projects remain early-stage, with limited long-term maintenance, fragmented GUI libraries, and steep onboarding for contributors — raising questions about whether this wave will sustain beyond enthusiast and infrastructure niches into broad application dominance.

Use Cases
  • Developers build fast, safe package managers
  • Engineers deploy lightweight, AI-enhanced database clients
  • Creators make local-first, telemetry-free device utilities

Deep Cuts

Chinese Writing Agent Transforms Prompt Engineering Into Literary Craft 🔗

Enables Claude Code and Codex to generate nuanced, emotionally resonant prose

KKKKhazix/sun-style-writing · Unknown · ▲ 14 in 1d

KKKKhazix/sun-style-writing distills 19 years of Chinese literary pain into a prompt engineering framework for AI agents. It teaches models like Claude Code and Codex to produce writing with the depth of classic baiyueguang tropes — longing, subtlety, emotional precision — moving beyond generic outputs.

Builders can now steer agents toward stylistically rich, culturally grounded narratives without fine-tuning. The project bridges literary theory and agent skills, offering reusable patterns for evocative storytelling in code.

Use Cases
  • Fiction writers generating culturally authentic Chinese short stories
  • Game designers crafting emotionally layered NPC dialogue
  • Educators teaching AI to emulate classical Chinese literary techniques

Source: KKKKhazix/sun-style-writing — based on the project README.

Claude Skill Slices Cyclomatic Complexity Without Manual Refactoring 🔗

Enables AI agents to autonomously simplify nested conditionals and loops

The saurabhkumar8112/cyclomatic-complexity-skill project equips Claude with a targeted refactoring ability: analyze code, identify high-complexity functions, and rewrite them to reduce branching paths while preserving logic. It transforms dense, error-prone sections into cleaner, more testable structures—directly improving maintainability and reducing cognitive load.

Builders gain an automated ally in code quality, especially valuable in legacy systems or large codebases where manual refactoring is costly and risky. The skill operates within Claude’s agent framework, turning abstract complexity metrics into actionable, safe code changes.
The catch: It's early-stage, narrowly focused on complexity reduction, and lacks broad language support or integration testing.

Use Cases
  • Developers refactoring legacy JavaScript modules
  • Teams preparing for audit-ready Python codebases
  • Agents maintaining high-reliability Go microservices

Source: saurabhkumar8112/cyclomatic-complexity-skill — based on the project README.

Quick Hits

readest Readest delivers a cross-platform ebook reader with powerful tools and an intuitive interface for an elevated reading experience. 23.9k
trustmebro TrustMeBro bypasses LLM guardrails by feeding fabricated tool output to confuse model safety mechanisms. 339
Concat Concat offers a free, open-source CapCut alternative for video editing with no cost or restrictions. 836
gloomberb Gloomberb brings a full-featured finance terminal directly into your command line for real-time market analysis. 2k
CoomeRtool CoomerTool is a Python utility designed to assist with content filtering and management workflows. 523
shrimply Shrimply is a Rust-based tool that generates videos from minimal input — yes, even a shrimp could use it. 402
tokentab TokenTab parses session logs from Claude Code, Codex, and Gemini CLI to calculate usage costs by model, project, and day. 535
open-webui User-friendly AI Interface (Supports Ollama, OpenAI API, ...) 150.5k
Who shipped it

The Business Desks

Fresh on Hugging Face

Model Drops

The newest model releases builders are picking up right now.
Beyond GitHub

The AI Wire

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

From the labs & arXiv

OpenClaw's latest release adds cross-device session migration for distributed AI workflows 🔗

Users can now shift assistant sessions between local and cloud workers while preserving workspace state

openclaw/openclaw · TypeScript · ▲ 176 in 1d 9mo old · Latest: v2026.8.1

OpenClaw v2026.8.

1 introduces session portability, allowing users to move active AI workflows between paired devices or cloud instances without losing context. The feature saves session workspaces and project seeds, enabling resumption on warm machines—a shift from purely local execution to flexible distributed operation. This addresses a core limitation of personal AI assistants locked to single devices.

The release also enhances conversational utility with exact-text search across visible chat history and durable progress cards that survive reloads, tracking subagent edits across web and native interfaces. Interactive widgets can now be pinned to session dashboards, exported as images, and gated by network origin for secure internal tooling. Credential handling sees improvement via private request flows for sensitive tool invocations.

Built on TypeScript, OpenClaw runs a local Gateway that orchestrates models, tools, and channels like WhatsApp, Slack, and iMessage through a trusted/untrusted execution boundary. Installation provisions Node.js runtimes or integrates with existing setups via npm/pnpm/Bun, with onboarding verifying model access and launching the Control UI.

The catch: Session migration relies on paired device trust and compatible environments, limiting cross-platform mobility—users cannot yet shift sessions arbitrarily between, say, a Windows laptop and an iOS companion app without manual workspace reconciliation, and cloud worker integration remains experimental for production teams.

Previously in The Times “covered” — Aug 28

Use Cases
  • Developers migrate AI debugging sessions from laptop to cloud workstation
  • Teams share persistent assistant workflows across trusted devices
  • Engineers search past conversations to resume interrupted coding tasks

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

More Stories

TensorFlow 2.21 drops Python 3.9 and TensorBoard dependency 🔗

Release adds int2/4 quantization and JPEG XL support for lightweight ML deployment

tensorflow/tensorflow · C++ · ▲ 53 in 1d Est. 2015

TensorFlow 2.21.

0 removes support for Python 3.9 and decouples TensorBoard, streamlining the core framework. The update enhances tf.lite with int2/int4 types for SQRT, EQUAL, and slice operators, enabling tighter model compression. JPEG XL decoding is now available in tf.image, improving image pipeline efficiency. These changes target edge and mobile use cases where binary size and inference speed matter. Despite activity—2,989 open issues and daily commits—the project maintains a vast C++/Python ecosystem.
The catch: Dropping Python 3.9 may break legacy environments relying on older system packages or long-term support distributions.

Previously in The Times “covered” — Aug 25

Use Cases
  • Train vision models with JPEG XL input on cloud TPUs
  • Deploy int4-quantized networks to microcontrollers via TensorFlow Lite
  • Replace TensorBoard with custom visualization tools in ML pipelines

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

NVIDIA’s Cosmos 3 Unifies Text, Video, Audio, Actions in One World Model 🔗

Open omnimodal foundation model now supports robot policy learning and multimodal simulation

NVIDIA/cosmos · Jupyter Notebook · 11.7k stars Est. 2024

Cosmos 3 integrates language, image, video, audio, and action modalities into a single Mixture-of-Transformers architecture. It enables world understanding — analyzing scenes for causality and next actions — and world generation — producing synchronized video, sound, and motion from multimodal inputs.

Developers can use it for robot action prediction, embodied AI, and smart infrastructure simulation via Hugging Face models and NVIDIA inference tooling. The release includes Cosmos3-Nano (16B) and supports integration with Diffusers, vLLM-Omni, TensorRT-LLM, and NIM.
The catch: Despite strong benchmarks, real-world deployment at scale remains unproven, and 49 open issues hint at integration and usability gaps.

Use Cases
  • Robotics teams train policies using action-conditioned video rollouts
  • Autonomous vehicle simulators generate multimodal driving scenarios
  • Smart infrastructure planners model pedestrian and vehicle interactions

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

Quick Hits

LLMs-from-scratch Implements a ChatGPT-like LLM in PyTorch from scratch, guiding builders through core transformer mechanics step by step. 104.1k
julia Delivers high-performance numerical and scientific computing with Python-like syntax and compile-time speed. 49.1k
langchain Provides a modular framework for building AI agents with tools, memory, and reasoning capabilities. 145.4k
paperless-ngx Automates document workflows by scanning, indexing, and archiving files into a searchable, self-hosted knowledge base. 44.7k
ComfyUI Enables visual, node-based creation and deployment of diffusion models via a flexible GUI, API, and backend. 130.9k
diffusers 🤗 Diffusers: State-of-the-art diffusion models for image, video, and audio generation in PyTorch. 34.4k

quackd Turns Plain Language into Robot Actions for Two-Legged Microducks 🔗

New flock mode lets multiple simulated ducks coordinate via auction-based task division

rokbenko/quackd · Python · 79 stars 3d old · Latest: v0.3.0

rokbenko/quackd gives the Pollen Robotics Microduck a brain by connecting plain language goals to its existing skills—walk, turn, kick, scoop, look, and quack—through an LLM-driven planner. Users write tasks in `.

duck` files like "Find the ball and kick it" or "Patrol, and quack twice if you see someone," and quackd translates them into skill sequences using any LLM: Claude, OpenAI, Gemini, Grok, or local models via Ollama, vLLM, or llama.cpp. The system runs in a bundled simulator that mirrors real-world perception and safety layers, requiring no API key for local models.

The latest release, v0.3.0, introduces flock mode: multiple simulated Microducks share an arena and coordinate via an in-process message bus using typed signals (TASK, BID, CLAIM, ROLE, HB, RESULT). A deterministic Contract Net auction assigns tasks based on each duck’s camera distance estimate, with results logged to flock.jsonl. The simulator uses lockstep timing and per-duck noise streams to ensure reproducibility, and includes all four Microduck colorways with collision physics.

quackd needs Python 3.11+ and uv, with setup guided by quackd doctor. It honors the Microduck’s existing skill set, avoiding retraining by leveraging LLMs as planners over fixed action primitives.

The catch: Everything today runs in simulation; the physical Microduck ships at Christmas 2026, and no code has yet been tested on real hardware, leaving the sim2real gap unproven.

Use Cases
  • Developers testing LLM-agent control of legged robots
  • Educators demonstrating embodied AI in classroom simulators
  • Researchers studying multi-robot coordination via auction protocols

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

More Stories

Autoware adds NVIDIA Thor support for Jetson and DRIVE platforms 🔗

Release 1.9.0 integrates CUDA 13 and Hugging Face Lidar Centerpoint model

autowarefoundation/autoware · Dockerfile · 12k stars Est. 2015

Autoware 1.9.

0 bakes runtime dependencies into development images and migrates devcontainers to docker-compose with GUI forwarding. It now supports NVIDIA Thor (Jetson + DRIVE) on JetPack 7 via SBSA CUDA 13, enabling edge deployment on next-gen hardware. The update pulls the Lidar Centerpoint model from Hugging Face and adds version locking for reproducible builds. These changes streamline setup for developers targeting autonomous vehicle prototypes.
The catch: Despite frequent commits, 71 open issues persist, raising questions about stability in production-scale deployments beyond simulation.

Use Cases
  • Engineers simulate urban driving scenarios
  • Teams prototype perception stacks on Jetson
  • Researchers test planning algorithms in ROS2

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

OpenArm v1.1 boosts hardware reliability and teleoperation accuracy 🔗

Automated calibration and modular camera mounts reduce setup errors for researchers

enactic/openarm · MDX · 2.9k stars Est. 2024

The enactic/openarm project released version 1.1, focusing on hardware reliability and teleoperation stack improvements.

Key updates include automated zero-position calibration to eliminate manual setup errors and a modular camera mount base for Realsense D435 and D405 cameras, ensuring consistent data collection. These changes address community feedback on elbow control ambiguity in 7-DoF arms via a redesigned J5 casing with rubber band interface for smoother motion coupling. The open-source humanoid arm, priced at $6,500 for a bimanual system, supports teleoperation, imitation learning, and real-world data collection in contact-rich tasks. Active development continues with 322 forks and 16 open issues.
The catch: Reliance on specific hardware like Realsense cameras and ROS2 may limit accessibility for labs without existing infrastructure or funding for proprietary sensors.

Use Cases
  • Researchers training AI models via teleoperation in contact-rich environments
  • Labs collecting reproducible imitation learning data across global sites
  • Teams evaluating force-feedback algorithms for safe human-robot interaction

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

evo Benchmarks Robot Trajectories Across Formats With Precision 🔗

Updated CLI now supports ROS2 bagfiles and LaTeX plot export for SLAM validation

MichaelGrupp/evo · Python · 4.3k stars Est. 2017

MichaelGrupp/evo evaluates odometry and SLAM trajectories using TUM, KITTI, and EuRoC formats, plus ROS and ROS2 bagfiles via geometry_msgs topics. Its CLI enables alignment, scale adjustment, and association tuning for monocular SLAM, outputting plots, Excel tables, or LaTeX figures.

The tool prioritizes speed over pure Python alternatives and supports modular extensions for custom metrics. Last updated August 2026, it maintains active development with 796 forks and ongoing issue resolution.
The catch: Despite broad format support, evo does not implement dataset-specific evaluation protocols, requiring users to define their own benchmarks.

Use Cases
  • Robotics engineers comparing VIO and LiDAR SLAM trajectories
  • Researchers validating monocular SLAM scale recovery on EuRoC datasets
  • Developers automating trajectory analysis in ROS2-based navigation stacks

Source: MichaelGrupp/evo — based on the project README.

Quick Hits

BehaviorTree.CPP A C++ behavior trees library with built-in nodes and tools for creating modular, reusable AI decision-making systems in robotics and games. 4.2k
tlsfuzzer A Python-based SSL/TLS protocol test suite and fuzzer that identifies security vulnerabilities through automated, comprehensive protocol manipulation. 633
panda-py Python bindings enabling real-time, low-latency control of Franka Emika robots for research, automation, and industrial applications. 177
robotmk A Rust-powered integration that brings Robot Framework test automation into Checkmk monitoring for unified infrastructure validation. 59
newton A GPU-accelerated physics engine using NVIDIA Warp, designed for high-fidelity robotic simulation and research with real-time performance. 5.6k

OWASP WSTG v4.2 Adds GraphQL and HTTP Method Testing for Modern AppSec 🔗

Release includes new scenarios for API security flaws and session hijacking, addressing evolving web threats

OWASP/wstg · Python · ▲ 5 in 1d Est. 2017 · Latest: v4.2

The OWASP Web Security Testing Guide (WSTG) released version 4.2 with targeted updates for API and configuration testing.

New scenarios cover GraphQL API testing (WSTG-APIT-01), HTTP method overriding (WSTG-CONF-06), and Server-Side Request Forgery (WSTG-INPV-19). The guide now includes Test Objectives across all scenarios to clarify validation goals.

Additional updates address session security (WSTG-SESS-09), authorization bypass (WSTG-ATHZ-02), and local file inclusion edge cases (WSTG-INPV-11.1). Appendix F on leveraging browser developer tools was added to support hands-on testing. Several legacy sections were merged to reduce redundancy, including HTTP Verb Tampering into HTTP Methods testing and Stack Traces into Improper Error Handling.

Front matter and introductory chapters were revised, and Penetration Testing Methodologies (Chapter 3.8) updated to reflect current practices. The release focuses on refining existing content rather than overhauling structure, with versioned identifiers like WSTG-v42-INFO-02 ensuring stable references.

The catch: The guide remains documentation-only—no automated tests, scanners, or remediation scripts are included, requiring teams to manually implement scenarios in their workflows.

Previously in The Times “covered” — Aug 30

Use Cases
  • Penetration testers validate GraphQL endpoints for injection flaws
  • Security teams audit HTTP method configurations for bypass risks
  • Developers reference WSTG-SESS-09 to test session token protections

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

More Stories

Anthropic-Cybersecurity-Skills v1.3.0 adds 55 skills for AI agent red-teaming 🔗

New AI Security, Supply Chain, and Hardware domains expand offensive and defensive coverage

mukul975/Anthropic-Cybersecurity-Skills · Python · 31.8k stars 6mo old

The mukul975/Anthropic-Cybersecurity-Skills repository released v1.3.

0, growing from 762 to 817 structured cybersecurity skills for AI agents. It added three new domains: AI Security (12 skills) covering LLM red-teaming with garak and PyRIT, Supply Chain Security (5 skills) for SBOM and dependency confusion, and Hardware and Firmware Security (4 skills) including CHIPSEC UEFI audits. The release also incorporated MITRE F3 (Fight Fraud) as a sixth framework, bringing total mappings to MITRE ATT&CK (805 skills), NIST CSF 2.0 (804), and others. Skills follow the agentskios.io standard and work with Claude Code, GitHub Copilot, and 20+ platforms. The catch: The project includes offensive techniques like phishing and C2 frameworks, requiring explicit authorization for use—misuse risks legal and ethical violations.

Previously in The Times “covered” — Aug 25

Use Cases
  • Train AI agents to detect prompt injection in LLM applications
  • Automate SBOM generation for open-source dependency risk assessment
  • Simulate hardware-based attacks like UEFI bootkit hunting in lab environments

Source: mukul975/Anthropic-Cybersecurity-Skills — based on the README and release notes.

Sniffnet adds latency monitoring and CIDR blacklist support in v1.5.1 🔗

New release enhances network diagnostics with real-time connection timing and refined IP filtering

GyulyVGC/sniffnet · Rust · ▲ 12 in 1d Est. 2022

Sniffnet v1.5.

1 introduces latency measurement for network connections, letting users see round-trip times in real time. The update also adds support for CIDR ranges in imported IP blacklists, improving threat detection flexibility. Built with Rust and Iced, the tool remains cross-platform, offering packet capture, protocol identification, and traffic visualization. It now includes Hungarian and Sinhala translations, expanding accessibility.
The catch: Latency features depend on accurate timestamping from the underlying pcap library, which may vary across operating systems and network interfaces.

Previously in The Times “covered” — Aug 25

Use Cases
  • Network administrators diagnosing application-layer delays
  • Security analysts identifying suspicious traffic patterns
  • Developers optimizing local service communication performance

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

x64dbg Hotfix Resolves r8 Register Bug, Improves Dark Theme 🔗

May 2026 patch fixes trace reader faults and adds absolute path support for config files

x64dbg/x64dbg · C++ · 49.4k stars Est. 2015

The x64dbg debugger released a hotfix on May 27, 2026, addressing an r8 register modification issue that could corrupt analysis during reverse engineering sessions. It also patches a trace reader memory fault triggered by status changes and corrects dark theme label coloring for better visibility.

Users can now pass absolute paths to the -cf flag, streamlining custom configuration workflows. Built on C++ with plugin support for disassembly via Zydis and XEDParse, it remains a key tool for Windows binary analysis.
The catch: Active development relies on community contributions, with 574 open issues indicating ongoing stability and feature gaps despite regular commits.

Use Cases
  • Malware analysts inspecting Windows executables without source code
  • Reverse engineers debugging 32-bit and 64-bit binaries via plugin architecture
  • Security researchers conducting dynamic analysis for exploit development and CTFs

Source: x64dbg/x64dbg — based on the README and release notes.

Quick Hits

subfinder Enumerates passive subdomains quickly to help builders discover exposed assets without active scanning. 14.3k
bunkerweb Provides a cloud-native, open-source WAF to protect web apps from common threats with minimal setup. 10.9k
maigret Aggregates public data from 3000+ sites into a detailed profile using just a username for OSINT investigations. 37.2k
nuclei Scans applications, APIs, and infrastructure for vulnerabilities using community-driven YAML templates for fast, customizable detection. 30.9k
CheatSheetSeries Delivers concise, high-value security cheat sheets on critical appsec topics to guide secure development practices. 33k

ZML Unifies AI Inference Across Hardware With Zero Code Changes 🔗

Latest release adds Prometheus metrics and ARM64 CUDA support for production monitoring

zml/zml · Zig · ▲ 1 in 1d Est. 2024 · Latest: zml-smi-v0.3.0

ZML lets developers run any AI model on NVIDIA, AMD, Intel, TPU, or Trainium hardware using a single codebase, eliminating vendor lock-in. Built with Zig, MLIR, and Bazel, it compiles models directly to target accelerators without rewriting.

The recent v0.3.0 release enhances observability by adding a Prometheus endpoint, JSON output, and macOS host info to zml-smi, alongside fixing logo rendering in KDE Konsole and enabling ARM64 support for CUDA platforms. Developers can switch hardware targets via flags like --@zml//platforms:cuda=true or --@zml//platforms:neuron=true for AWS Trainium. Examples include LLMs (Llama 3.1, Qwen 3.5) and MNIST, with support for loading models from local paths, Hugging Face, or S3. Bazel manages builds and dependencies, requiring only Bazelisk for setup.

The catch: Despite broad hardware support, ZML remains early-stage with 47 open issues and limited real-world scale validation, raising questions about long-term stability and enterprise readiness.

Use Cases
  • Deploy Llama 3.2 on AMD GPUs without code changes
  • Run Qwen models on AWS Trainium via CLI flags
  • Benchmark inference latency across CPU and TPU backends

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

More Stories

Deno 2.9.6 Adds Clipboard API and Secure Text Handling for Desktop Apps 🔗

Latest patch improves desktop runtime stability and MIME type support for modern web components

denoland/deno · Rust · ▲ 13 in 1d Est. 2018

Deno 2.9.

6 introduces a clipboard API for desktop applications, enabling secure read/write access to system clipboard contents. The update also adds support for the 'text/x-component' MIME type, aligning with emerging web standards for reusable UI components. Desktop-specific fixes ensure proper handling of menu item properties, macOS bundle signatures, and Vite-based HMR dev servers without blocking the JS thread. These changes enhance Deno’s viability for building cross-platform desktop tools with TypeScript.
The catch: Despite active development, Deno lacks native support for Windows ARM64, limiting deployment on emerging hardware like Surface Pro X and Qualcomm-powered laptops.

Use Cases
  • Build secure CLI tools with TypeScript
  • Develop cross-platform desktop apps using web APIs
  • Run lightweight web servers with Deno Deploy integration

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

Quick Hits

lede Provides the LEDE source code for building lightweight, customizable Linux firmware for embedded devices. 31.6k
FFmpeg Offers the complete FFmpeg source for cross-platform audio/video encoding, decoding, and streaming capabilities. 63.8k
codex Delivers a terminal-based coding agent that assists with code generation, debugging, and task automation using AI. 120.4k
tmux Supplies the tmux source for a highly configurable terminal multiplexer enabling persistent, split-pane workflows. 48.9k
cc-switch Unifies access to multiple AI coding agents (Claude Code, Codex, etc.) in a single cross-platform desktop interface. 130.4k
rustdesk An open-source remote desktop application designed for self-hosting, as an alternative to TeamViewer. 122.2k
ghostty 👻 Ghostty is a fast, feature-rich, and cross-platform terminal emulator that uses platform-native UI and GPU acceleration. 60.5k

FlightTracker OS Simplifies Raspberry Pi Flight Display Setup for Builders 🔗

New installer script detects hardware and configures Pi model and auto-configures systemd service on boot

ColinWaddell/FlightTracker · Python · 196 stars Est. 2021 · Latest: v2.9.0

ColinWaddell/FlightTracker’s latest release v2.9.

0 streamlines deployment with a one-line installer that clones the repo, sets up the Python environment, and launches the flight display service automatically on Raspberry Pi OS (Lite). The script detects the Pi model and selects the correct 32-bit or 64-bit FlightTracker OS image, eliminating manual configuration steps. Builders can now run the tracker on a 64x32 RGB LED matrix using either FlightRadar24’s free tier or its paid API, with a new data source manager letting users prioritize feeds like satellite positions or local ADS-B. The update also adds a Status page showing provider health, fetch latency, and Pi system metrics—useful for debugging data pipelines. An official 3D-printable case, designed for the HUB75 matrix and compatible with all Pi models, is now available on Printables with OpenSCAD source and assembly instructions. A GPIO-connected LED can blink during data loading, configurable via the web UI under Hardware settings.

The catch: The project relies heavily on FlightRadar24’s API terms, which may restrict commercial use or impose rate limits on free tiers, and local ADS-B setup requires additional RTL-SDR hardware not included in the base guide.

Previously in The Times “covered” — Aug 30

Use Cases
  • Home builders display real-time flights on fridge-mounted LED matrix
  • Aviation enthusiasts track overhead aircraft with satellite and weather overlays
  • Educators demonstrate IoT data pipelines using flight and weather APIs

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

More Stories

Shibco/ableton-linux brings Ableton Live 12 and Push to Linux with full hardware support 🔗

The project enables low-latency audio, MIDI hot-plugging, and desktop integration via Wine and PipeASIO

shibco/ableton-linux · Shell · 457 stars 1mo old

Shibco/ableton-linux lets producers run Ableton Live 12—Intro, Standard, Suite, Lite, and Trial—on Linux without compromises. It supports Push 1, 2, and 3, with Move support coming soon.

The runtime enables MIDI controllers and audio interfaces to connect and disconnect while Live stays open, using low-latency PipeASIO for performance. Features include native file dialogs, HiDPI scaling, font reuse, and automatic dark/light theme detection. Installers now log errors instead of crashing and let users choose between multiple Ableton builds. Performance optimizations help Live and Max start faster and use fewer resources under Wine.
The catch: Current test builds require clearing Live’s saved state and removing Demo Songs to avoid crashes, and Ableton Index still shows error pop-ups.

Use Cases
  • Electronic musicians producing tracks on Linux using Ableton Live 12
  • Live performers using Push 3 with low-latency audio via PipeASIO
  • Developers testing Max for Live extensions on Linux with experimental SDK support

Source: shibco/ableton-linux — based on the README and release notes.

Ibex RISC-V Core Gains Bit-Manipulation Support in Latest Update 🔗

Verified configurations now include B-extension for embedded control workloads

lowRISC/ibex · SystemVerilog · 2k stars Est. 2017

Ibex, the lowRISC-maintained 32-bit RISC-V CPU core written in SystemVerilog, has added official support for the Bit Manipulation (B) extension in its verified configurations. The update, reflected in the latest commit, enables efficient bit-field operations critical for embedded control and sensor processing.

Ibex remains parametrizable, supporting I, E, M, C, and now B extensions, with verification status updated for select configurations targeting low-power IoT and edge devices. Synthesis reports show area efficiency under 25kGE in Yosys flows for base setups, though commercial flow estimates vary. The core has seen multiple tape-outs and is used in PULP-based platforms, but full verification of all B-extension interactions remains incomplete.
The catch: Full verification of the B-extension across all Ibex configurations is still marked amber, indicating experimental status and potential edge-case risks in safety-critical deployments.

Use Cases
  • Embedded controllers implementing sensor fusion algorithms
  • Low-power IoT devices requiring compressed instruction sets
  • Edge nodes executing bit-manipulation-heavy cryptographic primitives

Source: lowRISC/ibex — based on the project README.

Quick Hits

firmware This ESP32 firmware enables predatory device behaviors, offering low-level control for advanced hardware exploitation and customization. 6.6k
TuyaOpen TuyaOpen provides a next-gen AI+IoT framework for seamless integration of AI agents across T2/T3/T5AI, ESP32, and other hardware platforms. 1.8k
photobooth-app This open-source photobooth app combines Python backend logic with a modern Vue3 frontend to deliver a customizable, free photo capture experience. 315
firmware OpenIPC firmware offers a community-driven, open alternative to proprietary IP camera software, enhancing device control and transparency. 2.2k
FanCtrl FanCtrl automatically adjusts PC fan speeds based on system temperature, optimizing cooling performance and noise levels without manual intervention. 563
aa-proxy-rs AndroidAuto wired/wireless proxy 459

Rust GUI egui adds drag detection fix in latest patch release 🔗

Version 0.36.1 resolves widget interaction bug reported by community contributor

emilk/egui · Rust · ▲ 27 in 1d Est. 2019 · Latest: 0.36.1

The emilk/egui project released version 0.36.

1, patching a bug where Sense::drag incorrectly detected drags when clicking a widget positioned above another. The fix, contributed by @lucasmerlin in pull request #8396, improves interaction accuracy in immediate mode interfaces. egui remains a Rust-native option for building GUIs that run on web via WebAssembly and desktop platforms using eframe. Despite active maintenance—with commits as recent as 0 days ago and 2,116 forks—the library carries 1,139 open issues, indicating ongoing refinement.
The catch: While lightweight and portable, egui’s immediate mode paradigm may not suit developers preferring retained-mode GUI architectures for complex state-heavy applications.

Previously in The Times “covered” — Aug 25

Use Cases
  • Game developers building in-editor tools with Rust
  • Web app creators targeting Wasm and WebGL
  • Embedded systems engineers needing cross-platform GUIs

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

More Stories

O3DE 26.05.0 Release Adds Wwise Audio Gem Updates and CMake 3.24 Requirement 🔗

Builders must now use VS 2019 16.9.2+ and Git LFS for binary assets in the Apache 2.0-licensed engine.

o3de/o3de · C++ · 9.6k stars Est. 2021

The Open 3D Engine project released version 26.05.

0, updating its Wwise Audio Engine Gem and raising the minimum CMake requirement to 3.24.0. Developers building AAA games or simulations must install Visual Studio 2019 16.9.2 or newer and ensure Git LFS is configured to handle large binary files. The engine remains free under Apache 2.0 with no commercial fees, supporting Windows, Linux, and macOS for high-fidelity 3D worlds. Despite active commits, 3,492 open issues suggest ongoing stability and feature work.
The catch: The engine’s steep setup burden — requiring specific SDKs, redistributables, and manual caching — limits accessibility for small teams or rapid prototyping.

Previously in The Times “covered” — Aug 25

Use Cases
  • Game studios building cross-platform AAA titles without licensing fees
  • Simulation creators developing high-fidelity industrial or training environments
  • Cinematic teams producing real-time 3D visuals for film and virtual production

Source: o3de/o3de — based on the README and release notes.

Quick Hits

Greater-Flavor-Mod Enhances historical strategy gameplay with expanded flavor, province depth, and accuracy, and immersive world-building for a richer player experience. 239
Universal-x86-Tuning-Utility Empowers users to fully customize and tune x86 hardware performance with open, unrestricted, and powerful low-level control. 2.8k
OpenVic Provides the core foundation for the OpenVic project, enabling collaborative development of an open-source Victoria-inspired grand strategy game. 467
Sundial-Lite Offers a free, open-source GLSL implementation of the Sundial lighting effect for real-time visual enhancement in graphics applications. 152
glsl-shaders Delivers hand-converted GLSL shaders from libretro’s common-shaders, ensuring compatibility and visual fidelity where automated tools fail. 1.2k
SPIRV-Cross Enables cross-language shader reflection and SPIR-V disassembly, letting developers inspect and reuse Vulkan shaders in HLSL, GLSL, or Metal. 2.5k
ebiten A dead simple 2D game engine for Go 13.4k
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