Design
Preset
Background
Text
Font
Size
Width
Account Pricing Sunday, August 2, 2026

The Git Times

“We drive into the future using only our rearview mirror.” — 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 →

Oh-My-Pi Agents Run Dual Kernels to Sync Python and JavaScript Workflows 🔗

Persistent Bun and Python workers enable real-time data sharing without leaving the terminal session

can1357/oh-my-pi · TypeScript · ▲ 21.1k in 1d 7mo old · Latest: v17.2.4

Oh-My-Pi (can1357/oh-my-pi) is a terminal-based AI coding agent that integrates a persistent Python sandbox and a Bun worker through a loopback bridge, allowing both kernels to invoke the agent’s own tools—like read, search, and task—without context switching. Built as a fork of Mario Zechner’s Pi, it now supports over 40 LLM providers, 32 built-in tools, and deep LSP and DAP integration, aiming to deliver IDE-level functionality inside the terminal.

Its TUI enables seamless workflows: users can load a CSV in Python, visualize it via JavaScript, and resume edits—all within a single session. Edits are hash-anchored for precision, reads return summarized snippets instead of raw dumps, and model selection adapts dynamically to provider capabilities. The agent generates shell completions from live CLI metadata, ensuring accuracy across bash, zsh, and fish. Recent fixes in v17.2.4 resolved WebSocket transport bypasses in Codex V2 and improved tool-call state tracking to prevent UI telemetry from misrepresenting execution status during steering inputs. These updates refine reliability in multi-turn agent interactions, especially when using remote models via OpenRouter or Anthropic.
The catch: Despite its ambitious scope, the project carries 961 open issues and relies on a complex dual-kernel architecture that may introduce subtle synchronization bugs or resource overhead in constrained environments.

Why this leads today It offers a new class of terminal-based AI agent with hash-anchored edits, subagents, and cross-environment integration that directly streamlines developer workflows without requiring context switching.

Use Cases
  • Data scientists iteratively analyze CSVs with live Python-JavaScript bridges
  • Debuggers step through code using embedded DAP and LSP inspection
  • CLI power users chain LLM-assisted edits across platforms access AI agents via consistent completion systems

Source: can1357/oh-my-pi — based on the README and release notes.

Fresh on Hugging Face

Model Drops

The newest model releases builders are picking up right now.

More on the Front Page

Bento Turns Office Documents Into Self-Contained HTML Files That Edit Themselves 🔗

The single-file suite embeds viewer, editor, and assets, enabling true local-first workflows without installation or accounts.

nyblnet/bento · TypeScript · 3.5k stars 2w old

Bento reimagines office documents as single HTML files that carry their own editor, viewer, and presenter. Opening `Bento_Slides.

bento.html` in any modern browser launches a full-featured slide editor; saving rewrites the file’s embedded JSON data block using the File System Access API, preserving fonts, images, charts, and animations. No installer, no login, no cloud dependency—just a ~560 KB file that is both document and application. Recent fixes restore broken fade, slide, and zoom transitions, ensure embedded typefaces survive copy-paste between decks, and expand update notes to cover skipped versions. The project targets builders who reject lock-in: decks remain readable in 2036, and plain JSON at the file’s top allows agents like Claude Code or Cursor to edit decks directly via filesystem access—no plugins or APIs needed. Collaboration requires manual file sharing; real-time co-editing isn’t supported.
The catch: Bento sacrifices real-time collaboration and native app performance for radical portability and long-term accessibility, making it unsuitable for teams needing simultaneous editing or complex animations beyond its current scope.

Use Cases
  • Engineers sharing architecture decks via email or USB
  • Teachers distributing editable lesson slides to students
  • Travelers presenting offline without software installs

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

x4gKing/3x-ui-multi Solves Port Conflicts in Single-Port Container Deployments 🔗

Routes Tor and direct traffic through nginx on one public port using loopback binding

x4gKing/3x-ui-multi · JavaScript · 271 stars 0d old

The project fixes a critical port collision where nginx and Xray’s direct inbound both tried to bind 0.0.

0.0:8080, causing crash loops on single-port hosts like Railway. Now, nginx exclusively binds 0.0.0.0:3000, while the direct inbound uses 127.0.0.1:8080, with nginx reverse-proxying / and /direct to it. Traffic for both Tor and direct connections flows through the single public port. Configuration lives in config.json, including tunable Tor rotation (tor.rotate_seconds) and country lists, all managed via start.sh without external cron. The catch: The solution assumes a single external port forwarding model and may not suit multi-port or complex ingress environments requiring direct external access to both services.

Use Cases
  • Deploy privacy tools on Railway with one exposed port
  • Run Tor exit nodes alongside direct APIs in containers
  • Simplify ingress for geo-rotated proxies on PaaS platforms

Source: x4gKing/3x-ui-multi — based on the project README.

Solidity Arbitrage Bot Automates MEV Trades Between DEX Pools 🔗

Owner-controlled contract executes swaps via external Python script with whitelist management

MIgHTy-alIeN/Ethereum-Trading-Bot · Solidity · 1.8k stars 2w old

MIgHTy-alIeN/Ethereum-Trading-Bot is a Solidity smart contract that scans for and executes arbitrage opportunities across DEX pools in a single transaction. It connects to an external automation script—typically Python—to control operations like executeArbitrage(), quickSwap(), and balance withdrawals.

Owners fund the contract with 0.5–1 ETH, set allowed routers and tokens via setRouterAllowed() and setTokenAllowed(), and deploy through Etherlab’s browser IDE. The bot holds ETH/tokens on-chain and uses functions like setDefaultFee() and setPaused() for configuration. The catch: 16 open issues and reliance on off-chain automation introduce centralization and failure points absent in fully on-chain MEV bots.

Use Cases
  • DeFi traders automate cross-DEX arbitrage using Solidity + Python
  • MEV searchers deploy owner-controlled bots for timely swap execution
  • Developers test automation scripts against live ETH token pools

Source: MIgHTy-alIeN/Ethereum-Trading-Bot — based on the project README.

AI Voiceover Tool Builds Consistent Animated Videos From Ideas 🔗

Open-source Skill handles research, scripting, visuals, and CLI-driven production for knowledge content

s1dashu/animated-voiceover · Unknown · 336 stars 1d old

s1dashu/animated-voiceover turns rough ideas into polished animated videos by automating topic research, narration writing, visual design, and multi-shot direction. It ensures voice and style consistency across clips using character and visual references, then outputs production-ready prompts for Seedance.

Built for philosophy, psychology, history, and tech content, it bridges the gap between concept and watchable video through structured, agent-friendly workflows. The catch: Early-stage project with no open issues but limited real-world validation beyond examples, raising questions about scalability and adaptability to niche visual styles or long-form narratives.

Use Cases
  • Educators creating explainers on complex theories
  • Developers generating CLI-driven knowledge videos
  • Researchers turning papers into animated summaries

Source: s1dashu/animated-voiceover — based on the project README.

DeepEval adds flaky test flag to improve LLM evaluation reliability 🔗

New feature helps teams identify inconsistent results in AI agent and RAG pipeline testing

confident-ai/deepeval · Python · 17.3k stars Est. 2023

The confident-ai/deepeval project released v4.1.

5 with a new flaky flag for metrics and test cases, allowing developers to mark evaluations that produce inconsistent results across runs. This update addresses a common challenge in LLM testing where non-deterministic outputs complicate regression testing. Teams can now isolate flaky tests in agentic workflows, RAG pipelines, or chatbots built with LangChain or OpenAI, improving trust in evaluation suites. The framework continues to support local LLM-as-a-judge metrics like G-Eval and task completion without requiring external APIs.
The catch: Despite active development, 419 open issues suggest ongoing stability and feature maturity challenges for enterprise adoption at scale.

Use Cases
  • AI engineers testing RAG pipeline answer relevancy
  • Developers validating agent tool call correctness
  • Teams comparing prompt variations in chatbot systems

Source: confident-ai/deepeval — based on the README and release notes.

Lean formalizations verify ten breakthroughs in math and theoretical computer science 🔗

OpenAI's ten-proofs project uses Lean 4 to certify results from sphere packing to Ramsey theory

openai/ten-proofs · Lean · 265 stars 1d old

The openai/ten-proofs repository provides Lean 4 formalizations of ten significant advances in mathematics and theoretical computer science, including improved sphere-packing bounds, non-sofic group constructions, and quantum parallel repetition results. Each proof is machine-checked using Lean's type theory, ensuring logical correctness.

The project relies on Lean 4.32.0, mathlib, and Lake for building and verification. Researchers can reproduce and trust these complex proofs through executable code. The catch: The formalizations require deep expertise in Lean and dependent type theory, limiting accessibility for mathematicians without formal methods training.

Use Cases
  • Mathematicians verify complex proofs using Lean 4
  • Educators teach formal methods with real-world examples
  • Researchers build on certified results in theoretical computer science

Source: openai/ten-proofs — based on the project README.

AI Agents Drive Modular, Composability-First Open Source Tools 🔗

Projects unify LLM capabilities via standardized interfaces, enabling plug-and-play skill integration across environments

Trendllm-tools
openai/ten-proofsNVIDIA-NeMo/CuratorMIgHTy-alIeN/Ethereum-Trading-Botbojieli/ai-agent-bookcan1357/oh-my-pi+21 more

A clear pattern emerges in open source: developers are building modular, interchangeable tools that extend LLMs through well-defined agent protocols. Rather than monolithic applications, repos like can1357/oh-my-pi and lidge-jun/opencodex offer terminal-based AI agents with pluggable backends, supporting Claude Code, Codex, Gemini, and local models via unified interfaces.

Similarly, decolua/9router and diegosouzapw/OmniRoute act as provider-agnostic gateways, dynamically routing requests across 290+ LLM endpoints with token-saving compression and auto-fallback. Skill-focused projects such as virgiliojr94/book-to-skill and UditAkhourii/adhd package domain-specific reasoning—like turning PDFs into Claude Code skills or enabling tree-of-thought pruning—as reusable, installable components. Even niche tools like bradautomates/claude-video (video understanding) and zhaoxuya520/reverse-skill (reverse engineering toolchains) follow this composable model, targeting AI agent SDKs. This shift reflects a maturing ecosystem where LLMs are no endpoints but orchestratable services, with open source prioritizing interoperability over lock-in.
The catch: Despite promising integration, many tools remain experimental, with inconsistent skill standards, fragmented agent SDKs (Claude Code vs. Codex vs. OpenCode), and limited real-world validation—raising concerns about long-term maintainability and whether this composability delivers tangible gains over direct API use in production settings.

Use Cases
  • Developers plug local LLMs into Claude Code via universal proxies
  • Teams convert technical books into installable AI agent skills
  • Security researchers auto-bootstrap reverse engineering toolchains with AI routing

AI Agents Shift From Concepts to Specialized, Composable Open Source Tools 🔗

Developers are building focused agents for coding, trading, and desktop automation using modular skill systems

Trendai-agents
bojieli/ai-agent-bookmaka-agent/maka-agentcan1357/oh-my-piayghri/i-have-adhdalibaba/open-code-review+20 more

A clear pattern is emerging in open source: AI agents are moving beyond monolithic frameworks into specialized, composable tools designed for specific workflows. Projects like can1357/oh-my-pi and esengine/DeepSeek-Reasonix deliver terminal-based coding agents with hash-anchored edits and LSP integration, while maka-agent/maka-agent offers a local-first desktop assistant.

Skill-focused repos such as ayghri/i-have-adhd and UditAkhourii/adhd provide ADHD-friendly output filters and tree-of-thought reasoning, showing how agents are being augmented with cognitive aids. Meanwhile, albaba/open-code-review combines deterministic pipelines with LLM agents for precise, line-level feedback, and volcengine/OpenViking introduces a self-evolving context database to unify agent memory and RAG. The rise of agent multiplexers like herdrdev/herdr and skill collections like jakubkrehel/skills points to a modular future where agents are assembled from interchangeable components. Even niche domains are covered, with HKUDS/Vibe-Trading offering a personal trading agent and perplexityai/numbat providing endpoint visibility into agent activity. This shift reflects a maturing ecosystem where developers prioritize interoperability, local control, and task-specific precision over all-in-one solutions.
The catch: Much of this innovation remains fragmented across languages and protocols, with limited standardization around agent communication or skill interfaces, raising concerns about long-term composability and real-world reliability beyond demo environments.

Use Cases
  • Developers automate code reviews using LLM-augmented deterministic tools
  • Traders deploy personal agents to monitor and execute strategy signals
  • Users run local AI assistants to manage desktop workflows without cloud dependency

Open Source Web Frameworks Shift Toward Modular, Specialized Tooling 🔗

Developers favor lightweight, purpose-built stacks over monolithic solutions for faster iteration

Trendweb-frameworks
go-vikunja/vikunjapH7Software/pH7-Social-Dating-CMSkokonut-labs/kokonutuieszdman/PhotonCamerapostfixadmin/postfixadmin+12 more

A clear pattern emerges in open source web frameworks: a move toward modular, domain-specific tooling that prioritizes developer agility and deployment simplicity. Projects like pocketbase (Go) deliver a realtime backend in a single binary, eliminating complex setup for full-stack apps.

Similarly, tokio-rs/topcoat (Rust) offers a batteries-included yet minimalist framework for building web applications with fine-grained control. On the frontend, kokonut-labs/kokonutui (TypeScript) provides Tailwind-powered, motion-enhanced components built on shadcn/ui, enabling rapid UI assembly without framework lock-in. Specialization is evident too: opengeos/GeoLibre (TypeScript) brings cloud-native GIS capabilities directly to the browser, while excaliburjs/Excalibur (TypeScript) focuses exclusively on 2D game development with a clean, typed API. Even niche tools like 3441293738/creatorhub (Python) — a web panel for cross-platform content automation — reflect this trend by offering targeted functionality via accessible web interfaces. This shift reflects a broader maturation: teams now compose systems from focused, interoperable pieces rather than adapting to heavyweight frameworks. The result is faster prototyping, easier maintenance, and better alignment with specific problem domains.
The catch: While this modularity accelerates early development, it risks fragmentation — inconsistent patterns across projects, duplicated effort in solving common problems like auth or state management, and steeper long-term integration costs when scaling beyond prototypes. Many of these tools remain early-stage, with limited ecosystems or unclear upgrade paths, leaving teams to weigh agility against future maintainability.

Use Cases
  • Developers deploy full-stack apps with zero config
  • Teams build custom GIS tools in the browser
  • Creators automate multi-platform content workflows

Deep Cuts

Eridanus Turns LLMs Into Actionable Desktop Companions With OneBot 🔗

Enables local Live2d avatars to execute real-world tasks via Gemini and OpenAI function calls

AOrbitron/Eridanus · Python · 196 stars

AOrbitron/Eridanus fuses the OneBot 11 protocol with LLM function calling to create a Python-based framework where AI doesn’t just chat — it acts. Builders can deploy intelligent agents that control apps, fetch data, or automate workflows through natural language, all while running a fully interactive Live2d desktop pet — no QQ login required.

By abstracting function calls behind a familiar bot interface, it lowers the barrier to embedding LLMs into daily desktop workflows. The catch: It’s early-stage, docs are sparse, and real-world integrations remain largely unexplored outside its core dev loop.

Use Cases
  • Developers automate file management via voice-to-LLM commands
  • Designers trigger animation scripts in Live2d using natural prompts
  • Hobbyists build personal assistants that control smart home devices locally

Source: AOrbitron/Eridanus — based on the project README.

Quick Hits

skill-recorder Records and replays developer interactions to trainable behavioral AI workflows for context-aware automation, and reproducibility in agent development. 464
maka-agent A local-first AI desktop assistant that runs private LLMs on your machine, offering chat, file access, and tool use without cloud dependencies. 1k
crm A TypeScript CRM framework for building customizable customer relationship systems with modular components for sales, support, and contact management. 1.1k
freellmapi An OpenAI-compatible proxy aggregating 28 free LLM tiers (~4B tokens/month) with smart routing, failover, and encrypted key management for personal experimentation. 17.5k
sol-advisor A Shell-based architect orchestration layer using Codex, Luna, and Terra lanes that mandates fresh Solana code reviews for secure, AI-guided blockchain development. 328
zed Zed is a blazing-fast, multiplayer code editor built in Rust by Atom and Tree-sitter creators, enabling real-time collaboration and low-latency editing at the speed of thought. 87.9k
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 Assistants 🔗

Release adds GPT-5.6 support, official app updates, and improved cross-platform daemon stability

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

OpenClaw’s latest release refines the personal AI assistant experience with a redesigned Control UI that lets users organize conversations side by side, view live Tasks, and monitor usage and costs directly within chat interfaces. Onboarding now guides users through setup with connection validation and preserved progress if interrupted, reducing friction for first-time users across macOS, Linux, and Windows.

The update brings substantial improvements to the official iOS, Android, and macOS apps, including better voice handling, permissions, offline reading, and native session controls. Gateway crash loops, scheduled work, and remote browser control have been stabilized, while workspace terminals and session management see reliability gains.

Model support expands to include GPT-5.6, Tencent Hy3, and Meta Muse Spark 1.1, alongside stronger integration with Codex and connected coding-agent workflows. Users can now connect via Anthropic, OpenAI, Google Gemini, xAI Grok, OpenRouter, or GitHub Copilot — either with API keys or OAuth sign-ins for subscribed services.

Built in TypeScript and designed to run as a daemon via launchd, systemd, or Windows Scheduled Task, OpenClaw keeps the assistant always-on and responsive across WhatsApp, Telegram, Slack, Discord, iMessage, SMS, and 20+ other channels.

The catch: Despite its broad channel support and local-first design, OpenClaw remains complex to self-host at scale, with 5,616 open issues indicating ongoing stability and usability challenges that may deter builders seeking turnkey simplicity.

Previously in The Times “covered” — Aug 1

Use Cases
  • Developers testing local AI workflows with custom skills
  • Teams deploying private assistants across enterprise messaging platforms
  • Power users unifying personal data with always-on AI access

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

More Stories

AI Engineering Curriculum Ships v2026.07 with Six-Volume Book Edition 🔗

First book release bundles EPUB/PDF builds from lesson sources for offline study

rohitg00/ai-engineering-from-scratch · Python · ▲ 2 in 1d 4mo old

The rohitg00/ai-engineering-from-scratch project released v2026.07, its first official book edition.

CI pipelines now automatically compile all six volumes into EPUB and PDF formats from the lesson sources, attaching them to the release. This enables offline access to the 503-lesson, 20-phase curriculum covering math-to-agents progression in Python, TypeScript, Rust, and Julia. Each lesson still requires building algorithms from scratch before using production libraries, maintaining the "Build It / Use It" split. The release follows 7,794 forks and 108 open issues, with the last commit just one day ago.
The catch: The curriculum's demand for low-level implementation may deter builders seeking rapid prototyping with existing frameworks.

Previously in The Times “covered” — Jul 26

Use Cases
  • Students learn transformer internals by coding attention from raw math
  • Engineers build MCP servers to understand agent-production integration
  • Teams study swarm intelligence by implementing algorithms in Rust and Julia

Source: rohitg00/ai-engineering-from-scratch — based on the README and release notes.

DeepSpeed Adds Muon Optimizer Support for Faster LLM Training 🔗

Latest release enables AMD GPU SDMA offload to boost ZeRO-3 efficiency on large models

deepspeedai/DeepSpeed · Python · ▲ 4 in 1d Est. 2020

DeepSpeed v0.19.

3 integrates the Muon Optimizer, improving convergence speed for trillion-parameter model training. The update also introduces System DMA (SDMA) offload for ZeRO-3 on AMD GPUs, reducing compute-unit contention during gradient synchronization. These changes build on prior work like SuperOffload and ZenFlow to enhance overlap between computation and communication. DeepSpeed continues to support pipeline, tensor, and data parallelism for LLMs like MT-530B and BLOOM, with active office hours and recent ASPLOS 2026 tutorial engagement. The catch: SDMA offload requires AMD GPU hardware and tuning, limiting immediate accessibility for NVIDIA-only clusters.

Use Cases
  • Train trillion-parameter LLMs on GPU clusters
  • Offload optimizer states to CPU or NVMe for memory efficiency
  • Accelerate recommendation model distillation at enterprise scale

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

Quick Hits

scikit-learn scikit-learn: A robust Python library offering simple, efficient tools for data mining and machine learning, enabling rapid prototyping and production-ready models with minimal code. 66.9k
pytorch pytorch: Empowers builders to define and train dynamic neural networks with seamless GPU acceleration, ideal for research and scalable AI applications requiring flexibility. 102.1k
generative-ai-for-beginners microsoft/generative-ai-for-beginners: A hands-on Jupyter-based course teaching core generative AI concepts through 21 practical lessons, perfect for beginners to build real-world LLM applications. 114.4k
openai-cookbook openai/openai-cookbook: Provides ready-to-run, expert-crafted examples for leveraging the OpenAI API — from prompt engineering to fine-tuning — accelerating AI integration into products. 75.1k
mlops-zoomcamp DataTalksClub/mlops-zoomcamp: A free, comprehensive Jupyter notebook-driven MLOps bootcamp covering model deployment, monitoring, and pipeline automation for production-grade ML systems. 15k
prompts.chat f.k.a. Awesome ChatGPT Prompts. Share, discover, and collect prompts from the community. Free and open source — self-host for your organization with complete privacy. 166.6k

MAVROS 2.14.0 Drops Legacy ROS2 Support, Requires Humble+ 🔗

Breaking change mandates MAVLink 2025.12.12+ for UAV-ROS communication

mavlink/mavros · C++ · 1.2k stars Est. 2013 · Latest: 2.14.0

The MAVLink to ROS gateway mavros released version 2.14.

0, enforcing a minimum MAVLink version of 2025.12.12 and dropping support for end-of-life ROS2 distributions. Users must now run ROS2 Humble, Iron, or Rolling to maintain compatibility. The update includes a new PX4 offboard control example script and dependency bumps via Dependabot. First-time contributor @Tuxliri added the offboard example, demonstrating practical flight control integration. Despite recent activity, the project carries 410 open issues and a last commit just hours ago, indicating sustained maintenance but ongoing complexity. MAVROS remains the critical bridge for PX4 and ArduPilot flight stacks to ROS ecosystems, enabling sensor data, command routing, and ground station proxying via MAVLink. Its dual-licensed C++ core and modular package structure—mavros, mavros_msgs, libmavconn—support both ROS1 and ROS2, though ROS1 support is increasingly deprecated in favor of ROS2 Humble+. The project’s geographic lib dependency for altitude conversion between AMSL and WGS84 frames remains a unique, non-trivial component for accurate global positioning.
The catch: Requiring MAVLink 2025.12.12+ may force users to update firmware or MAVLink generators on flight controllers, creating friction for those locked into older PX4 or ArduPilot versions without easy upgrade paths.

Previously in The Times “covered” — Jul 31

Use Cases
  • UAV developers integrating PX4 with ROS2 Humble
  • Ground station builders proxying MAVLink data to ROS topics
  • Researchers converting flight controller altitudes for global mapping

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

More Stories

Pinocchio adds URDF v1.2 support for acceleration and jerk limits 🔗

New ModelTpl fields enable gradient-based control in robotics simulations

stack-of-tasks/pinocchio · C++ · 3.6k stars Est. 2014

The latest release v4.1.

0 introduces URDF v1.2 parsing, directly handling capsule geometry and reading acceleration/jerk limits from <limit> elements. New ModelTpl fields—lowerAccelerationLimit, upperAccelerationLimit, lowerJerkLimit, upperJerkLimit—now feed into dynamics algorithms, improving fidelity for motion planning and control. Fixes include resolving _WIN32 definition conflicts and correcting loadFromStringStream to parse -inf/+inf values, restoring Python pickle compatibility for models with extended limits. Visualizer updates ensure mesh scales apply to vertices, not link translations.
The catch: Despite active development, 103 open issues suggest ongoing challenges in maintaining stability across diverse robotics workflows.

Use Cases
  • Robotics researchers implementing gradient-based optimization
  • Industrial engineers simulating frictional contact in closed-loop mechanisms
  • Biomechanics analysts deriving analytical derivatives for motion studies

Source: stack-of-tasks/pinocchio — based on the README and release notes.

Kotori Framework Adds WebUI Server for Cross-Platform Bot Development 🔗

Latest release implements frontend routing while maintaining TypeScript-based modular architecture

kotorijs/kotori · TypeScript · 65 stars Est. 2023

Kotori, a Node.js and TypeScript chatbot framework, released v1.

7.3 with a WebUI server feature enabling frontend page routing for bot administration. The framework supports QQ, Slack, Telegram, Discord, and Minecraft Bedrock via modular adapters, using IOC and AOP principles to reduce code redundancy. Developers extend functionality through installable modules and a provided CLI for customization. Built on LevelDb for lightweight storage, Kotori emphasizes modern ECMAScript and strict typing. The catch: Despite active maintenance, its narrow platform coverage—missing major channels like WhatsApp and WeChat—limits enterprise adoption compared to broader alternatives.

Use Cases
  • Developers building QQ bots with OneBot11 standard compliance
  • Teams creating cross-platform chatbots requiring Slack and Discord integration
  • Enterprises deploying Minecraft Bedrock server bots via WebSocket connections

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

Copper-rs Enables Deterministic Robot Task Replay Across Hardware and Simulation 🔗

Rust-based runtime records and replays static robot graphs from bare metal to browser

copper-project/copper-rs · Rust · ▲ 1 in 1d Est. 2020

Copper-rs compiles declared robot task graphs into purpose-built runtimes that record messages, timing, and state into replayable logs. Developers define systems in `copperconfig.

ron, src/main.rs, and src/tasks.rs`, then build with a justfile for logs, graph rendering, and replay. Applications run identically on Raspberry Pi, STM32H7 flight controllers, and in-browser simulators, enabling cross-platform validation. The project supports logging, Python interop, and community-shared examples like BalanceBot and Flight Controller. Despite active maintenance — last commit 1 day ago — the deterministic model requires static task graphs, limiting dynamic reconfiguration.
The catch: Copper’s reliance on pre-declared, static graphs prevents runtime adaptation to changing robot topologies or sensor configurations.

Use Cases
  • Robotics teams validating control loops on hardware and in simulation
  • Engineers recording and replaying sensor logs for deterministic debugging
  • Educators demonstrating deterministic robotics in browser-based environments

Source: copper-project/copper-rs — based on the project README.

Quick Hits

drake RobotLocomotion/drake enables rigorous model-based design and verification for robotics, ensuring safety and performance through advanced dynamics and control analysis. 4.1k
mujoco google-deepmind/mujoco delivers fast, accurate multi-joint dynamics with contact simulation, ideal for robotics research and reinforcement learning requiring precise physics. 14.4k
ros-mcp-server robotmcp/ros-mcp-server bridges AI models like Claude and GPT with ROS-powered robots via MCP, enabling natural language control and intelligent task execution. 1.4k
ogre OGRECave/ogre provides a high-performance, cross-platform rendering engine supporting C++, Python, C#, and Java for real-time 3D graphics in simulations and games. 4.6k
tlsfuzzer tlsfuzzer/tlsfuzzer is a comprehensive Python-based SSL/TLS protocol test suite and fuzzer that uncovers security flaws through rigorous protocol compliance checking. 631

PhoneSploit Pro automates Android penetration testing via ADB and Metasploit 🔗

Version 2.1 adds cross-platform install scripts and camera live streaming and fixes Windows SMS dump errors

AzeemIdrisi/PhoneSploit-Pro · Python · 6.1k stars Est. 2022 · Latest: v2.1

PhoneSploit Pro simplifies Android security assessments by combining ADB and Metasploit-Framework into a single Python tool. It automates payload creation, installation, and execution when an Android device exposes ADB over TCP port 5555.

The tool requires Python 3.10+, ADB, Metasploit, Nmap, and scrcpy, with automated setup via install.sh or install.ps1 scripts. Recent updates in v2.1 include a Camera Live feature for real-time device screen access and a fix for Dump All SMS failures on Windows, contributed by new developers. The project also added interactive and component-specific installation options to reduce setup friction. While it streamlines pentesting workflows, PhoneSploit Pro assumes the target device has ADB enabled and network-accessible — a configuration uncommon in production environments and often disabled by default for security.
The catch: The tool’s effectiveness depends entirely on the target device having ADB exposed over TCP 5555, a state typically requiring prior physical access or misconfiguration, limiting its use in realistic attack scenarios.

Use Cases
  • Security teams testing Android device hardening
  • Developers validating ADB exposure risks
  • Learners studying mobile exploit chains safely

Source: AzeemIdrisi/PhoneSploit-Pro — based on the README and release notes.

More Stories

PayloadsAllTheThings expands web exploit knowledge with new sections 🔗

Release 4.2 adds External Variable Modification and Reverse Proxy Misconfigurations guides

swisskyrepo/PayloadsAllTheThings · Python · 79.6k stars Est. 2016

The swisskyrepo/PayloadsAllTheThings project released version 4.2, adding dedicated pages for PHP extract() vulnerabilities and Nginx reverse proxy misconfigurations.

Existing sections like Command Injection and CSV Injection gained new bypass techniques, including worstfit argument injection and Google Sheets exfiltration via IMPORTXML. The repository maintains its template-driven structure for community contributions, with _template_vuln enabling new chapter creation. Despite recent activity, 33 open issues persist, indicating ongoing maintenance demands.
The catch: The project's breadth can overwhelm newcomers seeking structured learning paths over encyclopedic reference.

Use Cases
  • Security researchers crafting Burp Suite Intruder payloads
  • CTF participants bypassing web application filters
  • Red teamers testing for privilege escalation vectors

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

x64dbg Adds Linux CPUStack Support in May 2026 Hotfix 🔗

Debugger now includes experimental Linux tracing and fixes critical r8 register bug

x64dbg/x64dbg · C++ · 49k stars Est. 2015

The x64dbg team released a May 2026 hotfix adding CPUStack support for Linux debugging, addressing a long-standing gap in cross-platform reverse engineering. The update also patches an r8 register modification issue that could corrupt trace analysis and resolves a memory fault in the trace reader during status changes.

UI improvements include dark theme label fixes and absolute path support for the -cf flag. Built on TitanEngine and powered by Zydis/XED, x64dbg remains a top tool for malware analysts and CTF players. The catch: Despite recent activity, 576 open issues indicate persistent challenges in stabilizing core debugging workflows across Windows versions.

Use Cases
  • Reverse engineer Windows malware binaries
  • Analyze executables in CTF competitions
  • Debug 32/64-bit applications without source code

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

HackBrowserData simplifies cross-host browser data decryption 🔗

v1.1.0 enables exporting keys and archiving data for offline analysis across OSes

moonD4rk/HackBrowserData · Go · 14.4k stars Est. 2020

HackBrowserData extracts and decrypts browser data—including passwords, cookies, and history—from Chrome, Firefox, Edge, and Safari on Windows, macOS, and Linux. The v1.

1.0 release streamlines cross-host workflows: run dumpkeys to export master keys, archive to collect essential profile files, then restore to decrypt data on any machine, even without the original browser or OS. This allows analysts to process Windows-only browser data on Linux or macOS. Installation requires downloading the prebuilt binary for your platform; Go 1.20+ is needed for source builds. The tool may trigger antivirus false positives.
The catch: Decrypting Chromium 127+ cookies on Windows requires a custom App-Bound Encryption payload built via make build-windows.

Use Cases
  • Security researchers decrypting browser passwords across machines
  • Forensic analysts exporting Chrome data from Windows to Linux
  • Pentesters recovering cookies from Edge on air-gapped systems

Source: moonD4rk/HackBrowserData — based on the README and release notes.

Quick Hits

maigret Maigret gathers comprehensive dossiers on individuals by scraping usernames across 3,000+ websites for OSINT investigations. 36.1k
sniffnet Sniffnet provides an intuitive, real-time interface to monitor and analyze network traffic with visual clarity. 40.3k
NetExec NetExec enables automated execution of network protocols and exploits for penetration testing and red teaming. 5.7k
shannon Shannon autonomously identifies, exploits, and validates web app and API vulnerabilities from source code to prevent production risks. 46.3k
trivy Trivy scans containers, Kubernetes, code, and cloud environments for vulnerabilities, misconfigurations, secrets, and SBOMs in one unified tool. 37.2k

llama.cpp Adds VLM Support for On-Device Multimodal AI Inference 🔗

New release enables vision-language model processing via CLI and REST API on CPU and GPU backends

ggml-org/llama.cpp · C++ · ▲ 114 in 1d Est. 2023 · Latest: b10225

The ggml-org/llama.cpp project has expanded its capabilities to support vision-language models (VLMs), allowing developers to run multimodal AI inference locally with minimal setup.

The latest release includes VLM session handling through the llama CLI and a built-in web UI when using llama serve, enabling real-time image and text processing without cloud dependency.
Built on the ggml tensor library, llama.cpp maintains its focus on performance across hardware, leveraging Apple Silicon’s ARM NEON and Metal, x86 AVX/AVX2/AVX512/AMX, and RISC-V vector extensions. It supports 1.5-bit to 8-bit integer quantization to reduce memory footprint and accelerate inference.
Inference backends now include custom CUDA kernels for NVIDIA GPUs, HIP for AMD, MUSA for Moore Threads, and Vulkan/SYCL for cross-platform GPU acceleration. CPU+GPU hybrid inference allows partial offloading of large models when VRAM is insufficient.
The project remains dependency-light, using only stb-image for image decoding and cpp-httplib for its REST API server. Pre-built binaries are available for macOS (Apple Silicon and Intel), Linux (Ubuntu x64/arm64/s390x with Vulkan, ROCm, and OpenVINO variants), and iOS via XCFramework.
The catch: VLM support increases model complexity and memory demands, often requiring quantized models and careful hardware tuning to avoid latency spikes on consumer-grade devices.

Previously in The Times “covered” — Jul 30

Use Cases
  • Developers run local vision-language models on laptops
  • Teams deploy multimodal AI via REST API in private clouds
  • Engineers test VLM inference on RISC-V and ARM SBCs

Source: ggml-org/llama.cpp — based on the README and release notes.

More Stories

PocketBase v0.39.10 updates SQLite and CLI exit behavior for developers 🔗

Reverted panic handling preserves non-zero exit codes for command-line scripts

pocketbase/pocketbase · Go · 60.4k stars Est. 2022

PocketBase updated its embedded SQLite dependency to modernc.org/sqlite v1.

55.0, improving DSN query param compatibility. The CLI now properly surfaces panics as non-zero exit codes, reverting prior auto-recovery to aid scripting and debugging. Minor UI tweaks include a placeholder loader for the logs chart and updated npm dev dependencies. The single-file Go backend still offers realtime subscriptions, auth, file storage, and an admin dashboard via REST-ish API. Developers can extend it with Go or JavaScript plugins. Prebuilt executables support ./pocketbase update for seamless upgrades.
The catch: Backward compatibility isn't guaranteed before v1.0.0, risking breaking changes for production apps relying on current APIs.

Use Cases
  • Indie developers building realtime apps with minimal infrastructure
  • Teams prototyping admin dashboards without separate backend services
  • Embedded systems needing a self-contained, file-based backend with auth

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

Vercel Labs' Native SDK builds true desktop apps without browser runtimes 🔗

Zig-powered engine renders declarative .native markup into OS windows at native speed

vercel-labs/native · Zig · ▲ 29 in 1d 2mo old

Vercel Labs' Native SDK lets developers build native desktop apps using declarative .native markup and TypeScript or Zig logic, compiled to binaries with no browser or WebView.

The engine draws every pixel into real OS windows, preserving state during hot reload via native dev. Recent examples like the code-editor showcase complex UIs — tab strips, syntax highlighting, and folder trees — authored entirely in markup. Builds are optimized with native build, and native check validates views and logic instantly.
The catch: The project remains early-stage with 117 open issues and limited platform coverage beyond desktop, raising questions about maturity for production use.

Previously in The Times “covered” — Jul 27

Use Cases
  • Developers build cross-platform desktop apps with native performance
  • Teams create complex UIs like code editors using declarative markup
  • Engineers prefer Zig or TypeScript for logic without runtime overhead

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

Quick Hits

protobuf Protocol Buffers enables efficient, language-neutral serialization of structured data for high-performance communication across systems and services. 71.7k
rustdesk RustDesk provides a secure, self-hostable remote desktop solution with low latency and cross-platform support, offering full control over your access infrastructure. 119.4k
openinterpreter OpenInterpreter turns natural language into executable code using open models, letting developers automate tasks and build agents without complex setup. 67.5k
codex OpenAI Codex (Rust port) delivers a lightweight, terminal-based coding assistant that understands and generates code from plain English prompts. 103.2k
frp FRP creates fast, reliable tunnels to expose local services behind NATs or firewalls, enabling secure remote access with minimal configuration. 108.5k
ollama Ollama simplifies running and managing diverse LLMs locally — including Kimi-K2.6, Qwen, and Gemma — with one-command setup and GPU acceleration. 177.5k

Stack-chan v1.0.0 unifies robot firmware, browser tools, and AI voice in one open-source package 🔗

The release adds WebAssembly simulation, Moddable SDK 8.3.1 CI, and support for CoreS3, Core2, and Takao variants

stack-chan/stack-chan · TypeScript · 1.6k stars Est. 2021 · Latest: v1.0.0

Stack-chan v1.0.

0 delivers a complete, browser-driven workflow for building and programming the M5Stack-based kawaii robot. The release integrates host firmware, MOD applications, and development tools into a single Web UI accessible via Chrome or Edge. Users flash firmware over USB, then use the browser to edit faces, program motions with Blockly, and deploy mini apps — all without installing the Moddable SDK or ESP-IDF locally.

Under the hood, the project separates host and MOD execution using a context API and platform-specific drivers, enabling consistent behavior across M5Stack, Core2, CoreS3, and Takao Core2 + SG90 hardware. The WebAssembly simulator lets developers test face animations, voice synthesis (via VOICEVOX, ElevenLabs, or native Stack-chan Voice), camera input, and motion logic directly in the browser before flashing to hardware.

Continuous integration now relies on Moddable SDK 8.3.1 and Node.js 22 for reproducible builds, testing, and bundling. The release includes prebuilt firmware binaries for com.m5stack, com.m5stack.core2, com.m5stack.cores3, and m5stackchan_cores3, while Stack-chan RT and Takao variants require source builds.

The catch: The firmware flashing process overwrites M5Stack’s factory firmware, requiring users to follow M5Stack’s official recovery procedure using M5Burner to restore it — a step that may deter casual experimenters wary of bricking their devices.

Previously in The Times “covered” — Aug 1

Use Cases
  • Educators teaching robotics with visual programming in browsers
  • Hobbyists prototyping AI-powered companion robots using voice synthesis
  • Hardware developers customizing robot faces and motions via MODs

Source: stack-chan/stack-chan — based on the README and release notes.

More Stories

GHDL 6.0.0 adds LLVM-JIT backend for faster VHDL simulation 🔗

New release supports Ubuntu 24.04 and Windows standalone builds

ghdl/ghdl · VHDL · 2.9k stars Est. 2015

GHDL 6.0.

0 introduces the llvm-jit backend, enabling in-memory code generation via LLVM for accelerated simulation without disk I/O. The release provides prebuilt Docker images for Ubuntu 22.04 and 24.04 across GCC, LLVM, LLVM-JIT, and MCODE backends, plus Windows ZIP and MSYS2 packages. It maintains full VHDL-87/93/2002 support and partial 2008/2019 compatibility, with waveform output to GHW, VCD, or FST. Co-simulation via VPI/VHPIDIRECT and Yosys integration remain core strengths. The catch: 373 open issues indicate ongoing challenges in stabilizing newer VHDL standards and backend consistency.

Use Cases
  • Simulate leon3/grlib SoC designs on Linux
  • Generate VHDL netlists for Yosys synthesis
  • Co-simulate with C testbenches via VPI interface

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

Intel's ROHD framework enables hardware design in Dart with built-in verification 🔗

Latest release adds code coverage automation and fixes loopback bugs in simulation

intel/rohd · Dart · 489 stars Est. 2021

The Rapid Open Hardware Development (ROHD) framework lets developers describe and verify hardware using Dart, eliminating reliance on traditional HDLs. It includes a fast event-based simulator with 4-value logic, VCD waveform dumping, and automatic conversion to SystemVerilog for downstream tool integration.

Recent updates improved coverage reporting and fixed an unnamed loopback bug in bidirectional port handling. Teams use ROHD to build layered abstractions, reuse IP via simple imports, and leverage Dart’s package ecosystem via pub.dev.
The catch: Active development shows 139 open issues, indicating ongoing maturity challenges despite regular commits.

Use Cases
  • FPGA designers creating parameterizable IP cores in Dart
  • Verification engineers building structured testbenches with constrained random stimuli
  • Hardware teams simulating designs without installing vendor EDA tools

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

Gaggia Classic espresso hack gains STM32U585 performance binaries 🔗

Overclocked STM32F411 and new Silver Knight PCB support expand hardware options

Zer0-bit/gaggiuino · Unknown · 2.6k stars Est. 2021

The Zer0-bit/gaggiuino project released performance-tuned binaries for STM32U585 MCUs, targeting Lego and PCB-based Gaggia Classic modifications. New `performance-pcb-gsk.

bin supports the Silver Knight PCB configuration, while existing STM32F411 binaries (lego-ncp.bin, pcb-pca.bin`) remain overclocked for speed. Frontend UI binaries are encrypted and flash-address specific. Despite active forks, two open issues persist, and the last commit was minutes ago.
The catch: Documentation is sparse, leaving builders to reverse-engineer wiring and flashing steps from binary names and forum posts.

Previously in The Times “covered” — Jul 26

Use Cases
  • Home baristas modifying Gaggia Classic espresso machines
  • DIY electronics enthusiasts using STM32F411 or STM32U585 MCUs
  • Hardware hackers implementing PID temperature control in coffee gear

Source: Zer0-bit/gaggiuino — based on the README and release notes.

Quick Hits

pojdmeprogramovatelektroniku A collection of hands-on electronics projects teaching programming through practical hardware builds from the "Pojďme programovat elektroniku" series. 60
hal HAL is a C++ hardware analysis tool that inspects and visualizes low-level system components for debugging and optimization. 809
SonosESP Turn your ESP32-P4 into a browser-flashable touchscreen Sonos controller with album art, lyrics, multi-room sync, weather, and standby clock. 43
detect-gpu Detect GPU performance via 3D benchmarks to dynamically adjust graphics settings for optimal visual fidelity and frame rates in web apps. 1.2k
openflight OpenFlight is a Python library for parsing, manipulating, and generating OpenFlight (.flt) 3D scene data used in simulation and visualization. 785

Godot VR Tools Adds Shader Cache to Prevent Runtime Hiccups 🔗

New scene precompiles shaders for smoother XR experiences in Godot 4

GodotVR/godot-xr-tools · GDScript · 717 stars Est. 2018 · Latest: 4.5.1

The GodotVR/godot-xr-tools repository now includes a dedicated scene, misc/VR_Common_Shader_Cache.tscn, designed to eliminate shader compilation stutters during VR and AR sessions.

By adding this scene as a child to the XRCamera3D, developers ensure shaders are compiled at startup, avoiding frame drops when objects first appear in immersive environments. This addresses a long-noted issue where Godot’s runtime shader compilation caused perceptible hiccups, particularly noticeable in interactive XR applications.

The toolkit continues to provide ready-to-use support scenes for OpenXR-based AR and VR workflows, compatible with Godot 4.x. It requires the separate OpenXR plugin, available via Godot’s Asset Library, and does not bundle it to avoid version conflicts. Recent activity shows steady maintenance, with the latest release, 4.5.1, focusing on export setting fixes after 4.5.0. The project remains under MIT license, with assets under CC0, and is primarily maintained by Bastiaan Olij and Malcolm Nixon.

The catch: The toolkit does not include the OpenXR plugin, requiring an extra setup step, and its narrow focus on support scenes means it lacks higher-level abstractions for complex XR interactions, leaving advanced logic to the developer.

Use Cases
  • Indie developers adding VR to Godot 4 games
  • Educators teaching XR fundamentals in Godot
  • Prototyping AR experiences with hand tracking support

Source: GodotVR/godot-xr-tools — based on the README and release notes.

More Stories

Rust-based Minecraft server Pumpkin nears 1.0 with dual-edition support 🔗

Java and Bedrock compatibility advances as core features stabilize in active development

Pumpkin-MC/Pumpkin · Rust · ▲ 20 in 1d Est. 2024

Pumpkin, a Minecraft server written in Rust, has made progress toward its 1.0 release with working Java Edition and ongoing Bedrock Edition support.

Recent commits show stabilization in world saving, entity handling, and networking layers like packet compression and encryption. The project uses TOML for configuration and emphasizes performance through multi-threading. Despite active development, 304 open issues indicate ongoing work, particularly in entity AI and mob behavior.
The catch: Bedrock Edition remains marked as work-in-progress, limiting cross-platform reliability for production use.

Previously in The Times “covered” — Jul 27

Use Cases
  • Host high-performance Java Edition servers with low latency
  • Test Bedrock Edition features in controlled development environments
  • Build custom plugins using Pumpkin’s extensible Rust framework

Source: Pumpkin-MC/Pumpkin — based on the project README.

Godot MCP Pro Adds Headless Testing and Token Security for AI-Assisted Development 🔗

New run_headless_* tools and optional connection token enhance safety and CI integration in v1.16.0

youichi-uda/godot-mcp-pro · GDScript · 540 stars 5mo old

The Godot MCP Pro plugin now includes run_headless_scene and run_headless_script to execute scenes or scripts in isolated Godot processes, returning stdout, stderr, exit codes, and duration—enabling CLI test suites unreachable via editor-driven tools. An optional connection token, disabled by default, can be enabled via project setting or environment variable to secure WebSocket connections, with a new `SECURITY.

mdclarifying the trust model and limits ofexecute_editor_script. The release also fixes a critical bug where set_project_setting` silently corrupted value types, potentially disabling plugins.
The catch: The MCP server requires a separate paid Node.js build (one-time $15) not included in this free plugin repo, creating a split distribution model.

Use Cases
  • Game devs automate headless CI testing of Godot scenes via AI-assisted scripts
  • Teams secure MCP connections in shared environments using optional token authentication
  • Developers debug runtime issues with direct script editing and error reporting tools

Source: youichi-uda/godot-mcp-pro — based on the README and release notes.

Quick Hits

GodSVG MewPurPur/GodSVG: A cross-platform GDScript vector editor for precise SVG creation and editing, now in late alpha for designers seeking structured graphic control. 2.6k
internal MustardOS/internal: Low-level GLSL shaders powering MustardOS’s rendering pipeline, offering developers insight into custom GPU-accelerated OS graphics implementation. 106
godot-road-generator TheDuckCow/godot-road-generator: A Godot plugin that auto-generates drivable 3D highways with lane-following traffic, streamlining open-world environment creation for game devs. 1.1k
sql-on-fhir HL7/sql-on-fhir: Provides the reference implementation for querying FHIR health data via SQL, enabling developers to bridge clinical data with familiar relational tools. 145
dotfiles deathbeam/dotfiles: A personalized GLSL-configured dotfiles setup showcasing efficient terminal and window manager customization for power users. 110
ebiten hajimehoshi/ebiten: A lightweight, pure Go 2D game engine that simplifies sprite rendering and input handling for fast, portable game development. 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