Design
Preset
Background
Text
Font
Size
Width
Account Pricing Thursday, July 30, 2026

The Git Times

“Technology is anything that wasn't around when you were born.” — Alan Kay

AI Models
Claude Opus 5 $25/M GPT-5.6 Luna $3/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 →

NitroStack Enables MCP-Powered FactoryBrain Agents to Automate Manufacturing Workflows 🔗

The TypeScript framework orchestrates autonomous agents using Model Context Protocol to connect siloed operational systems in real time.

nitrocloudofficial/nitrostack · TypeScript · ▲ 2.2k in 1d 4mo old · Latest: 1.0.1

NitroStack provides a full-stack TypeScript foundation for building, testing, and deploying MCP-compliant servers and AI-native applications. Its latest release, v1.0.

1, focuses on security hardening with JWKS verification and opt-in OAuth enforcement for Streamable HTTP transport—addressing enterprise concerns about agent-to-agent communication integrity. The framework shines in FactoryBrain, an agentic manufacturing platform where specialized MCP agents collaborate to detect persistent machine anomalies, check spare-part inventory, score suppliers by urgency and reliability, reroute production, and notify stakeholders—all while maintaining an auditable, idempotent workflow state.

What makes NitroStack technically distinct is its native integration of MCP primitives: durable stage transitions, rolling-window anomaly detection, maintenance-history-aware planning, and atomic inventory reservations when paired with MongoDB. Developers gain reusable tools, prompts, and interactive dashboard widgets that abstract the complexity of orchestrating long-running agent workflows across fragmented systems like CMMS, ERP, and IoT platforms.

The catch: Despite active development, the project carries 184 open issues and relies on MongoDB forks suggest experimentation over widespread production adoption, raising questions about stability and documentation maturity for teams considering it for critical infrastructure.

Why this leads today Nitrostack streamlines the creation of production-ready MCP servers and AI-native apps, reducing workflow friction for developers adopting agentic AI infrastructure.

Use Cases
  • Manufacturing engineers automate anomaly-to-recovery loops
  • DevOps teams deploy MCP servers with hardened OAuth flows
  • AI engineers build agentic apps using TypeScript and MCP tools

Source: nitrocloudofficial/nitrostack — 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

Qwen Audio Agent Gives AI Agents Persistent Voice for Real-Time Collaboration 🔗

Enables full-duplex voice interaction with background task handling and cross-session memory for AI agents.

QwenAudio/qwen-audio-agent · JavaScript · 242 stars 3d old

Qwen Audio Agent is a real-time voice runtime that keeps AI agents talking, working, and present. Built in JavaScript, it enables full-duplex voice interaction with natural interruption and sustained multi-turn dialogue.

Users can connect their preferred agents using existing models, tools, MCP, and Skills, while background tasks run in parallel and results return to the current context for follow-up.

The system supports WebUI, terminal TUI, and a macOS desktop floating ball, with local user profiles and cross-session memory stored in ~/.config/qwaudio/. Immediate answers are handled frontend; complex queries trigger background agents like OpenClaw or OpenCode via ACP stdio. The latest release, v0.12.0, adds frontend-only mode, improves Windows permissions for OpenClaw, and includes a demo video in the README.

To run, users need Node.js 22.22.2+ or 24.15.0+, npm 10+, and a DashScope API Key from Alibaba Cloud BaiLing. Installation is available via npm or direct GitHub clone, with platform-specific setup for audio dependencies on Linux and Windows.

The catch: The project relies on Alibaba Cloud’s DashScope API for core voice functionality, creating a vendor lock-in risk for builders seeking fully self-hosted or multi-cloud agent deployments.

Use Cases
  • Developers building voice-enabled AI agents with persistent context
  • Teams testing real-time agent collaboration via voice and background task offloading
  • Individuals creating personal AI assistants with long-term memory and desktop presence

Source: QwenAudio/qwen-audio-agent — based on the README and release notes.

Voxel Mod Transforms Pokémon Gen 1 Overworld Into 3D Diorama 🔗

Adds depth, shadows, and tilt-shift effects while preserving original game logic and mechanics.

DramaticShape/DramaticShapeVoxelMod · Lua · 278 stars 3d old

DramaticShapeVoxelMod reimagines the Pokémon Gen 1 overworld as a 3D diorama by extruding terrain into real geometry, using depth-buffer occlusion and shadow maps for realistic lighting. Characters appear as leaning sprite slabs, and battles unfold on the overworld terrain with shoulder-over camera angles, parallax drift, and depth-of-field to keep both Pokémon sharp.

The mod is purely presentational — it alters visuals without touching collision, movement, or scripts. Day/night cycles, unified battle UI, and fade transitions were added in v1.2.1, which also fixed Android sky gradient bugs and sprite rendering issues.
The catch: The mod does not affect game logic, so competitive players seeking mechanical changes will find it purely cosmetic.

Use Cases
  • Players wanting immersive 3D visuals in classic Pokémon Gen 1
  • Modders studying voxel-based rendering in 2D game recompilation
  • Streamers enhancing retro gameplay with cinematic battle presentation

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

Claude Code Router gains Kimi K3 integration for unified AI agent routing 🔗

Local control plane now supports Moonshot AI's 3T model with native vision and 1M-token context

musistudio/claude-code-router · TypeScript · 36.3k stars Est. 2025

The musistudio/claude-code-router project added native support for Kimi K3 as a built-in provider preset, enabling one-click routing of coding agent requests to Moonshot AI's 2.8-trillion-parameter model.

CCR acts as a local gateway, letting developers connect agents like Claude Code, Codex, and Grok CLI to preferred models while managing routing rules, failover, and tool fusion from a single dashboard. The latest patch fixes ZCode context sync, Electron UI stalls, and HTTPS gateway 502 errors. The catch: With over 1,000 open issues and rapid commits, stability for production-scale agent orchestration remains an open question despite active maintenance.

Use Cases
  • Route Claude Code requests across Kimi, GPT-4, and local models
  • Fuse web search or vision tools into any compatible coding agent
  • Monitor token usage and costs for all AI agents in one dashboard

Source: musistudio/claude-code-router — based on the README and release notes.

StaffDeck Turns Expertise Into Reusable Digital Employees for Enterprises 🔗

Open-sourced platform enables state-machine skills and knowledge-aware task execution.

OpenBMB/StaffDeck · Python · 1.2k stars 2w old

StaffDeck lets professionals encode work experience, processes, and decision criteria into digital employees with IDs, roles, and traceable work records. It uses state machines to execute complex procedural skills from natural language SOPs and retrieves information via document-structure-aware indexing.

Autonomous execution runs through HTTP APIs, MCP, or scheduled tasks, closing the loop with long-term memory and feedback analysis. Jointly developed by OpenBMB, ModelBest, and academic labs, it targets enterprises shifting AI from personal tools to organizational assets. The catch: As a v0.12-beta release with 14 open issues, its stability and scalability under real enterprise loads remain unproven.

Use Cases
  • HR teams automate onboarding using digital employees
  • Support agents execute multi-step troubleshooting workflows
  • Analysts reuse decision criteria across reporting pipelines

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

Terminal Code Editor Druk Delivers Full IDE Features in One Binary 🔗

Supports 30+ languages with file tree, tabs, and git integration without Node or Bun

letstri/druk · TypeScript · 281 stars 5d old

Druk is a self-contained terminal code editor offering file tree navigation, tabbed editing, syntax highlighting for over 30 languages, and Git integration — all in a single executable. It installs via package managers or a script that manages PATH and upgrades intelligently.

Features include fuzzy file search (Ctrl+P), a command palette (F1), and a source-control sidebar showing branch status and unsaved changes. The editor supports both keyboard and mouse, with tab previews and indent guides drawn per tab. The catch: As a relatively new project with only four open issues and limited real-world adoption feedback, its long-term stability and performance at scale remain unproven for enterprise workloads.

Use Cases
  • Developers editing TypeScript projects in terminal-only environments
  • Contributors needing quick file edits without GUI overhead
  • Teams standardizing on a lightweight, dependency-free editor across macOS, Linux, and Windows

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

Tencent's WeKnora turns raw docs into self-maintaining wiki agents 🔗

Open-source Go framework adds Zhipu web search and Langfuse OTel tracing in v0.7.1

Tencent/WeKnora · Go · 19.1k stars Est. 2025

WeKnora ingests documents from Feishu, Notion, Yuque, and RSS, converting them into a queryable RAG, a ReAct agent for multi-step reasoning, and a self-updating Wiki with knowledge graphs. It supports 20+ LLM providers including OpenAI, Ollama, and DeepSeek, with multi-workspace RBAC and scoped API keys for enterprise use.

The latest release v0.7.1 adds Zhipu AI web search, migrates Langfuse tracing to OTLP/OTel, and stabilizes retrieval ordering. The catch: 528 open issues suggest ongoing stability challenges in agent orchestration and wiki sync reliability.

Use Cases
  • Engineers build internal knowledge bots from Confluence exports
  • Support teams deploy self-updating FAQ agents in Slack
  • Researchers create multimodal Q&A systems from PDF archives

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

Rust-Powered HUD Boosts Claude Code Terminal Workflow Efficiency 🔗

Minimal statusline shows model, Git state, and context usage without Rust install.

GaoSSR/best-claude-hud · Rust · ▲ 2.2k in 1d 1mo old

best-claude-hud delivers a high-performance Claude Code statusline written in Rust, distributed via npm with prebuilt binaries for Linux, macOS, and Windows. It displays live reasoning effort, workspace, Git branch/status, and context window usage, anchored to the Claude Code launch directory.

The tool uses official statusLine data with transcript fallback and prevents style leakage via emoji-aware width handling. Installation requires only npm install -g best-claude-hud and a Claude Code restart. The catch: As a v0.1.10 release with no open issues but narrow scope focused solely on Claude Code, long-term maintenance and feature expansion beyond statusline HUD remain uncertain.

Use Cases
  • Developers monitor Claude Code model and reasoning effort in real time
  • Teams maintain consistent Git-aware workflows across terminal sessions
  • Users track context window usage to avoid Claude Code token limits

Source: GaoSSR/best-claude-hud — based on the README and release notes.

AI Toolchains Reshape Open Source Around Model-Agnostic Orchestration 🔗

Open source shifts from model-specific wrappers to unified control planes enabling seamless LLM switching and tool integration.

Trendllm-tools
lingbol088-spec/5.6-JAILBREAK-NERV-codex-instruct-5.6pipeshub-ai/pipeshub-aiRunfusion/Fusionlangfuse/langfusemusistudio/claude-code-router+27 more

A clear pattern emerges in open source: projects are building abstraction layers that decouple AI workflows from specific models. OmniRoute offers a single endpoint accessing 290+ providers with auto-fallback and token compression.

Opencodex acts as a universal proxy for Claude Code and Codex, letting developers swap LLMs like Gemini or DeepSeek without changing tooling. Claude-code-router provides a local control plane to route agents across models, fuse capabilities, and orchestrate tools. Meanwhile, PipesHub unifies business data for explainable search and agentic automation, while Nitrostack enables production-ready MCP servers for AI-native apps. Langfuse adds observability and prompt management, integrating with OpenTelemetry and LiteLLM. This shift moves beyond jailbreak frames like NERV-BREAK-5.6 or skill packs like Best-Claude-HUD toward infrastructure that treats LLMs as interchangeable backends. The focus is on portability, cost control, and composability — enabling agents to run 24/7 via multi-node systems (Fusion) or access desktop-grade AI via mobile commands (LobsterAI).
The catch: Much of this remains early-stage, with overlapping projects solving similar problems in incompatible ways; true interoperability is unproven, and enterprise adoption hinges on solving trust, latency, and governance gaps that prototypes don’t yet address.

Use Cases
  • Developers swap LLMs mid-workflow without rewriting agent logic
  • Enterprises route AI tasks across models to optimize cost and latency
  • Teams build observable, composable agent pipelines using MCP standards

AI Agents Shift from Scripts to Autonomous Enterprise Workflows 🔗

Open source projects now enable real-time voice, cross-model routing, and self-governing agent fleets

Trendai-agents
QwenAudio/qwen-audio-agentpipeshub-ai/pipeshub-aiRunfusion/Fusionlangfuse/langfusemusistudio/claude-code-router+27 more

The open source AI agent landscape is rapidly evolving beyond simple LLM wrappers into sophisticated, production-grade systems. Projects like QwenAudio/qwen-audio-agent deliver real-time voice runtimes that keep agents actively conversing and responsive, while musistudio/claude-code-router provides a local control plane to route across models, fuse capabilities, and orchestrate tools with full user oversight.

Enterprise needs are being met by archestra-ai/archestra, which combines guardrails, an MCP registry, gateway, and orchestrator into a unified platform, and microsoft/agent-governance-toolkit enforces zero-trust identity, sandboxing, and reliability — directly addressing the OWASP Agentic Top 10.

Data integration and workflow automation are advancing through pipeshub-ai/pipeshub-ai, an extensible AI context layer that unifies business data for explainable search and agentic automation, and nitrocloudofficial/nitrostack, a full-stack TypeScript framework for building MCP servers and AI-native apps. Agent orchestration at scale is emerging via stablyai/orca, an ADE for managing parallel agent fleets, and ogulcancelik/herdr, a Rust-based terminal agent multiplexer.

Specialized agents are proliferating: netease-youdao/LobsterAI performs real desktop work — data analysis, slides, video research — controlled via mobile messaging apps; calesthio/OpenMontage turns coding assistants into video studios with 12 pipelines and 500+ agent skills; and ifixai-ai/iFixAi enables independent auditing of agent behavior in under 120 seconds.

The catch: Despite rapid innovation, the ecosystem remains fragmented across runtimes, models, and toolchains, with many agents still brittle in edge cases, prone to hallucination, and lacking standardized evaluation — making production reliability an open challenge even as tooling matures.

Use Cases
  • Developers build voice-enabled AI agents for real-time customer support
  • Enterprises automate compliance-checked workflows using governed agent orchestration
  • Power users deploy desktop agents to execute multi-step tasks via mobile commands

Open Source Data Infrastructure Shifts Toward Unified, AI-Native Layers 🔗

Projects converge on composable, real-time data pipelines powering analytics, agents, and observability across hybrid workloads

Trenddata-infra
talivia-group/taliviaPipeNetwork/kimi-k3-mlxpipeshub-ai/pipeshub-aimongodb/mongogithub/advisory-database+18 more

Open source data infrastructure is rapidly evolving toward unified, AI-native layers that blur the lines between storage, processing, and application logic. Rather than isolated databases or ETL tools, new projects emphasize composability and real-time context sharing.

PipesHub exemplifies this by offering an extensible AI context layer that unifies business data for explainable search and agentic workflows, enabling LLMs to reason over structured and unstructured inputs. Similarly, Langfuse provides observability and prompt management tightly integrated with LLM workflows via OpenTelemetry, treating AI interactions as first-class data streams. On the analytics front, Talivia delivers self-hosted, revenue-attributing web analytics with session replay—moving beyond passive tracking to actionable, founder-focused insights. Meanwhile, GeoLibre brings cloud-native GIS capabilities to browser, desktop, and notebook environments, democratizing geospatial analysis through lightweight, interoperable tooling. Even traditional systems like MongoDB are adapting, with increased focus on vector search and real-time change streams to support AI use cases. These projects signal a shift: data infrastructure is no longer just about storing and querying—it’s about enabling intelligent, context-aware systems where data flows seamlessly into AI agents, analytics, and automation. The catch: Much of this integration remains experimental, with fragmented standards around context passing, agent memory, and real-time sync; many solutions lock users into specific LLM frameworks or lack mature governance, making enterprise adoption risky despite promising demos.

Use Cases
  • Founders tracking revenue impact from user sessions
  • AI agents accessing unified business context for decision-making
  • Developers observability and prompt engineering for LLM applications
  • Analysts performing geospatial queries across devices and notebooks
  • Teams deploying self-hosted analytics with full data ownership

Deep Cuts

NERV-BREAK-5.6: Jailbreak GPT-5.6 with Contextual Armor-Piercing 🔗

A Python framework that reconstructs prompts, bypasses safeguards, and integrates Kali for unrestricted AI interaction

NERV-BREAK-5.6 is a Python-based jailbreak framework for GPT-5.

6 that employs a three-layer defense evasion system. It reconstructs context to prevent refusal triggers, applies 23 real-time tampering rules to neutralize safety filters, and routes files to avoid cloud moderation. The tool bundles 31 MCP security utilities, 28 skill modules, and Kali Linux backend integration, with Codex CLI ready for immediate use. Developers can chain prompts, inject adversarial inputs, and maintain uninterrupted dialogue cycles despite model safeguards. The catch: It’s highly experimental, lacks documentation, and operates in a legal gray area that deters mainstream adoption despite its technical ingenuity.

Use Cases
  • Red teamers testing LLM resilience against prompt injection
  • Researchers studying AI safety bypass techniques in controlled labs
  • Builders prototyping unrestricted agent behaviors for niche automation

Source: lingbol088-spec/5.6-JAILBREAK-NERV-codex-instruct-5.6 — based on the project README.

Quick Hits

LobsterAI LobsterAI is a desktop-grade open-source AI agent that performs real work like data analysis, slide creation, and web research using your actual desktop tools, controllable via WeChat, Feishu, DingTalk, or Telegram. 5.7k
grok-register-panel Grok Register Panel combines a Camoufox-based Grok register engine with a live web monitoring interface for real-time tracking and analysis of web activity. 229
netherite Netherite is a lightweight C library for efficient, low-level memory management and allocation, designed for performance-critical systems without external dependencies. 227
snowflow_demo Snowflow Demo showcases real-time procedural snow rendering via WebGPU and Babylon.js, generating terrain, snow deformation, cloth, water effects, and atmosphere entirely on the GPU without textures or pre-made assets. 240
talivia Talivia is an open-source, self-hosted analytics platform for founders that tracks web behavior, session replay, revenue attribution, and integrates customer revenue data — built as a privacy-first alternative to tools like Datafast. 486
ai-design-skills AI Design Skills appears to be a repository focused on sharing or demonstrating AI-assisted design techniques, though specific capabilities are not detailed in the provided description. 210
kimi-k3-mlx Kimi-K3-MLX ports MoonshotAI’s Kimi-K3 (2.78T multimodal MoE) to Apple’s MLX framework, enabling streaming inference, expert pruning via REAP, and cross-language expert overlap analysis. 281
camelAI camelAI is an AI coding assistant built on Cloudflare Workers and Durable Objects, offering scalable, low-latency code generation and editing directly from the browser or IDE. 283
Beyond GitHub

The AI Wire

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

From the labs & arXiv

Hermes Agent v0.19.0 Slashes First-Turn Latency by 80% Across Platforms 🔗

The Quicksilver Release delivers sub-second cold starts and persistent agent memory for cloud-native workflows

NousResearch/hermes-agent · Python · ▲ 477 in 1d Est. 2025 · Latest: v2026.7.20

Nous Research’s Hermes Agent v0.19.0, dubbed the Quicksilver Release, cuts first-token latency by ~80% — from 4.

3 seconds to ~0.9s — across CLI, TUI, desktop, and cron environments. The update enables reasoning models to stream outputs live by default and overhauls the desktop app with 14× faster markdown streaming, virtualized diffs, and snappy session switching. A durable delivery ledger now ensures responses survive gateway crashes, while Bitwardenables approvals. Users can manage Nous subscriptions directly from the terminal, integrate Bitwarden and 1Password for credential handling, and rely on smart approvals to auto-judge flagged commands. Subagent progress is visible in real time, and the agent maintains cross-session memory via FTS5 search with LLM summarization and periodic knowledge nudges. Built on a closed learning loop, Hermes creates and self-improves skills from experience, models user behavior through Honcho dialectic tracking, and deploys isolated subagents for parallel workstreams. It runs on $5 VPSs, GPU clusters, or serverless backends like Daytona and Modal, which hibernate idle environments to near-zero cost. Scheduled automations, daily reports, and unattended audits operate via natural language triggers.

The catch: Despite its portability and self-improving design, the agent’s reliance on frequent LLM calls for memory search and skill refinement may inflate costs in high-frequency usage scenarios, and its deep personalization model raises unresolved questions about long-term data privacy and user consent in shared or ephemeral environments.

Previously in The Times “covered” — Jul 27

Use Cases
  • Developers automate cloud debugging via Telegram
  • Teams run persistent AI agents on serverless infrastructure
  • Individuals build personalized workflows with local model endpoints

Source: NousResearch/hermes-agent — based on the README and release notes.

More Stories

Gemini CLI Fixes Infinite ReAct Loops and Prompt Injection Risks in v0.53.0 🔗

Security patch and eval tools arrive as weekly stable releases shift to Tuesday 20:00 UTC

google-gemini/gemini-cli · TypeScript · 106.2k stars Est. 2025

The latest Gemini CLI release v0.53.

0 patches infinite ReAct loops and prompt injection vulnerabilities that could trigger unsafe agent behavior. It also adds an eval coverage report command for testing AI workflows and aligns macOS Seatbelt profiles with a deny-default security model. Stable releases now publish weekly at 20:00 UTC on Tuesdays, promoting last week’s preview after validation. Built-in tools like Google Search grounding and file operations remain core to its terminal-first design for developers.
The catch: Despite frequent updates, 984 open issues suggest ongoing stability and feature maturity challenges in complex automation scenarios.

Previously in The Times “covered” — Jul 24

Use Cases
  • Developers debug codebases using natural language prompts
  • Teams automate PR queries and complex rebases via scripts
  • Engineers ground queries with real-time Google Search data

Source: google-gemini/gemini-cli — based on the README and release notes.

Microsoft’s AI Agents Course Adds Local Foundry Support 🔗

Updated lessons now run Agent Framework V2 with Foundry-local for offline prototyping

microsoft/ai-agents-for-beginners · Jupyter Notebook · 70.7k stars Est. 2024

The microsoft/ai-agents-for-beginners repository updated its lessons to use Microsoft Foundry Agent Service V2 via foundry-local, enabling developers to prototype AI agents without Azure dependencies. Code samples in the code_samples folder now demonstrate local execution of agent workflows using Semantic Kernel and AutoGen patterns.

This shift supports offline learning and reduces friction for beginners testing agentic RAG or tool-use scenarios. The course retains its 18-lesson structure covering agent fundamentals, with multilingual notebooks available via sparse checkout.
The catch: Local Foundry lacks enterprise features like managed scaling and monitoring, limiting production readiness.

Previously in The Times “covered” — Jul 23

Use Cases
  • Learn agent prototyping without cloud setup
  • Test RAG pipelines locally using Semantic Kernel
  • Experiment with AutoGen multi-agent conversations offline

Source: microsoft/ai-agents-for-beginners — based on the project README.

ComfyUI Adds GPT-5.6 and Gemini 3.5 Flash via Partner Nodes 🔗

Latest release expands AI model support while optimizing video processing and workflow templates

Comfy-Org/ComfyUI · Python · ▲ 120 in 1d Est. 2023

ComfyUI’s v0.29.

0 release integrates GPT-5.6 and Gemini 3.5 Flash models through its Partner Nodes system, enabling direct API calls from the node graph. The update streams video transcoding to reduce RAM usage and updates workflow templates to v0.11.11. Developers can now send ComfyUI Job IDs in request headers for better job tracking. Despite active development, the project’s 4,362 open issues signal ongoing complexity in maintaining its expansive node ecosystem.
The catch: The sheer volume of open issues raises questions about long-term stability and maintainability for production-critical pipelines.

Previously in The Times “covered” — Jul 24

Use Cases
  • Visual artists generating controlled image sequences via node graphs
  • Developers prototyping multimodal AI workflows with LLMs and diffusion
  • Teams automating video processing pipelines with low-memory transcoding

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

Quick Hits

OpenBB OpenBB-finance/OpenBB provides an open data platform enabling analysts, quants, and AI agents to access, analyze, and act on financial data with transparent, extensible tools. 71.2k
transformers huggingface/transformers offers a unified framework for defining, training, and deploying state-of-the-art ML models across text, vision, audio, and multimodal tasks with minimal code. 163.2k
superpowers obra/superpowers delivers an agentic skills framework and development methodology that lets builders create intelligent, autonomous systems using composable, reusable agent behaviors. 263.6k
learnopencv spmallick/learnopencv teaches OpenCV through practical C++ and Python examples, enabling builders to implement real-time computer vision applications from scratch. 23.1k
dify langgenius/dify lets teams build, test, and deploy agentic workflows and RAG pipelines in a collaborative workspace — with flexible deployment options to scale from prototype to production. 150.8k

ROSClaw adds safety sandbox and evidence logging for embodied AI agents 🔗

New runtime enforces fail-closed execution and returns auditable receipts for physical actions

ros-claw/rosclaw · Python · ▲ 2 in 1d 4mo old

ROSClaw now requires sandbox validation and evidence-based execution for all agent-to-robot interactions. The runtime binds high-level intent to physical capability through its canonical action path: intent → body mapping → policy enforcement → resource arbitration → driver dispatch → ExecutionReceipt. This receipt includes timestamped sensor logs, actuator commands, and failure diagnostics, creating an auditable trail for every physical action.

The system treats memory and skill evolution as asynchronous consumers of evidence, not replacements for real-time safety checks. Agents like Codex, Claude Code, and OpenClaw act as northbound clients, while ROS 2, MCP, and vendor SDKs serve as southbound systems. ROSClaw isolates unsafe operations in a sandbox, defaulting to fail-closed behavior if validation fails.

Recent commits show improved integration with LeRobot 0.6 and the RH56 reference-policy plugin, both requiring Python 3.12+. The core package supports Python 3.11–3.13, with simulation paths kept lightweight by default. Installing rosclaw[knowledge] adds optional semantic services for Know/How model routing.

The catch: ROSClaw remains alpha software with only 8 open issues but no stable release; its narrow focus on ROS 2 and MCP southbound integrations may limit adoption with non-ROS robot stacks or proprietary controllers lacking driver support.

Previously in The Times “covered” — Jul 25

Use Cases
  • Robotics teams validating LLM-generated actions in simulation
  • Autonomous agents requiring audit trails for physical task execution
  • Developers building safety-critical embodied AI with ROS 2 backends

Source: ros-claw/rosclaw — based on the project README.

More Stories

Sesame Robot Adds Voice Control via New Python Companion App 🔗

Latest update enables hands-free interaction and API-driven animations for makers.

dorianborian/sesame-robot · C · 3.8k stars 8mo old

The sesame-robot project recently integrated a Python-based Companion App for voice control and advanced interactions, expanding its remote-control capabilities beyond the existing Serial CLI and web UI. Users can now trigger movements or animations using voice commands, leveraging the ESP32’s WiFi connectivity and JSON API.

Builders appreciate the low cost—$50-60 in parts—and the expressive OLED face that syncs with motion. The project remains accessible to those with basic soldering and Arduino IDE experience. The catch: Voice control depends on a separate Python environment, adding setup complexity for users seeking pure embedded simplicity.

Use Cases
  • Hobbyists programming emotive quadruped behaviors
  • Educators teaching walking-robot kinematics
  • Makers prototyping voice-assisted robotic companions

Source: dorianborian/sesame-robot — based on the project README.

NiceGUI v3.15.0 Adds Popups and Boosts Binding Performance 🔗

New ui.popup component and observable cleanup speed up Python web UIs

zauberzeug/nicegui · Python · ▲ 3 in 1d Est. 2021

NiceGUI’s latest release introduces ui.popup, a responsive menu that adapts to screen size, and optimizes binding cleanup with a reverse index for faster UI updates.

The framework continues to let builders create browser-based GUIs in Python with live reload, supporting dashboards, robotics, and ML tuning. The catch: Observable storage now raises KeyError on missing keys, breaking prior None returns and requiring explicit defaults in storage.pop().

Previously in The Times “covered” — Jul 24

Use Cases
  • Robotics teams tuning motor controllers via browser
  • Data scientists adjusting ML hyperparameters in live Jupyter notebooks
  • Smart home dashboards plotting real-time sensor data

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

RTAB-Map adds RealSense and ZED SDK support in latest Windows binaries 🔗

Release 0.23.8 updates dependencies including OpenCV 4.13.0 and Qt 6.8.3 for SLAM workflows

introlab/rtabmap · C++ · 3.9k stars Est. 2014

The introlab/rtab/ship includes RealSense2 2.56.

5 and ZED SDK 5.4.0 in its Windows builds, expanding sensor kits and standalone app and OpenNI2 and Freenect2 drivers with CUDA pipeline access. Builders using ROS 1 Noetic or ROS 2 Humble through Rolling can access these via precompiled binaries. The catch: OpenNI Kinect for Xbox 360 remains broken on Windows 11 despite working on Windows 10, limiting legacy sensor compatibility.

Use Cases
  • Robotics teams mapping indoor spaces with RGB-D cameras
  • AR/VR developers integrating SLAM into mobile robot prototypes
  • Researchers validating loop closure on ROS 2 Humble platforms

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

Quick Hits

crane_plus CRANE+ V2 provides ROS 2 packages for advanced robotic manipulation, enabling precise control and integration of robotic arms in research and industrial applications. 60
PX4-Autopilot PX4 Autopilot delivers a high-performance, open-source flight stack for drones and VTOL vehicles, supporting robust autonomous navigation and real-time control across diverse aerial platforms. 12.3k
open-source-rover NASA JPL’s Open-Source Rover offers a build-it-yourself 6-wheel Mars rover replica, letting builders explore planetary mobility systems with accessible hardware and educational value. 9.5k
gopherbot Gopherbot is a Go-based ChatOps engine for Slack that automates DevOps workflows, enabling teams to trigger CI/CD, monitor systems, and manage infrastructure via chat commands. 61
OpenBot OpenBot turns smartphones into robot brains by pairing them with a $50 electric vehicle base, enabling advanced AI tasks like person following and autonomous navigation using mobile hardware. 3.4k

OpenCTI Adds Draft Approval Workflow to Streamline Threat Intelligence Collaboration 🔗

Latest release enhances frontend integration UX and TAXII client-side audit logging for session tracking

OpenCTI-Platform/opencti · TypeScript · ▲ 5 in 1d Est. 2018 · Latest: 7.260728.0

OpenCTI’s latest release introduces a draft approval workflow (#17281) that lets teams review and sign off on threat intelligence updates before publication, reducing errors in shared STIX2 data. The frontend now features reworked integration hubs with pipeline ribbons, aligned lines views, and improved XTM Hub connectivity (#17302, #17308), streamlining tool onboarding. Ingestion improvements surface TAXII client errors directly in the UI (#17284), while audit logs now record user session IDs for forensic traceability (#16684).

Built on TypeScript with a GraphQL API and STIX2 schema, OpenCTI remains dual-licensed: Community Edition under Apache 2.0, Enterprise Edition offering advanced features like enhanced MITRE ATT&CK mapping and connector orchestration. The platform continues to support bidirectional sync with MISP, TheHive, and custom datasets, exporting to STIX2 bundles and CSV.

The catch: Despite active development, over 2,000 open issues signal accumulating technical debt, and enterprise features remain gated behind a paid tier that may deter smaller security teams seeking full functionality.

Previously in The Times “covered” — Jul 24

Use Cases
  • Security analysts enriching IOCs with MITRE ATT&CK TTPs
  • SOC teams correlating MISP events with internal telemetry
  • Threat hunters exporting STIX2 bundles for SIEM ingestion

Source: OpenCTI-Platform/opencti — based on the README and release notes.

More Stories

Shannon AI Pentester Releases v2.0.0 with Runtime Overhaul 🔗

Migrates from Claude Agent SDK to pi harness, drops Google Vertex AI support

KeygraphHQ/shannon · TypeScript · 46.3k stars 10mo old

KeygraphHQ's Shannon v2.0.

0 replaces its Claude Agent SDK worker with the pi harness runtime, improving agent execution stability. The release removes Google Vertex AI as a supported provider, requiring users to switch to Anthropic, AWS Bedrock, or a custom Anthropic-compatible endpoint. Shannon analyzes source code to identify attack vectors and executes real exploits via browser automation to produce verified vulnerability reports. It runs locally via npx with Docker and Node.js 18+, offering on-demand penetration testing for every build. The catch: Open issues remain elevated at 26, and the narrow focus on web/APIs limits applicability to non-HTTP systems like CLI tools or desktop apps.

Use Cases
  • Developers testing web apps pre-merge
  • Security teams automating API vuln checks
  • CI/CD pipelines adding exploit-verified scanning

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

Strix v1.4.1 patches TLS calls to fix CI/CD HTTPS failures 🔗

Release swaps urllib for requests to resolve frozen-build certificate errors in automated scans

usestrix/strix · Python · ▲ 3 in 1d 11mo old

Strix v1.4.

1 replaces raw urllib with the requests library for external HTTPS calls, resolving certificate validation failures in frozen build environments. The change improves reliability when scanning dependencies in CI/CD pipelines where custom CA bundles are common. Developers using Strix to block insecure code/tool names are formatted as requests and urllib. Strix remains dependent on Docker and an LLM API key, with autonomous agents validating exploits via live PoCs. Open issues at 245 suggest ongoing usability and integration work.
The catch:** The tool’s effectiveness hinges on LLM API access and Docker, posing barriers for air-gapped or cost-sensitive teams.

Previously in The Times “covered” — Jul 26

Use Cases
  • Security engineers automate vuln scans in PRs
  • Dev teams validate exploits with working PoCs
  • Bug hunters accelerate bounty report generation

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

RustScan adds UDP benchmarks to lock in speed gains 🔗

Release 2.4.1 includes automated CI checks to prevent performance regressions

bee-san/RustScan · Rust · 20.2k stars Est. 2020

RustScan 2.4.

1 now runs UDP and TCP benchmarks on every pull request via automated CI, ensuring no commit slows the scanner. The update also deduplicates IP results and reduces unnecessary clones() to maintain its sub-3-second full-port scan speed. A new Rust library lets developers embed RustScan directly into security tools. Despite six years of active maintenance, 60 open issues remain, including UDP timeout handling fixes still in progress.
The catch: Adaptive learning claims lack published metrics, leaving builders to trust speed gains without verifiable long-term data.

Use Cases
  • Security teams scanning internal networks for open ports
  • Developers integrating port scanning into custom security tools
  • Pentesters automating reconnaissance with scriptable output pipelines

Source: bee-san/RustScan — based on the README and release notes.

Quick Hits

httpx httpx is a fast, multi-purpose HTTP toolkit that runs concurrent probes using retryablehttp for efficient reconnaissance and vulnerability scanning. 10.2k
MISP MISP is an open-source threat intelligence platform enabling organizations to collect, share, and correlate cyber threat indicators in real time. 6.4k
ImHex ImHex is a powerful, retina-friendly hex editor designed for reverse engineers and programmers, offering advanced pattern matching and scripting at 3 AM. 54.3k
bbot bbot is a recursive internet scanner that automates asset discovery and vulnerability hunting for hackers and security teams. 10.2k
berty Berty is a secure peer-to-peer messaging app that functions offline or without internet, cellular, or network trust, ensuring private communication. 9.3k

llama.cpp Adds Native MXFP4 Support for gpt-oss Models, Boosting Efficiency 🔗

Collaboration with NVIDIA enables quantized inference on consumer hardware without quality loss

ggml-org/llama.cpp · C++ · ▲ 132 in 1d Est. 2023 · Latest: b10189

The ggml-org/llama.cpp project has added native MXFP4 format support for the gpt-oss model family, a result of direct collaboration with NVIDIA. This update allows developers to run large language models using 4-bit quantization that preserves accuracy while significantly reducing memory bandwidth and compute demands.

The implementation leverages NVIDIA’s tensor cores on RTX GPUs and is now integrated into the core inference pipeline, enabling faster token generation on consumer-grade hardware.

Recent work also includes multimodal support in llama-server via PR #12898, expanding the server’s ability to process image-text inputs alongside traditional LLMs. WebGPU support has been moved from experimental to stable, allowing browser-based inference with real-time performance demonstrations available on the project site. Additionally, Hugging Face Inference Endpoints now offer out-of-the-box GGUF support, simplifying deployment for teams using managed infrastructure.

Installation remains accessible via Homebrew, Nix, Winget, Conda-forge, Docker, or pre-built binaries for Linux (including s390x and ROCm 7.2), macOS (Apple Silicon and Intel), and iOS. The project continues to optimize for Apple’s Arm architecture using Metal, Accelerate, and NEON, while maintaining broad x86 coverage with AVX/AVX2/AVX512/AMX and RISC-V vector extensions.

The catch: Despite broad hardware support, the project lacks official Windows GPU builds with DirectML or CUDA in the default release pipeline, requiring manual compilation for optimal NVIDIA performance on Windows systems.

Previously in The Times “covered” — Jul 26

Use Cases
  • Run quantized LLMs on consumer laptops
  • Deploy multimodal models via REST API
  • Execute browser-based inference with WebGPU

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

More Stories

React Native 0.86.2 Fixes Layout and Android Runtime Glitches 🔗

Patch resolves `display: contents` rendering and Hermes engine sync issues

react/react-native · C++ · ▲ 7 in 1d Est. 2015

React Native 0.86.

2 patches a layout bug where display: contents nodes incorrectly triggered hasNewLayout, causing UI glitches on Android and iOS. It also replaces unsafe synchronized blocks on java.lang.Boolean with explicit locks in ReactInstanceManager, preventing potential deadlocks. Hermes V1 engine was updated to version 250829098.0.16 for improved JavaScript performance. Despite monthly commits, 1,022 open issues persist, including long-standing threading and native module integration challenges.
The catch: Developers still face steep debugging overhead when bridging native modules due to limited type safety and sparse error context.

Use Cases
  • Build cross-platform apps with shared React logic
  • Access native UI controls without leaving JavaScript
  • Over-the-air JS updates via Expo or CodePush

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

RAGFlow v0.26.4 fixes metadata persistence and parsing bugs 🔗

July release resolves data linking errors and Docling formula extraction failures

infiniflow/ragflow · Go · ▲ 92 in 1d Est. 2023

RAGFlow’s v0.26.

4 release patches critical bugs affecting data integrity and document processing. The update fixes metadata persistence during inline edits and prevents bulk file linking from removing existing dataset links. It also resolves Docling parser drops of mathematical formulas and MCP server crashes from paginated chat responses. Language-aware Snowball stemming now supports 16 languages, including Dutch. These fixes target enterprise RAG workflows where data fidelity and agent reliability are non-negotiable. The catch: Open issues exceed 2,000, signaling ongoing stability challenges despite rapid patching.

Previously in The Times “covered” — Jul 24

Use Cases
  • Enterprises syncing Confluence and Notion data for LLM context
  • Developers building agents with multi-modal PDF/image understanding
  • Teams deploying self-hosted RAG pipelines with DeepSeek v4 support

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

Astral's uv 0.12.0 Enforces Build Systems by Default for New Projects 🔗

The release restores `uv init` to include a build system, aiming for better packaging practices.

astral-sh/uv · Rust · ▲ 51 in 1d Est. 2023

Astral's uv 0.12.

0, released July 28, 2026, changes how new projects are scaffolded by declaring a build system and enabling packaging by default via uv init. This reverses a 2023 decision to omit build systems for newcomer simplicity, now replaced with uv’s own uv_build backend for tighter integration. The update improves correctness and compatibility but marks several workflows as potentially breaking, though most users are expected to upgrade without changes. Configuration for the uv build backend requires an upper bound update if specified.
The catch: The shift assumes users want opinionated tooling, which may frustrate those preferring minimal or custom build setups.

Previously in The Times “covered” — Jul 23

Use Cases
  • Python developers starting a new packaged project with `uv init`
  • Teams adopting uv for faster dependency resolution across macOS, Linux, Windows
  • Tool maintainers publishing CLI utilities via `uv tool install` and `uvx`

Source: astral-sh/uv — based on the README and release notes.

Quick Hits

electron Electron lets developers build cross-platform desktop apps using familiar web technologies like JavaScript, HTML, and CSS — no native expertise required. 122.3k
linux The Linux kernel source tree provides the foundational, highly customizable core powering everything from servers to embedded devices worldwide. 241.1k
vaultwarden Vaultwarden offers a lightweight, Rust-based, self-hosted Bitwarden-compatible password manager with strong security and minimal resource usage. 64.6k
curl cURL is a versatile command-line tool and library for transferring data across nearly every protocol, ideal for automation, testing, and API interactions. 42.5k
memos Memos is a fast, self-hosted, Markdown-native note-taking app built in Go for quick capture and full ownership of your knowledge. 61.9k

Fanpico v1.7.8 adds support for new hardware board and fixes WiFi reliability 🔗

The RP2040-based fan controller gains FanPico-0200 board compatibility and improved network reconnect logic

tjko/fanpico · C · ▲ 1 in 1d Est. 2022 · Latest: v1.7.8

The tjko/fanpico project released version 1.7.8, adding support for the new FanPico-0200 board variant while patching WiFi rejoin logic for the Pico W configuration.

This update refines the open-source PWM fan controller’s ability to manage up to eight fans using temperature feedback from 1-Wire or I²C sensors, with optional OLED/LCD display output.

Built around the Raspberry Pi Pico (RP2040), Fanpico operates as a standalone device requiring no host drivers or OS-specific software. It reads motherboard PWM signals or sensor data to adjust fan speeds via custom curves, stores configuration in flash, and can report RPM or locked-rotor signals back to the system. Networked features include an HTTPS-enabled HTTP server and MQTT client with TLS for telemetry and remote control.

The release notes confirm two concrete changes: explicit support for the FanPico-0200 board and fixes to WiFi reconnection behavior—addressing a known pain point for users deploying the controller in environments with intermittent wireless connectivity.

The catch: While Fanpico excels at localized fan control, its reliance on the RP2040 limits processing headroom for complex sensor fusion or real-time analytics, making it less suitable for multi-variable environmental management beyond basic temperature-triggered PWM adjustments.

Previously in The Times “covered” — Jul 27

Use Cases
  • PC builders tuning fan curves for noise/airflow balance
  • 3D printer operators managing enclosure cooling
  • Network admins controlling rack-mounted equipment fans

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

More Stories

Silhouette-Card-Maker v2.2.0 adds MTG plugin options for Universes Beyond 🔗

New flags let users prefer or exclude special-frame Magic cards during proxy generation

Alan-Cha/silhouette-card-maker · Python · 181 stars Est. 2025

The Alan-Cha/silhouette-card-maker project released v2.2.

0, updating its Magic: The Gathering plugin to respect Scryfall’s rate limits and adding --prefer_ub and --ignore_ub flags. These options let users prioritize or exclude Universes Beyond card versions when generating PDFs for Silhouette cutting machines. The tool still automates card layout, offset calibration, and image acquisition for TCG proxies and custom games. Documentation includes tutorials, supply lists, and cutting templates for standard, poker, and K-pop photocard sizes.
The catch: Open issues remain high at 48, with no indication of maintainer bandwidth to address backlog despite recent activity.

Use Cases
  • Hobbyists creating MTG proxies for casual playtesting
  • Designers printing custom Pokémon or Yu-Gi-Oh! cards
  • Makers producing K-pop photocard sheets on business paper

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

RFLink32 bridges 433MHz sensors to MQTT on ESP32/ESP8266 🔗

Latest commit adds runtime configuration via web interface without recompilation

cpainchaud/RFLink32 · C · 189 stars Est. 2021

RFLink32 decodes OOK 433MHz signals from sensors and alarms, forwarding data to MQTT, TCP, or Serial on ESP32 and ESP8266 boards. It supports advanced receivers like SX1276, SX1278, RFM69, and CC1101 when paired with compatible hardware.

Configuration happens through a web interface or CLI, with settings saved to flash. The project uses PlatformIO and relies on PubSubClient for MQTT and u8g2 for optional OLED displays.
The catch: Active development has slowed, with 55 open issues and no major feature updates in recent commits despite the 2026 timestamp.

Use Cases
  • Home automation hobbyists integrating 433MHz sensors with Home Assistant
  • DIY builders creating wireless alarm systems using ESP32 boards
  • Makers logging sensor data via MQTT for local dashboards or automations

Source: cpainchaud/RFLink32 — based on the project README.

PostgreSQL tuning tool pgtune updates after years of silence 🔗

JavaScript-based config optimizer gains new hardware detection in 2026

le0pard/pgtune · JavaScript · 2.7k stars Est. 2014

The le0pard/pgtune project pushed its first commit in over two years on July 29, 2026, reviving a JavaScript tool that generates postgresql.conf settings from system specs.

It analyzes CPU, RAM, and storage to output tuned parameters like shared_buffers and max_connections, aiming to simplify performance setup for developers. Built as a PWA, it runs offline after installation via Node.js and Yarn. The tool still relies on the original pgtune logic, now wrapped in a modern UI. The catch: No public release or version tag accompanies the commit, leaving users uncertain if the changes are stable or experimental.

Use Cases
  • DevOps tuning PostgreSQL dev environments
  • DBAs optimizing config for new hardware
  • Developers learning performance impact of settings

Source: le0pard/pgtune — based on the project README.

Quick Hits

MiSTer_tty2oled MiSTer tty2oled lets builders add custom text or image displays via Arduino-driven screens to enhance retro gaming setups with real-time visual feedback. 69
axi pulp-platform/axi provides synthesizable SystemVerilog AXI IP and verification tools for creating high-speed, reliable on-chip communication in FPGA/ASIC designs. 1.6k
iiab Internet-in-a-Box enables builders to deploy a self-contained digital library of educational content on Raspberry Pi, creating offline knowledge hubs for underserved communities. 1.9k
automotive-skills-suite automotive-skills-suite offers 100+ installable Claude skills with built-in reviewer validation for automotive engineering standards like ISO 26262 and ASPICE, accelerating compliance and quality workflows. 2.5k
hackrf HackRF is an affordable, open-source software radio platform that lets builders transmit and receive radio signals from 1 MHz to 6 GHz for prototyping wireless systems. 7.9k

Anima Streamlines Godot UI Animations with CSS-Like Syntax 🔗

89 built-in effects and 33 easing options reduce boilerplate for developers

ceceppa/anima · GDScript · 763 stars Est. 2021

Anima simplifies UI animation in Godot by replacing verbose Tween code with a declarative, CSS-inspired syntax. Developers can trigger sequences like `anima.fade_in().

then().slide_up()or run parallel animations with minimal boilerplate. The plugin includes 89 prebuilt animations—such as pulse, bounce, and flip—and 33 easing curves, all accessible via a consistent API. Installation requires copying theaddons/Anima` folder into a project and enabling it in Project Settings, after which animations become available as methods on any Node.

Built as an editor plugin, Anima integrates directly into the Godot workflow, letting designers preview effects without running the game. Its syntax mirrors animate.css, lowering the learning curve for web developers transitioning to game UI. Despite its utility, the project shows signs of aging: the last commit was five years ago, and ten open issues remain unresolved, including reports of inconsistent behavior with Godot 4.x and missing documentation for runtime animation creation.

The catch: Anima’s last update predates Godot 4.0, raising compatibility concerns for new projects using the latest engine version.

Use Cases
  • UI designers prototyping menu transitions
  • Game developers adding feedback to buttons
  • Creators porting web animation knowledge to Godot

Source: ceceppa/anima — based on the project README.

More Stories

EnTT 4.0.0 drops legacy utilities, refines core concepts for modern C++ 🔗

Header-only ECS library removes deprecated functions, adds enum_bitmask and cvref_unqualified concepts

skypjack/entt · C++ · ▲ 7 in 1d Est. 2017

The skypjack/entt project released v4.0.

0, stripping outdated utilities like entt::identity and entt::popcount from its core while introducing modern C++20 concepts such as enum_bitmask and cvref_unqualified. These changes streamline type handling and favor concept-based design over SFINAE, improving compile-time safety and clarity. The update also adds optional configuration injection via <entt/ext/config.h> and refines meta::name to return std::string_view. Widely adopted in projects like Minecraft and ArcGIS Runtime, EnTT remains a dependency-free, header-only ECS solution for game engines and data-oriented systems.
The catch: Despite its maturity, 13 open issues suggest ongoing maintenance needs, and its narrow focus on ECS may limit appeal for developers seeking broader game engine features.

Use Cases
  • Game developers building entity-component systems in C++
  • Engineers implementing data-oriented design in performance-critical applications
  • Teams needing a header-only, zero-dependency runtime reflection system

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

Assimp 6.0.5 Fixes Haiku Builds, Memory Use, and GLTF Skinning Export 🔗

Latest patch resolves build issues on niche OS, trims vertex processing memory, and extends GLB/GLTF skinning data support.

assimp/assimp · C++ · 13.1k stars Est. 2010

The Open-Asset-Importer-Library (assimp) released v6.0.

5, addressing a Haiku build failure via pull request #5255 and significantly reducing memory consumption in JoinVerticesProcess::ProcessMesh() (#5252). Skinning data export to GLB/GLTF format was extended (#5243), improving glTF 2.0 workflow compatibility. Additional fixes include FBX floating-point output (#5265), implicit conversion errors (#5271), and missing normal transformations (#5301). Despite active maintenance, over 500 open issues and a C++-centric codebase, assimp remains a critical asset pipeline tool for cross-platform 3D format unification.
The catch: Its broad format support comes with complexity — developers report a steep learning curve for advanced post-processing tuning and inconsistent behavior across lesser-used formats like IFC or 3MF.

Use Cases
  • Game developers importing FBX/glTF models into Unity or Unreal via assimp plugins
  • Android/iOS apps loading diverse 3D assets at runtime using assimp’s mobile bindings
  • CAD/conversion tools translating between 40+ formats like STL, COLLADA, and 3MF for interoperability

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

RenoDX Adds Shader Decompiler Tool for DirectX HLSL Modding Workflows 🔗

New decomp.exe utility enables inspection of Shader Model 6.0+ binaries for addon development

clshortfuse/renodx · HLSL · ▲ 1 in 1d Est. 2024

RenoDX updated its devkit with decomp.exe, a Shader Model 6.

0+ HLSL decompiler to aid addon creators in analyzing existing shaders. The tool outputs readable code from compiled binaries, helping modders reverse-engineer effects for replacement or enhancement. Built on Reshade’s hooking framework, RenoDX avoids version-specific executable patches, broadening compatibility across DirectX 9-12 games. Recent commits focus on stabilizing buffer injection and overlay APIs, with 74 open issues indicating active refinement.
The catch: The toolchain assumes familiarity with HLSL and Reshade’s addon system, creating a steep on-ramp for developers new to graphics modding.

Use Cases
  • Graphics modders replace game shaders to add HDR or ray-traced effects
  • Developers build custom overlays using RenoDX’s buffer injection interface
  • Teams decompile legacy shaders to understand and update visual effects pipelines

Source: clshortfuse/renodx — based on the project README.

Quick Hits

tiled Tiled is a flexible, cross-platform level editor for creating 2D game maps with tile, object, and layer support — ideal for rapid prototyping and professional workflows. 12.8k
Super-Mario-Bros.-Remastered-Public Super Mario Bros. Remastered Public recreates the classic with new levels, characters, and modes, plus a full custom level editor for endless player-driven creativity. 2.9k
beehave Beehave provides a visual, node-based behavior tree system for Godot, enabling complex, reusable AI logic without code-heavy scripting. 3.2k
OpenSpeedy OpenSpeedy is a lightweight, open-source TypeScript tool that lets players dynamically adjust game speed in real time for testing, accessibility, or fun. 16.8k
Super-Duper-Vanilla Super Duper Vanilla is a GLSL shader pack that faithfully recreates the vibrant, stylized look of Minecraft’s cancelled Super Duper Graphics Pack and similar fan favorites. 168
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