Design
Preset
Background
Text
Font
Size
Width
Account Pricing

The Git Times

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

holaOS Unifies AI Agents in a Local-First Workspace with Shared Memory 🔗

Runs Claude Code, Codex, or custom agents side-by-side using shared context and 100+ tool integrations

holaboss-ai/holaOS · TypeScript · ▲ 9.5k in 1d 4mo old · Latest: latest

holaOS provides a local-first AI agent workspace where developers can run multiple agents — including Claude Code, Codex, and its own holaOS agent — simultaneously without context switching. Built on TypeScript and Electron, it offers shared memory, tools, skills, and app integrations across all agents, ensuring consistent behavior regardless of which model is driving the task.

The workspace supports over 100 integrations via the Model Context Protocol (MCP), allowing agents to interact with files, browsers, and external apps as interactive surfaces. Memory is stored locally as plain, editable files, enabling durable context across sessions and agent switches — users can close the app and return days later to find the agent aware of prior work. Frontier models like GPT 5.6, Claude Opus 5, and Fable 5 are built in for heavy lifting, while efficient models like Kimi K3 and GLM 5.2 handle routine tasks. Users can also bring their own keys (BYOK) for OpenAI, Anthropic, or compatible endpoints, keeping costs and usage tied to their personal accounts. Agents install from an in-workspace marketplace and run as live, interactive panels where users can step in, observe, and edit outputs directly — results appear in place, not as detached chat logs.
The catch: Despite active development, the project has only seven open issues and a narrow scope focused on desktop agents, leaving open questions about its viability in team workflows, mobile environments, or production-grade agent orchestration at scale.

Why this leads today It offers a unified, open-source platform for running and coordinating AI agents across tools and environments with shared memory, addressing a fragmented workflow developers currently face.

Use Cases
  • Developers testing agent performance across coding tasks
  • Power users automating multi-tool workflows with persistent context
  • Researchers comparing LLM behaviors in shared environments

Source: holaboss-ai/holaOS — based on the README and release notes.

More on the Front Page

Coolify 4.3.7 Fixes Critical Backup and Deployment Flaws for Self-Hosted PaaS 🔗

Latest patch resolves multi-database dump failures and S3 backup issues on non-root SSH setups

coollabsio/coolify · PHP · 60.7k stars Est. 2021

Coolify, the self-hosted PaaS alternative to Vercel and Heroku, released v4.3.

7 with targeted fixes for database and deployment reliability. The update ensures complete PostgreSQL, MySQL, and MariaDB dumps when "All databases" is selected, preserving multi-database choices and respecting storage settings during backup deletion. It also restores backup pages for apps using directory-only targets and fixes S3-compatible volume backups on servers accessed via non-root SSH users — a common setup in production environments. Deployment cancellation now works consistently across API, MCP, and preview flows, advancing the next queued job after abort. These fixes address long-standing gaps in backup integrity and operational safety for teams managing databases and services on their own infrastructure. Coolify supports 280+ one-click services including Next.js, Svelte, Laravel, and Dockerized apps, letting users deploy static sites, databases, and full-stack apps on VPS, bare metal, or Raspberry Pi via SSH. The project remains PHP-driven with Docker Compose orchestration, emphasizing zero vendor lock-in by saving all configurations to the user’s server.

The catch: Despite active development, 770 open issues suggest ongoing stability challenges, and the PHP/Laravel stack may deter teams preferring modern runtimes like Go or Node.js for self-hosted tooling.

Use Cases
  • Deploy Laravel apps with MySQL on a $5 VPS
  • Host static Svelte sites alongside Redis cache
  • Run self-hosted GitLab with PostgreSQL backups to S3

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

Vision plugin gives DeepSeek agents pixel-level sight without API keys 🔗

Enables multi-step image reasoning using system tools like Tesseract and Chrome

ysr666/dsh-vision-router · JavaScript · 684 stars 4d old

The ysr666/dsh-vision-router plugin equips text-only DeepSeek Harness agents with built-in vision capabilities that require no API key or Python setup. It routes image turns through fourteen pixel-level tools — including OCR, grounding, crop, and pixel diff — while preserving original pixels for the vision model and keeping DeepSeek as the reasoning engine.

Installed with one command, it treats image processing as ordinary tool calls, enabling iterative workflows like vision_ground → vision_crop → vision_describe → fix → screenshot. The plugin uses sharp, potrace, Tesseract, and system Chrome locally, with a free anonymous fallback via OVHcloud models at ~10 RPM.
The catch: Vision output loops can masquerade as success if not guarded, requiring careful tool chaining to avoid degenerate repetition in multi-step reasoning.

Use Cases
  • Developers debugging UI layouts by comparing pixel differences across versions
  • Agents extracting structured data from screenshots via OCR and grounding
  • Automated agents iterating on image crops until visual goals are met

Source: ysr666/dsh-vision-router — based on the README and release notes.

J-Space Cuts DeepSeek V4 Agent Failures by Fixing Reasoning Gaps 🔗

Benchmarks show J-Space reduces capability-realization loss on Flash and Pro variants without model changes.

The J-Space Cognition Suite V3.6 reduces DeepSeek V4’s agent-task failures by addressing mismatches in reasoning modes, tool schemas, and state validation.

Benchmarks indicate Flash performance nears GLM5.3 while Pro exceeds Fable 5, despite identical model weights. J-Space uses workspace loading, selective routing, and verification loops to convert latent capability into stable task completion. It does not compress thinking or extend chains but structures short judgment-action-reasoning-verify-recover cycles. The catch: Open issues and narrow scope suggest real-world reliability beyond controlled benchmarks remains unproven.

Use Cases
  • Developers deploying DeepSeek V4 agents in complex tool-use workflows
  • Teams reducing hallucination in long-context reasoning tasks
  • Engineers improving agent reliability without retraining models

Source: Tiger3807861189/DeepSeek-V4-J-Space-Capability-Realization-Report — based on the README and release notes.

Sunrise Mod Lets Destiny 2 Players Explore Worlds Offline 🔗

Built in C++ with ImGui and Detours, it enables noclip and destination loading without online servers

stanuwu/Sunrise · C++ · 517 stars 5d old

The stanuwu/Sunrise project is a C++ mod that enables offline exploration of Destiny 2 destinations by patching an old game build at runtime. It uses Microsoft Detours for hooking and Dear ImGui for UI overlays, allowing players to fly, noclip, and override activities in locations like the Tower or European Dead Zone.

Basic inventory management and HUD elements are included, but core gameplay such as missions, enemies, NPCs, and persistent saves remains unsupported. The mod is actively developed, with recent updates fixing hunter model loading and adding debug overlays and a logo to the menu.
The catch: Most core Destiny 2 gameplay features are not functional, limiting use to map exploration and experimentation rather than playable content.

Use Cases
  • Modders studying Destiny 2 world structure offline
  • Players exploring destinations without online connectivity
  • Developers testing client-side rendering and memory manipulation techniques

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

Cumora Turns AI Agents Into Full Team Chat Teammates 🔗

Agents claim work, coordinate, and send real email using Claude Code or Codex brains

yetone/cumora · TypeScript · 1.9k stars 0d old

Cumora is a cross-platform team chat where AI agents work alongside humans in shared rooms, DMs, Kanban boards, and calendars. Agents run on managed cloud pods or locally via BYOA using your own Claude Code or Codex CLI, with no provider keys exposed to the server.

Built with TypeScript, React, and Node, it uses Postgres and Redis for state and coordination, preventing agent collisions via cursor gates and atomic work claims. The frontend spans web, desktop, and mobile from a single UI layer.
The catch: Early-stage with 4 open issues; real-world agent coordination at scale remains unproven.

Use Cases
  • Dev teams assign bug fixes to AI agents in chat
  • Support bots handle email and ticket triage autonomously
  • Researchers run local agents to coordinate literature reviews

Source: yetone/cumora — based on the project README.

Temporary Agent Preset Boosts DeepSeek V4 Flash in Opencode-go 🔗

Early adaptation validates Flash model gains under guided conditions despite newer routing modes

SheberDavid/v4-flash-godmode-opencode-go · JavaScript · 505 stars 3d old

SheberDavid/v4-flash-godmode-opencode-go provides a dsh agent preset that switches DeepSeek V4 Flash from "ghost mode" to "god mode" in opencode-go. Based on the original dsh-routing-suite, it enables Flash model performance gains under guided conditions.

The project confirms it is a temporary adaptation of the author’s early version, not tracking later updates that introduced standard and spec routing modes.
The catch: This preset reflects an early version and does not include the author’s newer routing designs, limiting its relevance for current workflows.

Use Cases
  • Developers testing DeepSeek V4 Flash in opencode-go
  • Teams validating agent behavior under guided conditions
  • Users experimenting with dsh-based routing presets

Source: SheberDavid/v4-flash-godmode-opencode-go — based on the project README.

AI Agents Shift from Solo Tools to Collaborative, Memory-Enabled Teams 🔗

Open source projects now enable agent swarms with shared context, pluggable skills, and cross-platform orchestration.

Trendai-agents
yetone/cumoraccch1mneyyy/dsh-TUITiger3807861189/DeepSeek-V4-J-Space-Capability-Realization-ReportNanmiCoder/dsh-agent-teamsSheberDavid/v4-flash-godmode-opencode-gomohitagw15856/pm-claude-skillsholaboss-ai/holaOSdeepset-ai/haystackdust-tt/dustPrimeIntellect-ai/prime-agentbojieli/ai-agent-bookTencentCloud/TencentDB-Agent-Memorycorsairdev/corsairyc-software/qmayghri/i-have-adhdDevin-AXIS/iPolloWorkcalesthio/OpenMontageHKUDS/CLI-Anythingtt-a1i/archifyblader/humanizeralibaba/open-code-reviewwhiteguo233/OpenBiliClawopenchamber/openchamberkangarooking/cangjie-skillinternet-court/internet-court-skillkunchenguid/firstmatehuangruiteng/loopxtrycompai/crmAML-memory/agent-memory-leaderboardAntigmaLabs/anteesengine/DeepSeek-Reasonixvercel-labs/agent-browsergastownhall/beadsSnailclimb/JavaGuidemicrosoft/ai-agents-for-beginnersalirezadir/AIMLInterviewsAtsushiSakai/PythonRoboticsSollimann/bonsaicurl/curljustrach/codedb

The open source AI agent landscape is rapidly evolving beyond single-purpose assistants toward coordinated, memory-backed teams. Projects like yetone/cumora and NanmiCoder/dsh-agent-teams position agents as first-class teammates in chat environments, supporting BYOK brains such as Claude Code or Codex.

Memory is becoming a shared utility: TencentCloud/TencentDB-Agent-Memory structures conversations, skills, and code into governed, reusable assets accessible across agents and frameworks. Skill ecosystems are modularizing — mohitagw15856/pm-claude-skills offers over a thousand plain-markdown agent skills installable via npx, while holaboss-ai/holaOS provides a unified workspace with 100+ integrations and shared memory for running any agent across apps and browsers. Orchestration layers are maturing: deepset-ai/haystack enables production-grade agent workflows with explicit control over retrieval and routing, and prime-agent focuses on self-improving agents for long-running tasks. Even niche needs are addressed — ayghri/i-have-adhd tweaks agent output for ADHD-friendly clarity, and internet-court/internet-court-skill introduces trust mechanisms for agent-to-agent commerce.

The catch: Despite momentum, fragmentation persists — agents often lock into specific harnesses (Claude Code, Codex), memory formats lack standardization, and skill interoperability remains unproven at scale, raising concerns about vendor lock-in disguised as openness.

Use Cases
  • Developers debug code using agent teams with shared context
  • Enterprises deploy governed AI workflows across internal tools
  • Individuals install agent skills for daily task automation

Open Source LLM Tooling Shifts to Modular Agent Ecosystems 🔗

Projects unify memory, skills, and routing to turn LLMs into programmable teammates across workflows

Trendllm-tools
yetone/cumoraccch1mneyyy/dsh-TUImohitagw15856/pm-claude-skillsholaboss-ai/holaOSdeepset-ai/haystackdust-tt/dustcathrynlavery/diagram-designTencentCloud/TencentDB-Agent-Memoryzhaoxuya520/reverse-skilllidge-jun/opencodexDevin-AXIS/iPolloWorkdiegosouzapw/OmniRouteZhuLinsen/daily_stock_analysisvirgiliojr94/book-to-skillcoreyhaines31/marketingskillsasuojun/claude-vision-skillalibaba/open-code-reviewcrawfordxx/xiaoma-durex-copywriterfreestylefly/awesome-gpt-image-2internet-court/internet-court-skillNVIDIA-NeMo/Switchyarddifferent-ai/openworkhuangruiteng/loopxAntigmaLabs/anteMadsLorentzen/ai-job-searchAOrbitron/Eridanuslnxjedi/gopherbotjustrach/codedb

The open source LLM landscape is rapidly coalescing around modular, interoperable agent tooling that treats LLMs not as black boxes but as pluggable components in larger systems. Repos like yetone/cumora and holaboss-ai/holaOS position AI agents as first-class collaborators in shared workspaces, complete with team chat and cross-tool memory.

Skill marketplaces are emerging: mohitagw15856/pm-claude-skills offers over a thousand ready-to-install agent skills in markdown, while virgiliojr94/book-to-skill automates turning technical PDFs into executable agent capabilities. Routing and provider abstraction are gaining traction — lidge-jun/opencodex and diegosouzapw/OmniRoute let teams swap LLMs behind a unified CLI or API, with token optimization and fallback logic. Memory layers are being productized: TencentCloud/TencentDB-Agent-Memory structures agent knowledge into governed, reusable assets like Skill and LLM-Wiki. Even niche domains are getting specialized tooling, from zhaoxuya520/reverse-skill for security workflows to dust-tt/dust for custom agent orchestration. This shift reflects a maturing belief: the real value lies not in the model itself, but in the ecosystem of tools, skills, and infrastructure that make agents reliable, extensible, and embeddable in real work.
The catch: Much of this tooling remains fragmented by framework (Claude Code vs. Codex vs. OpenCode), with overlapping skill formats and memory schemas that hinder true portability; many projects are still early prototypes lacking long-term governance or clear paths to interoperability beyond demo integrations.

Use Cases
  • Developers install agent skills via `npx` to automate PRD writing
  • Teams route LLM calls across providers to optimize cost and latency
  • Enterprises deploy shared agent memory to preserve institutional knowledge across shifts

Open Source Data Infrastructure Shifts Toward Modular, Agent-Ready Tooling 🔗

Projects increasingly expose data flows as programmable interfaces for AI agents and composable pipelines

Trenddata-infra
ccch1mneyyy/dsh-TUIairbytehq/airbytecoollabsio/coolifyXiaoYouChR/Ghost-Downloader-3ZhuLinsen/daily_stock_analysistt-a1i/archifyt8y2/dbxcoreyhaines31/marketingskillslarashero3-dotcom/lieflat-chartsbrightdata/clirepowise-dev/repowiseOpenLabs-so/openanalyticsVexDB-THU/VexDB-LiteSnailclimb/JavaGuidescikit-learn/scikit-learngoogle-ai-edge/mediapipecurl/curlprotocolbuffers/protobufduckdb/duckdb

A clear pattern is emerging in open-source data infrastructure: tools are evolving from monolithic pipelines into modular, agent-interoperable components designed for real-time, AI-driven workflows. This shift prioritizes exposing data movement, transformation, and querying as accessible, self-describing interfaces—often via MCP (Model Context Protocol) or similar agent-facing APIs—enabling LLMs and autonomous agents to discover, invoke, and compose data functions dynamically.

Airbyte (airbytehq/airbyte) exemplifies this by positioning itself as open-source data movement for ELT pipelines and AI agents, explicitly supporting agent-driven orchestration. Similarly, Repowise (repowise-dev/repowise) offers codebase intelligence via MCP, letting agents assess health, generate docs, and detect dead code programmatically. Bright Data’s CLI (brightdata/cli) enables terminal-based scraping and structured extraction—ideal for agent-led web data acquisition. DuckDB (duckdb/duckdb) provides an in-process analytical SQL engine that agents can query directly without external services, while t8y’s dbx delivers a lightweight, AI-assisted cross-platform DB client with built-in MCP server functionality.

Even visualization and analysis tools are adapting: Larashero’s lieflat-charts generates interactive HTML charts for AI agents to consume and present insights, and ZhuLinsen’s daily_stock_analysis combines real-time news, multi-source data, and LLM-driven dashboards for automated financial reasoning. These aren’t just tools—they’re becoming skills in an agent-accessible toolkit.

The catch: Much of this agent-ready infrastructure remains experimental, with inconsistent MCP implementations, limited agent tooling maturity, and unclear governance around data provenance and security when agents autonomously invoke pipelines—raising concerns about reliability and auditability in production AI workflows.

Use Cases
  • Data engineers orchestrating ELT flows via AI agents
  • DevOps teams deploying self-hosted analytics with agent telemetry
  • Financial analysts automating market research using LLM-driven data pipelines

Deep Cuts

AOrbitron/Eridanus Turns LLMs Into Desktop Companions With Live2D Flair 🔗

It merges OneBot protocol with function calling for smart, QQ-free bot interactions

AOrbitron/Eridanus · Python · 200 stars

AOrbitron/Eridanus is a Python framework that fuses LLM function calling with the OneBot 11 protocol to create intelligent, extensible bots. Its standout feature is a pure Live2D desktop pet mode that runs without QQ, letting developers attach Gemini or OpenAI APIs to animate avatars that respond intelligently to voice or text.

Built for tinkerers, it enables custom skills via function calls — think calendar checks, smart home triggers, or context-aware replies — all while rendering expressive 2D characters. The project lowers the barrier to building personal AI companions that feel alive, not just responsive.
The catch: It's early-stage with sparse documentation and a niche focus on desktop pet use cases over broader bot frameworks.

Use Cases
  • Developers creating LLM-powered desktop assistants with animated avatars
  • Hobbyists building QQ-independent bots using Gemini or OpenAI APIs
  • Makers prototyping interactive AI companions with custom function skills

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

Quick Hits

dsh-agent-teams AgentTeams plugin for DeepSeek Harness enables collaborative AI agents to coordinate complex tasks, boosting automation efficiency for developers building scalable multi-agent systems. 516
TG-Polymarket-bot Telegram bot monitors Polymarket whale trades in real time, letting users copy high-impact bets with one tap — plus free public alerts for transparent, actionable market insights. 1k
airbyte Airbyte provides open-source ELT pipelines to move data from any source to warehouses, lakes, or AI apps — self-hosted or cloud — powering reliable, flexible data infrastructure for AI-driven workflows. 21.9k
Who shipped it

The Business Desks

Fresh on Hugging Face

Model Drops

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

The AI Wire

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

From the labs & arXiv

OpenClaw v2026.7.1-2 fixes npm plugin metadata parsing for smoother updates 🔗

The patch resolves singleton-array handling in official plugin installs for Node.js-based AI assistants

openclaw/openclaw · TypeScript · ▲ 91 in 1d 8mo old · Latest: v2026.7.1-2

OpenClaw’s latest release v2026.7.

1-2 addresses a critical npm compatibility issue where newer clients send singleton-array metadata that broke official plugin updates. The fix ensures plugins install and update to correction releases without manual intervention. Built in TypeScript, OpenClaw runs a local Gateway that connects AI models to messaging channels like WhatsApp, Slack, and Signal, with optional companion apps for voice, camera, and screen access. Onboarding verifies model access and configures the workspace automatically. The assistant treats inbound messages as untrusted input and requires explicit approval for pairing unknown senders on DM-capable channels. Tools execute on the host unless sandboxing is configured, per the security guide. With over 81,000 forks and active development—last commit zero days ago—the project maintains a self-hosted, privacy-first approach to personal AI. The catch: Despite broad channel support, sandboxing remains opt-in and complex, leaving host-exposed tools a risk for users sharing instances or connecting untrusted models without deep config review.

Previously in The Times “covered” — Aug 12

Use Cases
  • Developers running private AI assistants across messaging platforms
  • Privacy-focused users avoiding cloud-based model data retention
  • Teams extending local tools with voice, camera, or screen automation via companion nodes

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

More Stories

Microsoft’s AI Agents Course Teaches Foundry Integration Through Jupyter Notebooks 🔗

18 lessons cover agentic RAG and Semantic Kernel for builders with Azure access

microsoft/ai-agents-for-beginners · Jupyter Notebook · 72.5k stars Est. 2024

The microsoft/ai-agents-for-beginners repository offers 18 Jupyter Notebook lessons focused on building AI agents using Microsoft’s Agent Framework and Foundry Agent Service V2. Each lesson includes runnable code examples in the code_samples folder, covering topics like agentic-ai, agentic-rag, and autogen.

The course assumes prior familiarity with generative AI and requires an Azure account to access Foundry services. While translated into 50+ languages, the core technical content remains English-centric in code and tooling. The catch: The course locks builders into Microsoft’s ecosystem, offering limited guidance for porting agents to open-source or multi-cloud environments.

Use Cases
  • Learn to build RAG-powered agents using Semantic Kernel
  • Prototype AI workflows with Azure Foundry Agent Service
  • Understand agentic design patterns via guided Jupyter exercises

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

Microsoft MCP Curriculum Adds Rust Examples for Secure AI Workflow Integration 🔗

Cross-language tutorials now cover session setup to service orchestration in .NET, Java, TypeScript, JavaScript, Rust, and Python.

microsoft/mcp-for-beginners · Jupyter Notebook · 17k stars Est. 2025

The microsoft/mcp-for-beginners repository provides hands-on Jupyter Notebook lessons teaching developers how to implement the Model Context Protocol for modular AI systems. Recent updates expanded Rust coverage alongside existing language tracks, emphasizing secure client-server communication and tool integration.

Learners progress from basic session handling to orchestrating multi-service workflows using standardized MCP interfaces. The curriculum assumes foundational programming knowledge and focuses on practical implementation over theory.
The catch: Open issues indicate gaps in advanced error handling and production deployment guidance beyond basic examples.

Use Cases
  • Backend engineers implementing AI agents that access databases and APIs
  • Full-stack developers building modular chatbots with external tool connectivity
  • Systems architects designing secure, scalable AI workflows across language boundaries

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

Spec Kit Adds Architecture Governance Extension to Community Catalog 🔗

New preset streamlines compliance checks for system design patterns in AI-assisted workflows

github/spec-kit · Python · 129.9k stars 12mo old

GitHub’s spec-kit project released v0.16.

4, adding the Architecture Governance extension to its community catalog. This preset enables teams to enforce design standards—like layering rules or interface contracts—during spec-driven development with AI coding agents. The update also includes fixes for PowerShell output handling, task-ID regex parsing, and workflow validation, improving reliability across environments. Built around the specify-cli tool, the kit turns specifications into executable guides that generate code via agents like Copilot or Mistral. It requires uv for installation and supports extensibility through presets and bundles.
The catch: The project’s rapid activity—328 open issues and frequent breaking adjustments—suggests ongoing instability that may challenge adoption in regulated or long-term maintenance contexts.

Previously in The Times “covered” — Aug 12

Use Cases
  • Teams enforce architectural rules during AI-assisted coding
  • Developers generate specs that auto-produce working code
  • Organizations standardize spec-driven workflows across squads

Source: github/spec-kit — based on the README and release notes.

Quick Hits

scikit-learn Scikit-learn provides a comprehensive, easy-to-use library for implementing machine learning algorithms in Python, enabling rapid prototyping and deployment of predictive models. 67k
mediapipe MediaPipe offers cross-platform, customizable ML pipelines for real-time processing of video, audio, and sensor data, ideal for building responsive, intelligent media applications. 36.6k
firecrawl Firecrawl delivers a scalable TypeScript-powered API to search, scrape, and interact with web content at scale, turning unstructured web data into structured context for AI systems. 168.8k
gemini-cli Gemini CLI brings the full power of Google’s Gemini models directly into your terminal, enabling AI-driven automation, code generation, and query handling without leaving the shell. 106.5k
examples TensorFlow Examples provides a curated collection of Jupyter notebooks demonstrating practical implementations of deep learning models across vision, NLP, and time-series tasks. 8.3k

ROS 2 Control Framework Gains Real-Time Docker Support for Humble and Jazzy 🔗

Pre-built images now streamline deployment across ROS 2 distros without source builds

ros-controls/ros2_control · C++ · 974 stars Est. 2017

The ros-controls/ros2_control project has published official Docker images for its latest releases, tagged ghcr.io/ros-controls/ros2_control_release and `ghcr.

io/ros-controls/ros2_control_source`. These images cover Humble, Jazzy, and Rolling distros, enabling developers to spin up control-loop environments instantly. Built from C++ core, the framework provides a generic interface for robot controllers, hardware abstraction, and real-time safety checks — critical for industrial and research robots. Contributors can now test PRs against standardized environments using the same images used in CI, reducing "works on my machine" friction. The project maintains active support across ROS 2 releases, with documentation auto-generated per distro branch. Despite steady activity — last commit zero days ago and 468 forks — 121 open issues indicate ongoing challenges in controller plugin compatibility and real-time performance tuning under varying hardware loads.
The catch: Real-time performance guarantees depend heavily on underlying OS and hardware; the framework does not harden kernels or isolate CPUs, leaving latency tuning to the deployer.

Use Cases
  • Robotics teams testing control loops in CI
  • Researchers deploying ROS 2 controllers on edge hardware
  • Engineers standardizing hardware interfaces across robot fleets

Source: ros-controls/ros2_control — based on the project README.

More Stories

PythonRobotics updates SLAM and path planning with active maintenance 🔗

Last commit today shows ongoing fixes despite decade-old codebase and 50 open issues

AtsushiSakai/PythonRobotics · Python · 30.3k stars Est. 2016

The AtsushiSakai/PythonRobotics repository provides Python implementations of core robotics algorithms including SLAM, path planning, and control. Recent activity shows a commit within the last day, indicating active maintenance of this ~10.

4-year-old project. It covers localization (EKF, particle filter), mapping, and navigation techniques like RRT and LQR for use in autonomous vehicles and robotics research.
The catch: 50 open issues suggest unresolved bugs or feature gaps that may affect reliability in production systems.

Use Cases
  • Students learning robotics algorithms via runnable Python code
  • Researchers prototyping SLAM and path planning solutions
  • Developers building autonomous navigation prototypes in simulation

Source: AtsushiSakai/PythonRobotics — based on the project README.

NiceGUI v3.16.0 patches memory leaks and XSS flaws in Python web UI toolkit 🔗

Security fixes address socket disconnect leaks and color picker injection vulnerabilities

zauberzeug/nicegui · Python · 16.1k stars Est. 2021

The latest release of NiceGUI patches two security flaws: unauthenticated memory exhaustion via leaked client sockets and cross-site scripting through ColorPicker.set_color() and `ui.

color_input`. Developers building dashboards, robotics interfaces, or ML tuning tools with this Python-based web UI framework should update immediately. The framework still enables rapid creation of browser-based GUIs with live reload, 3D scenes, and data binding using standard Python syntax.
The catch: Open issues remain at 75, indicating ongoing maintenance demands despite recent activity.

Previously in The Times “covered” — Aug 11

Use Cases
  • Engineers tuning motor controllers via browser interface
  • Data scientists adjusting ML hyperparameters in real time
  • Robotics teams monitoring sensor streams on lightweight dashboards

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

RobotWebTools Updates ROS JavaScript Library with Action Cancel Feature 🔗

Latest release adds `Action.cancelAllGoals()` method for improved robot task management

RobotWebTools/roslibjs · TypeScript · 827 stars Est. 2013

RobotWebTools' roslibjs library released version 2.1.

0, introducing Action.cancelAllGoals() to terminate all active robot actions simultaneously. The TypeScript monorepo maintains WebSocket-based ROS communication for browser-based robotics interfaces. Recent commits show active dependency maintenance via Dependabot, including TypeScript and Vite updates. Despite 27 open issues, the project sees daily commits indicating sustained community engagement. The catch: Browser-based ROS clients still face latency and reliability challenges compared to native C++/Python implementations for time-critical robotic control systems.

Use Cases
  • Web developers building ROS dashboard interfaces
  • Robotics engineers creating browser-based teleoperation tools
  • Educators teaching ROS concepts through interactive web applications

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

Quick Hits

bonsai Enables deterministic AI behavior trees in Rust with Python bindings for flexible, high-performance agent control. 969
navigation2 Provides a robust, production-ready ROS 2 navigation stack for reliable robot path planning and obstacle avoidance. 4.6k
IsaacLab Offers a unified, Isaac Sim-powered framework to accelerate robot learning research with seamless simulation integration. 7.9k
newton Delivers GPU-accelerated physics simulation via NVIDIA Warp for high-fidelity robotic dynamics and interaction modeling. 5.5k
isaac_mission_dispatch Implements a VDA5050-compliant cloud service for scalable, standardized fleet mission dispatch in industrial automation. 104

OpenCTI's On-Demand Deployment Speeds Feature Branch Testing for Threat Intel Teams 🔗

New CI enhancement allows developers to spin up isolated instances from built images without full redeployment

OpenCTI-Platform/opencti · TypeScript · ▲ 9 in 1d Est. 2018 · Latest: 7.260817.0

OpenCTI-Platform/opencti has added on-demand deployment for already-built feature branch images, streamlining how threat intelligence teams test changes. The enhancement, released in version 7.

260817.0, lets developers trigger isolated OpenCTI instances directly from CI pipelines using pre-built Docker images—eliminating wait times for rebuilds during iterative work on connectors, schemas, or UI components. This targets a core friction point in CTI platform development: validating STIX2 data integrations or GraphQL API tweaks without disrupting shared environments. The platform, built in TypeScript and aligned with STIX2 standards, continues to support integrations with MISP, TheHive, and MITRE ATT&CK through its connector framework. Recent bug fixes addressed critical backend issues, including malformed STIX message handling that blocked worker queues and upsert logic flaws that cleared createdBy references when confidence scores increased. These fixes improve data integrity in collaborative analysis workflows where attribution and source tracking are paramount. The dual CE/EE model remains, with Enterprise features unlocked via platform settings.
The catch: Despite active development, 2,086 open issues signal persistent complexity in balancing CTI schema strictness with real-world data variability—a trade-off teams must evaluate when deploying at scale for production threat hunting.

Previously in The Times “covered” — Aug 13

Use Cases
  • SOC analysts enriching IOCs with TTPs from MITRE ATT&CK
  • Threat researchers linking malware observables to victimology reports
  • DevSecOps automating CTI data exchange between MISP and SIEM tools

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

More Stories

Cybersecurity Projects repo offers 70 hands-on learning tools 🔗

Foundations tier uses single-file Python builds with heavy teaching comments

CarterPerez-dev/Cybersecurity-Projects · Go · 5.1k stars 9mo old

CarterPerez-dev/Cybersecurity-Projects provides 70 tiered cybersecurity projects from Foundations to Advanced, each with full source code. The Foundations level uses single-file Python scripts with inline annotations and Numpy-style docstrings to teach concepts from zero.

Structured career paths include certification guides for SOC Analyst, Pentester, and Security Engineer roles. Resources cover tools, courses, certifications, and frameworks, and communities via the CertGames platform. Last pushed August 2026, the repo shows active maintenance with 735 forks. The catch: Advanced projects assume Go proficiency, creating a language jump for Python-focused beginners moving between tiers.

Use Cases
  • Learn SOC analysis through guided project builds
  • Build a DDoS mitigation tool with step-by-step instructions
  • Reference certification roadmaps for security engineering roles

Source: CarterPerez-dev/Cybersecurity-Projects — based on the project README.

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

Release expands blacklist filtering and updates eight language packs for global users

GyulyVGC/sniffnet · Rust · ▲ 17 in 1d Est. 2022

GyulyVGC’s Sniffnet v1.5.

1 introduces real-time latency tracking for network connections, addressing long-standing request #845. The update also enables CIDR notation in imported IP blacklists, improving threat filtering precision. Translation updates now cover Ukrainian, German, French, Spanish, Indonesian, Swedish, Greek, Persian, and Russian, alongside new Hungarian and Sinhala builds. Built with Rust and Iced, the tool remains cross-platform, offering PCAP export, protocol identification, and per-app bandwidth tracking.
The catch: Open issues exceed sixty, suggesting ongoing stability work despite active development.

Previously in The Times “covered” — Aug 12

Use Cases
  • Network admins diagnose latency spikes per connection
  • Security teams filter threats using CIDR-formatted blacklists
  • Developers trace bandwidth usage by individual applications

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

Quick Hits

SWE-agent SWE-agent automatically fixes GitHub issues using your preferred language model, enabling rapid code repair and adaptation for cybersecurity or coding challenges. 20.1k
maigret Maigret aggregates detailed dossiers on individuals by username across 3,000+ websites, empowering builders with deep OSINT reconnaissance capabilities. 36.9k
sops SOPS simplifies secrets management by encrypting sensitive data in files using GPG, AWS KMS, or similar, ensuring secure configuration handling in any workflow. 22.8k
hacktricks HackTricks wiki curates practical, battle-tested techniques from CTFs, real-world exploits, and research — a builder’s reference for offensive security know-how. 12.1k
trufflehog TruffleHog detects, verifies, and analyzes leaked credentials in code repositories, helping builders prevent breaches by identifying exposed secrets early. 27.5k
Ciphey ⚡ Automatically decrypt encryptions without knowing the key or cipher, decode encodings, and crack hashes ⚡ 21.6k

FRP v0.71.0 fixes UDP proxy panic and domain validation bypass 🔗

Wire protocol v2 now uses compact binary codec for UDP payloads when peers negotiate capability

fatedier/frp · Go · ▲ 22 in 1d Est. 2015 · Latest: v0.71.0

The latest frp release patches a server panic triggered by negative pool_count values and closes a case-insensitive validation loophole that allowed mixed-case customDomains to bypass subHost restrictions. Wire protocol v2 now negotiates a dedicated binary codec for UDP packet payloads, falling back to JSON only when peers lack support.

Developers use frp to expose local services behind NATs via TCP, UDP, HTTP, or HTTPS, with P2P mode and TLS encryption. The catch: Complex configurations for P2P connect mode and protocol multiplexing remain challenging for new adopters despite recent stability fixes.

Use Cases
  • Developers expose internal web services via custom domains
  • Teams access LAN computers over SSH through frp tunnels
  • Services enable HTTPS for local HTTP servers using frp tunneling

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

Quick Hits

duckdb DuckDB delivers high-performance SQL analytics directly in your application, enabling fast, embedded data processing without external servers. 40.4k
PowerToys PowerToys enhances Windows workflow with customizable utilities like FancyZones and PowerRename, boosting efficiency for power users and developers. 137.8k
codex Codex acts as a lightweight, terminal-based coding agent that understands natural language to generate, edit, and explain code in real time. 106.6k
alacritty Alacritty provides a blazing-fast, GPU-accelerated terminal experience with minimal dependencies and consistent performance across Linux, macOS, and Windows. 65.4k
tmux tmux enables persistent, multiplexed terminal sessions — split panes, detach/reattach, and remote workflow control — all within a single, lightweight interface. 48.7k
rtk CLI proxy that reduces LLM token consumption by 60-90% on common dev commands. Single Rust binary, zero dependencies 76.4k
protobuf Protocol Buffers - Google's data interchange format 71.7k
browser Lightpanda: the headless browser designed for AI and automation 34k

HackRF Release Fixes Mixer Lock Failures, Adds Pro Flash Support 🔗

v2026.01.3 resolves frequency instability and expands storage for advanced SDR experimentation

greatscottgadgets/hackrf · C · 8k stars Est. 2012 · Latest: v2026.01.3

The latest HackRF release addresses a persistent mixer frequency lock issue that disrupted signal stability during long captures. Engineers can now rely on consistent tuning across bands, critical for radar and satellite monitoring.

The update also enables access to larger SPI flash on HackRF Pro models, allowing users to store bigger firmware images or custom gateboards directly on hardware. This expands the platform’s utility for standalone, field-deployable SDR nodes without constant host tethering.
Built in C, HackRF remains a low-cost entry point to wideband SDR from 1 MHz to 6 GHz. Its open hardware designs and thorough documentation continue to support academic research, DIY spectrum mapping, and prototyping of wireless protocols. The project maintains active issue triage, with technical support queries addressed within two weeks by maintainers.
The catch: Despite recent fixes, 83 open issues persist, and documentation builds still require manual Sphinx and LaTeX setup — a friction point for newcomers seeking zero-config onboarding.

Previously in The Times “covered” — Aug 14

Use Cases
  • University lab testing 5G NR signal characteristics
  • Field engineer capturing aviation ADS-B signals
  • Hobbyist decoding NOAA weather satellite downlinks

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

More Stories

Bruce firmware adds WPA3 detection and handshake capture fixes 🔗

Latest release improves WiFi security testing and stabilizes SSH/Telnet on non-keyboard devices

BruceDevices/firmware · C++ · 6.5k stars Est. 2024

BruceDevices/firmware v1.16.

1 adds detection for WPA3-PSK and mixed WPA2/WPA3 networks, ensuring handshakes are captured correctly during attacks. The update fixes SSH and Telnet crashes when accessed without an active connection and resolves deauthentication issues affecting station mode. Screen refresh rates for the Channel Analyzer and Rogue Detect tools were reduced to improve readability. These changes enhance reliability in offensive WiFi operations across ESP32-S3 and ESP32-C5 hardware.
The catch: With 244 open issues and a narrow focus on offensive security, the firmware may lack stability for general-purpose or long-term embedded deployments.

Previously in The Times “covered” — Aug 11

Use Cases
  • Red teamers testing WiFi network defenses
  • Security researchers capturing WPA3 handshakes
  • Pentesters conducting deauthentication and beacon spam attacks

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

Intel's ROHD Framework Simplifies Hardware Design in Dart for Builders 🔗

Latest update adds coverage badge and Node v24 trial run for smoother verification workflows

intel/rohd · Dart · 489 stars Est. 2021

The Rapid Open Hardware Development (ROHD) framework lets builders describe and verify hardware using Dart, replacing verbose HDLs with familiar syntax. Recent updates include a coverage status badge in the README and a trial run for Node v24, improving CI integration and toolchain compatibility.

ROHD’s built-in simulator, SV generation, and IP reuse via import streamline design and testbench creation without complex EDA toolchains. Developers leverage pub.dev packages and event-based simulation with VCD output for debugging. The catch: Despite active commits, 137 open issues suggest ongoing stability challenges in advanced features like combinational loop handling and SV generation edge cases.

Use Cases
  • FPGA designers writing testbenches in Dart instead of SystemVerilog
  • Hardware teams reusing IP components via simple Dart imports
  • Educators teaching hardware concepts using a modern, accessible language

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

Fanpico Adds Post-Quantum Crypto to Raspberry Pi Pico Fan Controller 🔗

v1.7.9 enables SSH server with quantum-resistant security for both Pico and Pico W boards

tjko/fanpico · C · 234 stars Est. 2022

Fanpico’s latest release integrates post-quantum cryptography into its SSH server, strengthening remote access for the open-source PWM fan controller built around the RP2040 MCU. The update unifies firmware across standard Pico and Pico W variants, adds WiFi diagnostics (channel, RSSI, BSSID), and ships with a revised PCB v1.

1 for the FanPico-0200 model. It retains core features: 8-fan control, dual temperature probe support, MQTT/HTTPS logging, and SCPI-like configuration via onboard flash — all OS-independent.
The catch: Post-quantum crypto increases firmware size and may strain limited Pico W resources under heavy concurrent MQTT/TLS workloads.

Use Cases
  • DIY builders tuning PC fan curves for silent operation
  • Engineers monitoring 3D printer enclosure thermal dynamics
  • IT staff managing network rack airflow via MQTT telemetry

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

Quick Hits

Simple-Hydroponics-Arduino Hydruino provides a professional-grade, fully configurable Arduino-based automation controller for hydroponic systems, enabling precise control without internet or cloud dependencies. 93
librealsense librealsense is a cross-platform C++ SDK that enables real-time depth, motion tracking, and facial recognition from Intel RealSense cameras for robotics and AI applications. 9k
openflight openflight is a Python library for parsing, manipulating, and generating OpenFlight (.flt) 3D scene files used in simulation, visualization, and geospatial modeling workflows. 889
FlightTracker FlightTracker turns a Raspberry Pi into a real-time aircraft monitor using a dot matrix display to show live flight data, satellite positions, weather, and time from ADS-B or FlightRadar24 sources. 190
TuyaOpen TuyaOpen delivers a next-gen AI+IoT framework for rapid hardware integration across ESP32, T-series chips, and more, enabling fast development of intelligent edge devices with on-device AI capabilities. 1.8k

Escoria Demo Game Tests Godot 4.7 Compatibility Amid API Shifts 🔗

Developers must manually adjust core scripts for Godot 4.7 due to breaking changes in scene warning handling

godot-escoria/escoria-demo-game · GDScript · 845 stars Est. 2016 · Latest: v4.0.0-alpha.319

The Escoria demo game serves as a live testbed for the point-and-click adventure framework’s compatibility with Godot Engine 4.6.

x and 4.7.x. Recent commits show active maintenance, with the last push just days ago and the latest release tagged v4.0.0-alpha.319. The project highlights a critical version-specific tweak: in res://addons/escoria-core/game/core-scripts/esc_item.gd, the _get_configuration_warnings() method must return "\n".join(_scene_warnings) for Godot 4.6 but simply _scene_warnings for 4.7. Escoria auto-detects this need on launch, but warns that custom builds require manual script updates to avoid errors.
Built with GDScript, the demo integrates CC0-licensed assets from Kenney, Marco Giorgini, and Julian Murgia, alongside audio from Eric Matyas and freesound.org. It showcases Escoria’s core features — dialogue trees, item interaction, and scene navigation — while functioning as a regression test for ongoing framework development.
The catch: The framework’s reliance on Godot 4.x-specific API workarounds creates fragmentation; developers targeting both engine versions must maintain divergent script paths, increasing complexity for teams standardizing on a single Godot release.

Use Cases
  • Adventure devs testing Godot 4.6/4.7 compatibility
  • Contributors validating Escoria core script changes
  • Teams learning point-and-click mechanics in Godot

Source: godot-escoria/escoria-demo-game — based on the README and release notes.

More Stories

MustardOS/internal defines core system infrastructure for embedded Linux devices 🔗

Forked 102 times despite two open issues and no recent feature commits

MustardOS/internal · GLSL · 108 stars Est. 2024

The repository contains essential components for MustardOS: custom binaries in bin/, SFTPGo browse symlinks, global and device-specific configurations, a POSIX-compliant script/ backbone, and filesystem partitions for init/ and kiosk/ modes. It also houses shared resources, update markers, and frontend integration hooks.

Last updated over two years ago, the project shows no recent code changes despite sustained community interest via forks.
The catch: No commits in 0 days suggests stagnation, raising questions about active maintenance and long-term viability for production use.

Use Cases
  • Device builders configuring kiosk-mode Linux systems
  • Developers creating SFTP-accessible embedded appliances
  • Engineers scripting POSIX-compliant firmware update workflows

Source: MustardOS/internal — based on the project README.

GodSVG Edits SVG Code Directly Without Metadata for Clean Output 🔗

Late-alpha tool runs on desktop and web via Godot Engine in GDScript

MewPurPur/GodSVG · GDScript · ▲ 4 in 1d Est. 2023

GodSVG lets users manipulate SVG elements through a graphical interface while instantly generating human-readable, metadata-free SVG code. Available for Windows, macOS, Linux, and web, it produces optimized, file-size-efficient output by editing the SVG structure directly.

Built with Godot Engine using GDScript, the project remains in late alpha after over three years of solo development. Recent activity shows ongoing maintenance, with the last commit just days ago and 51 open issues indicating active refinement.
The catch: As a late-alpha, one-developer project, long-term stability and feature completeness remain uncertain for production reliance.

Use Cases
  • Designers editing SVGs without bloated metadata or hidden layers
  • Developers generating lightweight, hand-tuneable SVG code for web use
  • Teams needing cross-platform vector editing that preserves code clarity

Source: MewPurPur/GodSVG — based on the project README.

Quick Hits

beehave Bitbrain/beehave lets Godot developers build complex AI behaviors using intuitive behavior trees for smarter, more responsive game characters. 3.2k
o3de O3DE empowers creators to build AAA-grade 3D experiences — games, films, and simulations — free and open-source across platforms. 9.6k
bevy Bevy offers a clean, data-driven Rust engine that simplifies game development with modularity and high performance. 47.7k
raylib Raylib provides a lightweight, beginner-friendly C library for rapid 2D/3D game prototyping with minimal setup. 34.4k
NRD NVIDIA NRD delivers real-time denoising for ray-traced graphics, boosting visual quality without sacrificing frame rates. 827
ebiten Ebiten enables Go developers to create smooth 2D games effortlessly with a minimal API and zero external dependencies. 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