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 $3.37/M Qwen3.8 2.4T A95B $6/M Kimi K3 $15/M
Full Markets →

Tiny Zig Proxy Masks Telegram Traffic as HTTPS to Evade Censors 🔗

v1.13.0 adds Hysteria2 egress support and fixes live-server bugs undetected by CI

sleep3r/mtproto.zig · Zig · ▲ 95 in 1d 4mo old · Latest: v1.13.0

The sleep3r/mtproto.zig project delivers a dependency-free MTProto proxy written in Zig that disguises Telegram traffic as ordinary TLS 1.

3 HTTPS, allowing users in censored regions to maintain access without installing additional software. At just 177 KB and under 1 MB RAM, it runs on minimal VPS instances and requires only a single command to deploy. The proxy generates a shareable link that users tap once in Telegram to establish a connection—no client-side configuration, no sign-ups, and no logging of secrets or metadata.

Recent activity centers on the v1.13.0 release, which integrates Hysteria2 as an egress option (hysteria2:// or hy2:// schemes), enabling UDP-based tunneling for improved performance where supported. More critically, the update patches five bugs uncovered during live-server testing that CI pipelines failed to catch, including a critical flaw where running setup egress without prior setup tunnel left the proxy accepting connections but relaying no data—while falsely reporting success. The release also strengthens error handling for invalid URI parameters like unsupported obfuscation or port hopping, ensuring clear feedback instead of silent degradation.

Built for simplicity and stealth, the project avoids the overhead of traditional VPNs by focusing solely on Telegram traffic and mimicking benign HTTPS patterns to bypass deep packet inspection. Its MIT-licensed, zero-dependency design appeals to developers seeking lightweight, self-hosted anti-censorship tools.

The catch: While effective for Telegram, the proxy’s narrow scope limits reuse for other protocols, and its reliance on TCP desync and TLS mimicry may face evolving DPI countermeasures not yet tested at scale.

Why this leads today It offers a practical, self-hosted Telegram proxy in Zig that evades censorship by blending into HTTPS traffic, enabling open communication in restrictive environments without reliance on third parties.

Use Cases
  • Activists accessing Telegram in regions with state-level blocks
  • Families sharing a simple link to restore cross-border messaging
  • Developers deploying lightweight anti-censorship tools on cheap VPS instances

Source: sleep3r/mtproto.zig — based on the README and release notes.

More on the Front Page

How-To-Fish Trainer Lets Players Skip Grind in Steam Co-op Fishing Game 🔗

Open-source C++ tool injects memory to unlock gear, disable losses, and tweak physics for 1–4 player sessions

MediatorAxle/How-To-Fish-Trainer · C++ · 250 stars 0d old

MediatorAxle’s How-To-Fish Trainer is a memory-injection tool for the Steam game How to Fish, offering 45+ cheat options across five sections: Angler, Arsenal, Island, System, and Field of view. Built in C++, it launches an overlay via the Insert key, allowing players to enable infinite money and ammo, prevent fish from escaping, stop birds from stealing catches, and unlock every weapon and island instantly.

Host-only mode is enabled by default, ensuring guest players in 1–4 player co-op cannot alter shared world settings like creature AI or boss health—preserving the intended chaotic fun of cooperative failure. The trainer skips the punishing early-game grind by removing the red-bar failure state while keeping the reel-in fight intact, and guarantees rare fish variants from the next cast. Known issues include island options failing to bind from the boat (requiring players to land first), physics forces above 10x launching quest items out of bounds, and Insert key conflicts with Steam, Discord, or RTSS overlays. Windows Defender may flag the executable due to its memory-reading behavior, though the source is available for self-compilation.

The catch: As a single-day-old release built against the game’s launch version, long-term compatibility with future How to Fish updates is untested and may break without rebasing.

Use Cases
  • Players bypass early-game resource scarcity in solo play
  • Co-op groups enable shared progression without grinding
  • Streamers reduce frustration during long fishing sessions

Source: MediatorAxle/How-To-Fish-Trainer — based on the README and release notes.

Bryl Lim releases 302-exercise workout library with typed npm package 🔗

Provides framework-neutral assets for React Native, Expo, and static sites via MIT-licensed code

bryllim/workout-guide · Astro · 253 stars 1d old

Bryl Lim’s workout-guide offers 302 exercises, each with three transparent 512×512 SVG frames, sourced from Everkinetic and expanded under CC BY-SA 4.0.

The framework-neutral npm package delivers typed ESM and CommonJS APIs with manifest search, lookup, and CDN URL helpers. Direct asset imports and literal require() examples support React Native and Expo integration. A searchable Astro-powered gallery hosts detail pages and an integration guide. All code and documentation are MIT-licensed; visual assets retain CC BY-SA 4.0 with full attribution in LICENSES.md. The catch: The project relies on a single maintainer and lacks automated testing for cross-platform asset rendering in complex app builds.

Use Cases
  • Developers add animated exercise demos to React Native fitness apps
  • Designers build static workout galleries with search and detail views
  • Teams integrate licensed pose illustrations into Expo-based wellness tools

Source: bryllim/workout-guide — based on the README and release notes.

Vibe-Trading Adds Options Lab and Factor Research Tools 🔗

New UI features enhance strategy analysis without altering core trading engine

HKUDS/Vibe-Trading · Python · 31.7k stars 4mo old

The HKUDS/Vibe-Trading project rolled out an Options Lab web UI with payoff diagrams, IV scenario matrices, and live US options chain data, all read-only and backed by existing MCP tools. A Factor Research tab now displays IC statistics, equity curves, and correlation heatmaps from run artifacts via a new /runs/{run_id}/factor endpoint.

These additions improve strategy visualization for quantitative traders using Python-based agents. The catch: 24 open issues suggest ongoing stability challenges in MCP integrations and OAuth flows with brokers like IBKR.

Use Cases
  • Quant analysts backtesting options strategies
  • Traders visualizing factor model performance
  • Developers extending multi-agent trading systems

Source: HKUDS/Vibe-Trading — based on the README and release notes.

Open-source design canvas Doop lets humans and AI agents co-create in real time 🔗

Built-in MCP server enables live collaboration with Claude Code or any MCP client

kgoedecke/doop · TypeScript · 336 stars 2d old

Doop is a multiplayer design canvas where humans edit in the browser and AI agents stream designs via a built-in MCP server. Every frame renders real HTML in sandboxed iframes, with live cursors, presence, and activity feeds visible to all.

The project includes a resident Doop Agent that designs autonomously when @mentioned or queued, and supports style distillation from pinned exemplar frames. Self-hosting requires one command via Docker Compose or npm run dev, using embedded Postgres for zero-config persistence. Agents inherit human access controls, keeping canvases private by default unless link sharing is enabled.
The catch: As a v0.1.0 release with no open issues but limited real-world usage, its long-term reliability under heavy multi-agent load remains unproven.

Use Cases
  • Design teams co-create UI layouts with Claude Code suggesting live edits
  • Solo developers prototype frontends while AI agents generate accessible components
  • Educators run live design workshops where students and AI build together in shared canvases

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

backpass turns agent transcripts into gradient descent for AGENTS.md 🔗

Proposes evidence-backed memory edits from local session logs without external APIs

kunchenguid/backpass · JavaScript · 241 stars 3d old

backpass treats your AGENTS.md as model weights and agent sessions as forward passes, using session transcripts as loss signals to propose memory updates.

It reads local transcript stores from seven agent harnesses—like CLI and ACP sessions—directly from disk, never uploading data. Each proposed edit includes verbatim quotes from at least two independent sessions, with a maximum of five edits per run. You review and accept changes via backpass apply, keeping humans in the loop. The tool requires Node >=22.5 and acpx, with no built-in API keys.
The catch: It only works with Hermes v26 CLI and ACP sessions, excluding gateway, cron, and WhatsApp logs due to shared working directories.

Use Cases
  • AI engineers refining agent memory from local CLI sessions
  • Developers correcting agent behavior using ACP session transcripts
  • Teams auditing agent decisions with evidence-backed memory updates

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

macOS Terminal cmux Adds Remote Workspace Sync for AI Agent Teams 🔗

Latest release fixes notification handling and SSH startup for Claude Code integration

manaflow-ai/cmux · Swift · 26.4k stars 6mo old

cmux v0.64.

22 resolves a bash noclobber error and preserves Claude’s secure storage directory during agent restores. It now clears Dock notifications when focused and fixes SSH startup syntax in retry loops. The terminal supports vertical/horizontal splits, in-app browsing, and remote workspaces via cmux ssh, with browser panes routing through remote networks. AI coding agents trigger blue rings on panes and tab lights for attention. Native macOS app built in Swift, GPU-accelerated via libghostty.
The catch: Open issues exceed 4,400, indicating significant unresolved bugs despite active maintenance.

Use Cases
  • Developers managing multiple AI coding agents locally
  • Teams running Claude Code in teammate mode with shared context
  • Engineers accessing remote machines via SSH with synchronized browser state

Source: manaflow-ai/cmux — based on the README and release notes.

Tencent's WeKnora Turns Raw Docs Into Self-Maintaining Knowledge Wiki 🔗

Latest release adds revision history, diff, and rollback for agent-edited wiki pages

Tencent/WeKnora · Go · 20.6k stars Est. 2025

WeKnora, Tencent's open-source LLM knowledge framework in Go, converts documents into queryable RAG, autonomous ReAct agents, and a self-updating Wiki mode. The v0.

7.2 release introduces page revision history with diff, revert, and manual editing—letting teams track and roll back AI-generated knowledge changes. It ingests from Feishu, Notion, Yuque, and RSS, handles 10+ formats, and integrates with 20+ LLMs including OpenAI and Qwen. Multi-workspace RBAC and Langfuse observability support enterprise deployment.
The catch: With 546 open issues and rapid commits, stability for high-scale production use remains unproven despite active development.

Use Cases
  • Enterprise teams sync internal wikis for AI-powered Q&A
  • Developers build autonomous agents that reason over document collections
  • Knowledge curators maintain version-controlled, interlinked markdown knowledge bases

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

AI Agent Toolkits Reshape Open Source Around LLM-Driven Automation 🔗

Projects now package skills, agents, and verifiers as reusable, interoperable components for autonomous workflows

Trendllm-tools
LB623/no-negative-echokgoedecke/dooprome-os/romebam-bam-2/solo-skillsmanaflow-ai/cmuxbojieli/ai-agent-book1Panel-dev/MaxKBHKUDS/Vibe-TradingTencent/WeKnoracathrynlavery/diagram-designvirgiliojr94/book-to-skillfreestylefly/awesome-gpt-image-2Wei-Shaw/sub2apizhaoxuya520/reverse-skillTiger3807861189/J-Space-Cognition-Suite-V3.7citrolabs/ego-liteAlishahryar1/free-claude-codellm-as-a-verifier/llm-as-a-verifierlightningpixel/modlyinternet-court/internet-court-skillFailproofAI/failproofaicoreyhaines31/marketingskillsMadsLorentzen/ai-job-searchjundot/omlxlidge-jun/opencodexshy3130/tickflow-stock-panelMemTensor/memmy-agentbradautomates/claude-videoWenyuChiou/awesome-agentic-ai-zhsunchaokun/PPT-Design-Skillnuyoah-ai-works/nuyoah-xiezhen-promptZhuLinsen/daily_stock_analysisdob323/session-kitrasbt/LLMs-from-scratchray-project/rayGoogleCloudPlatform/generative-aiAwarexone/Agentic-Bug-Hunterpbatard/rufus

A clear pattern is emerging in open source where LLMs are no longer just models but foundational layers for composable, agentic tooling. Repos like llm-as-a-verifier/llm-as-a-verifier provide frameworks to critique agent outputs without retraining, while MemTensor/memmy-agent introduces a shared memory hub so multiple AI agents maintain synchronized context—enabling true collaboration.

Skill marketplaces are forming: bam-bam-2/solo-skills offers 26 ready-to-run AI agent skills for solopreneurs, and virgiliojr94/book-to-skill converts technical books into executable Claude Code abilities. Platforms such as 1Panel-dev/MaxKB and Tencent/WeKnora go further, turning documents into self-maintaining wikis or reasoning agents via RAG. Even niche domains are agentified: HKUDS/Vibe-Trading deploys LLMs as personal trading agents, and zhaoxuya520/reverse-skill routes pentest tools via AI-driven skill selection. Interoperability is key—lidge-jun/opencodex lets users plug any LLM into Codex or Claude Code, and citrolabs/ego-lite shares browser state with agents for seamless automation. These projects treat LLMs not as endpoints but as runtime engines for pluggable, verifiable, memory-aware agents—shifting open source from static libraries to dynamic, goal-directed systems.
The catch: Much of this remains experimental, with fragmented standards for skill interfaces, agent communication, and safety guarantees; many tools rely on fragile prompt engineering rather than robust architectures, and real-world agent reliability in complex, long-horizon tasks is still unproven outside controlled demos.

Use Cases
  • Solo founders automate admin, coding, and outreach with pre-built agent skills
  • Teams deploy shared memory hubs to synchronize context across AI agents in workflows
  • Developers convert technical books into instant, executable AI agent references for on-the-job learning

AI Agents Evolve from Assistants to Autonomous Open Source Systems 🔗

New frameworks unify memory, skills, and reasoning for self-improving agent workflows across domains

Trendai-agents
LB623/no-negative-echoApodexAI/FrontierAgentrome-os/romeinkboard/system-atlasbam-bam-2/solo-skillsGamePhanes/GamePhanesbojieli/ai-agent-book1Panel-dev/MaxKBHKUDS/Vibe-Tradingclacky-ai/openclackyTencent/WeKnoravolcengine/OpenVikingakitaonrails/ai-memoryayghri/i-have-adhdchaitanyagiri/munder-difflintt-a1i/archifyifixai-ai/iFixAiblader/humanizerHiThink-Tech/Financial-APIcan1357/oh-my-picalesthio/OpenMontagellm-as-a-verifier/llm-as-a-verifierinternet-court/internet-court-skillvectorize-io/hindsighttitanwings/distillyMemTensor/memmy-agentPrimeIntellect-ai/prime-agentWenyuChiou/awesome-agentic-ai-zhhenryqin1997/statemvercel-labs/agent-browsernuyoah-ai-works/nuyoah-xiezhen-promptchuspeeism/dashi-ppt-skillSnailclimb/JavaGuideGoogleCloudPlatform/generative-aiAwarexone/Agentic-Bug-Huntercurl/curl

Open source is shifting from isolated AI helpers to integrated agent ecosystems. Projects like FrontierAgent offer native TUI and ReAct modes with zero Docker dependency, while rome-os/rome positions itself as an agentic OS.

Memory unification is advancing via volcengine/OpenViking, which merges RAG, skills, and context into a self-evolving database, and MemTensor/memmy-agent provides a shared memory hub for multiple agents. Skill marketplaces are emerging: bam-bam-2/solo-skills ships 26 ready-to-use agent skills for solopreneurs, and titanwings/distilly extracts reusable skills from human-agent interactions. Specialized agents are gaining traction—HKUDS/Vibe-Trading acts as a personal trading agent, ifixai-ai/iFixAi enables rapid agent auditing in under 120 seconds, and calesthio/OpenMontage launches as the first agentic video production system with 100+ tools. Verification is also being baked in: llm-as-a-verifier/llm-as-a-verifier delivers SOTA feedback without retraining. These repos signal a move toward composable, self-improving agents that operate across coding, finance, media, and commerce.
The catch: Despite rapid innovation, the ecosystem remains fragmented—skills often lack cross-agent compatibility, memory systems diverge in implementation, and few agents demonstrate reliable long-term autonomy without human oversight, raising questions about real-world robustness beyond demos.

Use Cases
  • Developers automate code reviews using agent skills
  • Traders deploy personal agents for market analysis
  • Teams generate architecture diagrams from natural language prompts

Open Source Embraces Modular, Agent-Ready Web Frameworks 🔗

Projects unify frontend/backend with plugin architectures for AI-driven, real-time, and composable web experiences

Trendweb-frameworks
ApodexAI/FrontierAgentChenShuo2004/cs-boardbryllim/workout-guidegetsentry/sentry-react-nativemedusajs/medusaproxifly/free-proxy-listTypeCellOS/BlockNotewundergraph/cosmofrappe/frappegronxb/hot-updaterweb-infra-dev/rspresszhu1090093659/dsh-web-uiHiThink-Tech/Financial-APIdnshe/DNSHE-FreeDomainsGoogleCloudPlatform/generative-aiOGRECave/ogreprojectdiscovery/httpxTheSpeedX/PROXY-Listcurl/curljstrieb/github-statsdmtrKovalenko/odiffkarlseguin/http.zig

A clear pattern emerges in open source web frameworks: a shift toward modular, extensible architectures designed for AI agents, real-time updates, and cross-platform deployment. Rather than monolithic stacks, new projects prioritize composability and low-friction integration.

Frappe exemplifies this as a low-code framework blending Python and JavaScript to accelerate real-world app development with built-in workflow automation. MedusaJS takes modularity further, offering a headless commerce engine where developers plug in custom services via APIs, enabling AI agents to orchestrate shopping experiences. Rspress leverages Rust-based tooling for lightning-fast static site generation, integrating seamlessly with modern frontend toolchains while supporting plugin extensions. Meanwhile, BlockNote provides a block-based, Prosemirror-powered rich text editor that’s framework-agnostic yet deeply extensible — ideal for embedding in agent-driven content tools. The rise of agent-centric design is evident in FrontierAgent, which ships a native TUI and ReAct modes for CLI-based AI agent teams, hinting at frameworks evolving to support autonomous workflows. Even infrastructure reflects this shift: Cosmo enables federated GraphQL at scale as an open alternative to proprietary studios, while Hot-Updater offers self-hosted OTA updates for React Native, reducing reliance on closed services. These projects signal a move toward web frameworks that are not just tools for building sites, but foundational layers for AI-augmented, dynamically composable systems.
The catch: Much of this innovation remains fragmented across language ecosystems (TypeScript, Python, Rust), with limited cross-project standards for agent communication or plugin interoperability, risking siloed adoption despite promising individual advances.

Use Cases
  • Developers build AI-powered storefronts using MedusaJS plugins
  • Teams deploy self-hosted React Native updates via Hot-Updater
  • Content creators extend BlockNote for agent-assisted writing workflows

Deep Cuts

Chinese Codex Skill Generates PR Titles, Comments, Commits from Results 🔗

Reduces rejected solutions in delivery by aligning output with final outcomes

LB623/no-negative-echo · Python · ▲ 48 in 1d

LB623/no-negative-echo is a Python-based Codex skill that automatically generates titles, comments, commit messages, and PR descriptions based on the final result of code generation. By anchoring output to the actual outcome rather than intermediate steps, it minimizes mismatches that lead to rejected solutions during delivery.

Built for agent-skills workflows, it streamlines prompt engineering for Chinese-speaking developers using Codex. The skill ensures consistency between generated artifacts and what actually ships, cutting down rework. It’s lightweight, focused, and designed to plug into existing agent pipelines with minimal setup.
The catch: Early-stage and niche, it lacks broad documentation and real-world validation beyond initial adopters.

Use Cases
  • Chinese dev teams reducing PR rejections in agent-driven workflows
  • Automating commit messages that reflect actual code outcomes
  • Aligning Codex-generated comments with final functionality for clarity

Source: LB623/no-negative-echo — based on the project README.

Quick Hits

rome Rome provides an agentic operating system built in TypeScript, enabling autonomous AI agents to interact with system resources and workflows natively. 302
selfdb SelfDB is a Python-based personal knowledge database that lets users store, query, and reflect on private data using natural language and local LLMs. 263
FrontierAgent FrontierAgent offers a native command-line TUI for building and running AI agents with ReAct reasoning and team collaboration modes, requiring no Docker or preinstall on macOS and Linux. 355
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

TensorFlow 2.21 drops Python 3.9 support and decouples TensorBoard 🔗

Release adds int2/int4 quantization and JPEG XL decoding for lightweight ML deployment

tensorflow/tensorflow · C++ · ▲ 234 in 1d Est. 2015 · Latest: v2.21.0

TensorFlow 2.21.

0 removes support for Python 3.9 and separates TensorBoard from the core framework, streamlining dependencies for production use. The update enhances tf.lite with int2, int4, uint4, and int16x8 types for operators like SQRT, EQUAL, and slice, enabling tighter model quantization for edge devices. tf.image now decodes JPEG XL images directly, reducing preprocessing overhead for modern web formats. tf.data introduces NoneTensorSpec to explicitly handle undefined tensor shapes in pipelines, improving debuggability. These changes target developers optimizing ML workloads where binary size, latency, and dependency control matter. The project maintains stable C++ and Python APIs while shifting non-guaranteed language bindings to community maintenance. Nightly builds remain available via tf-nightly packages for early testing. Despite active development, the repository shows 2,937 open issues and a last commit just hours ago, indicating sustained but strained maintenance pressure.

The catch: Aggressive dependency removal and version drops may force costly upgrades in enterprise environments locked to older Python versions or TensorBoard-integrated tooling.

Previously in The Times “covered” — Aug 22

Use Cases
  • Quantizing models for microcontrollers using int2/int4 precision
  • Decoding JPEG XL in web-native ML pipelines without external libraries
  • Debugging data pipelines with explicit None tensor shape handling

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

More Stories

Hands-on LLM book code stays current with PyTorch updates 🔗

Last commit one day ago shows active maintenance despite 2023 launch

rasbt/LLMs-from-scratch · Jupyter Notebook · 103.7k stars Est. 2023

rasbt/LLMs-from-scratch provides Jupyter notebooks implementing a ChatGPT-like LLM from scratch in PyTorch, matching the Manning book Build a Large Language Model (From Scratch). It covers pretraining, finetuning, and loading larger model weights for educational use.

The project supports instruction tuning and tokenizer customization, mirroring foundational model approaches. Despite its ~3.1-year age, recent activity indicates ongoing relevance for builders seeking deep LLM internals understanding. The catch: The narrow focus on PyTorch and educational scale may limit direct production adoption compared to industrial frameworks like Hugging Face Transformers.

Use Cases
  • ML engineers learning LLM architecture fundamentals
  • Students experimenting with custom pretraining pipelines
  • Researchers studying attention mechanisms step-by-step

Source: rasbt/LLMs-from-scratch — based on the project README.

Spec Kit 1.0.1 refines spec-driven workflows for AI coding agents 🔗

Release adds Azure Cosmos DB extension and fixes workflow condition handling

github/spec-kit · Python · 131.2k stars Est. 2025

GitHub’s spec-kit project released v1.0.

1, focusing on documentation improvements and dependency updates rather than core功能. The update includes an Azure Cosmos DB extension for the community catalog and resolves issues where workflow switch steps incorrectly offered inverted condition corrections. Teams use Spec Kit to define executable specifications that AI agents like GitHub Copilot can directly implement, reducing rework. The tool remains Python-based and installable via uv tool install specify-cli. Despite 11,786 forks, 347 open issues indicate ongoing refinement needs.
The catch: The project’s value hinges on AI agent reliability, which varies across models and introduces unpredictability in spec-to-code translation.

Previously in The Times “covered” — Aug 18

Use Cases
  • Developers define features before coding with AI agents
  • Teams adopt spec-driven processes using preset bundles
  • Organizations extend workflows with community-built extensions

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

Quick Hits

ColossalAI ColossalAI optimizes large AI model training by reducing costs and accelerating performance through efficient parallelization techniques. 41.4k
AutoGPT AutoGPT enables autonomous AI agents that break down complex goals into actionable steps, letting users focus on outcomes rather than implementation. 186.9k
hermes-agent Hermes-Agent adapts and improves its behavior over time through continuous learning from user interactions and feedback loops. 236k
generative-ai Generative AI on Google Cloud provides ready-to-run notebooks and sample code for building and deploying Gemini-powered applications. 17.6k
ray Ray scales AI and ML workloads across clusters with a unified runtime and libraries for distributed training, tuning, and serving. 43.6k
yolov5 Ultralytics YOLOv5 in PyTorch for object detection, instance segmentation, classification, training, and export. 57.9k

Rerun 0.36.2 fixes external link bugs in multimodal data viewer 🔗

Patch improves usability for engineers debugging sensor streams across robotics pipelines

rerun-io/rerun · Rust · 11.3k stars Est. 2022 · Latest: 0.36.2

The latest Rerun release resolves a bug that prevented opening external links from within its embedded viewer, a regression affecting users who rely on cross-referencing documentation or issue trackers during data inspection. Built in Rust with column-chunk storage for multi-rate physical data, Rerun ingests synchronized multimodal streams—images, point clouds, joint states, and video—from sources like MCAP, rrd, and LeRobot logs.

Its SDKs in Python, Rust, and C++ enable logging and querying via dataframes or SQL, while the browser-based viewer renders everything in real time, allowing engineers to scrub episodes, compare sensors side-by-side, and observe CV pipelines like SLAM or hand tracking as they run. Data streams directly into training workflows without export jobs or stale copies, reducing friction in robotics ML iteration. The project maintains active development, with the last commit just zero days ago and 1,276 open issues indicating ongoing community engagement.
The catch: While the viewer runs in the browser, the C++ and Rust SDKs require a separate binary install for full functionality, adding setup complexity compared to the all-in-one Python experience.

Previously in The Times “covered” — Aug 20

Use Cases
  • Robotics engineers syncing lidar and camera data for debugging
  • ML teams streaming multimodal sequences directly into training loops
  • Researchers querying synchronized sensor data via SQL or dataframes

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

More Stories

Kornia v0.8.3 adds XFeat, ALIKED features and half-precision support 🔗

Patch release improves geometry ops performance and fixes MPS, ONNX export issues

kornia/kornia · Python · 11.3k stars Est. 2018

Kornia v0.8.

3 delivers ~175 commits since v0.8.2, adding XFeat and ALIKED feature detectors/descriptors, batched VisualPrompter inference, and float16/bfloat16 support. Core geometry ops like depth_to_normals now run ~3x faster, with transform points and denormalize_points_with_intrinsics seeing ~2x speedups. The release includes MPS device fixes, ONNX export corrections, expanded docstrings, and DeprecationWarnings for kornia.utils utilities slated for removal. The catch: Despite active development, 99 open issues persist, raising questions about triage velocity and long-term maintenance sustainability for a project of this scale.

Use Cases
  • Implement differentiable image augmentation in PyTorch training pipelines
  • Perform GPU-accelerated geometric transformations for robotics perception
  • Integrate pre-trained vision models like YuNet or LoFTR into spatial AI workflows

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

PythonRobotics adds Lidar-to-grid mapping for autonomous navigation 🔗

New k-means clustering and rectangle fitting enhance SLAM pipeline accuracy

AtsushiSakai/PythonRobotics · Python · 30.4k stars Est. 2016

The project updated its mapping module with Lidar-to-grid conversion and k-means object clustering, improving environmental perception for path planning. Rectangle fitting now refines obstacle boundaries in grid maps, reducing false positives in dynamic window approach navigation.

These changes tighten the SLAM loop by feeding cleaner map data into ICP matching and FastSLAM 1.0. Builders using the repo for autonomous vehicle prototyping gain sharper localization without switching frameworks. The catch: Minimum dependency claims falter as cvxpy and SciPy introduce version-locking risks in long-term deployments.

Previously in The Times “covered” — Aug 18

Use Cases
  • Students implement RRT path planning in simulation
  • Researchers test EKF localization with real Lidar streams
  • Developers prototype drone 3D trajectory following using PSO optimization

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

Pinocchio Adds URDF v1.2 Support for Acceleration and Jerk Limits 🔗

Release v4.1.0 enables parsing extended joint constraints from URDF files for motion planning

stack-of-tasks/pinocchio · C++ · 3.7k stars Est. 2014

Pinocchio v4.1.

0 now parses URDF v1.2 <limit> tags to extract acceleration and jerk bounds, feeding them into ModelTpl fields like lowerAccelerationLimit and upperJerkLimit. This update improves fidelity in simulating real-world actuator constraints for robots with speed-limited joints. The library also fixed a string-stream parsing bug that broke Python pickle for models with infinite limits and adjusted Windows build checks to use _WIN32. Built on Eigen and FCL, Pinocchio remains central to tools like Crocoddyl and Stack-of-Tasks for dynamics and control.
The catch: Despite active development, 112 open issues suggest ongoing challenges in maintaining stability across its broad feature set and platforms.

Use Cases
  • Simulate robot arms with joint speed and jerk limits
  • Compute analytical derivatives for trajectory optimization
  • Enable collision-aware motion planning in humanoid robots

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

Quick Hits

copper-rs Build, run, and replay robot behavior deterministically with a Rust-based OS for reliable robotics development. 1.5k
ogre Render high-performance graphics across platforms using a flexible C++ engine with bindings for Python, C#, and Java. 4.6k
ardupilot Develop and deploy autonomous vehicles — planes, copters, rovers, and subs — with a unified open-source flight stack. 15.7k
navigation2 Enable robust robot navigation in ROS 2 with advanced path planning, localization, and obstacle avoidance capabilities. 4.6k
ontology-development-kit Manage the full ontology lifecycle — from creation to deployment — using Dockerized tools for consistent, reproducible knowledge modeling. 363

Agent-less Go vulnerability scanner vuls adds multi-OS support in latest release 🔗

v0.40.1 resolves tag-push conflicts while maintaining NVD, OVAL, and container scanning

future-architect/vuls · Go · 12.2k stars Est. 2016 · Latest: v0.40.1

The vuls project, a Go-based agent-less vulnerability scanner for Linux, FreeBSD, containers, and language libraries, released v0.40.

1 on August 25, 2026. This update fixes a release-tagging issue where multiple tags pointed at HEAD, ensuring reliable automated deployments via CI/CD pipelines. The tool continues to scan systems without agents by leveraging local package managers (yum, zypper, pkg-audit) and vulnerability feeds including NVD, JVN, OVAL, and distro-specific advisories like RHSA and Debian Security Tracker.
Vuls generates regular reports through cron, identifying affected servers and linking vulnerabilities to exploit databases such as Metasploit and Exploit Database. It supports major distributions including Alpine, Amazon Linux, Debian, Ubuntu, RHEL, and SUSE, alongside Windows, macOS, and cloud environments.
The catch: Despite broad OS coverage, vuls lacks native Windows agent-less scanning and relies on external tools for Windows CVE correlation, creating a visibility gap in heterogeneous environments.

Previously in The Times “covered” — Aug 24

Use Cases
  • Sysadmins patching Linux containers weekly
  • Security teams auditing FreeBSD servers
  • DevOps automating vulnerability reports in CI/CD

Source: future-architect/vuls — based on the README and release notes.

More Stories

Agentic Bug Hunter v6.0.0 expands AI provider freedom for ethical hackers 🔗

Adds OpenRouter and OrcaRouter support, improves Ollama model handling, and streamlines install/uninstall workflows

Awarexone/Agentic-Bug-Hunter · Python · 4.4k stars 5mo old

Awarexone/Agentic-Bug-Hunter’s v6.0.

0 release lets security researchers run the terminal-based bug bounty toolkit on local or cloud AI models without a Claude subscription. It now supports OpenRouter and OrcaRouter as OpenAI-compatible providers, with smarter Ollama model selection that remembers user choices and avoids silent fallbacks. The installer refreshes the managed bughunter command path on rerun, preventing older installations from shadowing updates, while uninstall.sh preserves config by default. The toolkit automates recon, testing, validation, and report generation for platforms like HackerOne and Bugcrowd, learning from past targets to improve future scans. The catch: Reliance on external AI providers means scan quality and speed vary significantly based on the chosen model’s capabilities and API availability, with no guaranteed consistency across free tiers.

Use Cases
  • Security researchers testing web apps for bounties using local LLMs
  • Penetration teams automating vuln validation across multiple targets
  • Independent hunters generating submission-ready reports without paid AI tiers

Source: Awarexone/Agentic-Bug-Hunter — based on the README and release notes.

Strix v1.5.3 Fixes Browser Session Leaks in AI Pentesting Containers 🔗

OpenRouter attribution added; autonomous agents now validate exploits with working PoCs

usestrix/strix · Python · 57.9k stars Est. 2025

Strix released v1.5.

3, patching four container-related bugs that caused abandoned browser sessions and dropped environment variables during autonomous penetration tests. The update ensures browser profiles persist as root and fixes LLM tool-call handling for OpenRouter and tool-less requests. Developers using Strix’s AI agents gain more reliable dynamic scanning, where exploits are validated via actual proofs-of-concept—not static false positives. Integrated with GitHub Actions, it blocks insecure code in PRs and auto-generates remediation patches. The tool requires Docker and an LLM API key, saving results to strix_runs/.
The catch: 307 open issues suggest scalability and edge-case stability remain concerns for enterprise workloads.

Previously in The Times “covered” — Aug 20

Use Cases
  • Security teams automate vulnerability validation in CI/CD pipelines
  • Developers generate working exploit PoCs for bug bounty submissions
  • AppSec engineers remediate flaws with AI-suggested pull requests

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

AI Pentester Shannon Fixes Failed Scan Termination in v2.5.3 🔗

Open-source tool now surfaces error reasons for aborted security tests in CI pipelines

KeygraphHQ/shannon · TypeScript · 47.2k stars 11mo old

Shannon v2.5.

3 resolves a critical workflow issue where failed scans in Temporal would hang without explanation. The fix ensures terminated scans surface clear failure reasons, improving reliability for teams running automated pentests on every build. Developers using the TypeScript-based agent with Node.js 18+ and Docker gain faster feedback on API and web app vulnerabilities. Shannon still requires manual setup of AI provider credentials and browser automation tools, limiting plug-and-play adoption. The catch: Real exploit execution demands careful sandboxing to avoid unintended system impact during testing.

Previously in The Times “covered” — Aug 20

Use Cases
  • DevSecOps teams automating security checks in CI/CD
  • API developers validating endpoint protections pre-release
  • Security engineers reproducing findings from source analysis

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

Quick Hits

Anthropic-Cybersecurity-Skills Provides 817 structured cybersecurity skills mapped to six major frameworks for AI agents to enhance threat detection and response across 29 domains. 31k
sniffnet Enables comfortable, real-time network traffic monitoring with an intuitive interface for developers and security practitioners. 40.7k
SWE-agent Automatically resolves GitHub issues using language models, with applications in software engineering, offensive security, and competitive coding. 20.1k
maigret Aggregates personal data from over 3,000 websites using just a username to build detailed dossiers for investigative purposes. 37k
opennhp Delivers a lightweight, cryptography-driven toolkit to enforce Zero Trust security across infrastructure, apps, and data in AI-centric environments. 13.9k

x64dbg-MCP Server Brings Agentic Debugging to Reverse Engineering Workflows 🔗

v1.1 adds 12 new tools and blocking commands to improve AI agent reliability during live debugging sessions

duty1g/x64dbg-mcp-server · Zig · ▲ 250 in 1d 2d old · Latest: v1.1

The duty1g/x64dbg-mcp-server project enables AI assistants to control x64dbg over HTTP via the Model Context Protocol, turning the debugger into a programmable agent. Built with Zig, it produces a zero-dependency, single-binary plugin that cross-compiles for both x32 and x64 architectures.

The server exposes 84 MCP tools for full debugger control — including breakpoint management, memory reading, register dumping, disassembly, and PE analysis — along with 22 event callbacks for real-time responses to debugger state changes.

Version 1.1 focuses on reliability for AI agents in automated workflows. New tools like WaitForEvent let HTTP-only clients poll for debugger events without SSE, while SetMemoryBreakpoint and SetExceptionBreakpoint grant fine-grained control over execution halts. Critical behavior updates make run and ExecuteDebuggerCommand blocking — pausing until the target halts — preventing agents from issuing commands while the debuggee runs freely. The server also now intercepts mistaken MCP tool names passed as debugger commands, redirecting users to the correct tool and reducing agent errors.

Authentication is mandatory via bearer token, auto-generated on first run, and the server auto-starts when x64dbg launches. Configuration — including IP, port, and token — is adjustable through a dialog in x64dbg’s Plugins menu, which triggers an automatic restart.

The catch: While the plugin eliminates runtime dependencies and simplifies deployment, its reliance on x64dbg limits use to Windows environments, and agent-driven debugging still requires careful tool chaining to avoid state conflicts during complex malware analysis or binary hunting sessions.

Previously in The Times “covered” — Aug 24

Use Cases
  • Malware analysts automate breakpoint setting and memory inspection via AI agents
  • Reverse engineers use natural language to step through code and extract strings
  • Security researchers automate PE analysis and OEP detection in headless debug sessions

Source: duty1g/x64dbg-mcp-server — based on the README and release notes.

More Stories

RTK’s latest hooks now compress Mistral Vibe CLI output for agents 🔗

Rust-based proxy adds transparent filtering to cut LLM token use by up to 90%

rtk-ai/rtk · Rust · ▲ 124 in 1d 7mo old

The rtk-ai/rtk project released v0.45.

0 with hook support for Mistral Vibe CLI, enabling automatic command rewriting that strips noise and groups similar output before it reaches LLMs. Written in Rust with zero dependencies, the single binary now handles over 100 dev commands like git status or npm test, applying smart filtering, truncation, and deduplication to reduce token consumption by 60-90%. Agents using hook-based or plugin-based workflows receive compact output without explicit rtk calls, streamlining AI-assisted coding.
The catch: Hooks only work on Bash tool calls; Claude Code’s built-in Read, Grep, and Glob tools bypass the proxy unless invoked via shell commands like rtk read.

Previously in The Times “covered” — Aug 18

Use Cases
  • Developers reducing LLM costs during agentic coding sessions
  • Teams using Claude Code or Hermes to minimize context window bloat
  • Rust enthusiasts deploying zero-dependency CLI proxies in CI pipelines

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

Awesome Go curates essential libraries and frameworks for Go developers worldwide 🔗

Maintained since 2014, it guides tool selection across 50+ categories from AI to IoT

avelino/awesome-go · Go · 182.2k stars Est. 2014

Avelino/awesome-go aggregates vetted Go packages, helping builders avoid dependency fatigue. Its Slack community via Golang Bridge enables real-time discussion, while contribution guidelines ensure quality.

Recent activity shows sustained upkeep with a commit today and 216 open issues reflecting ongoing refinement. The list spans standard CLI tools, database drivers, machine learning, and hardware interfaces, updated through community pull requests.
The catch: Curation relies on volunteer judgment, so niche or emerging tools may lag inclusion despite merit.

Use Cases
  • Devs pick web frameworks like Gin or Echo for rapid API development
  • Teams select database drivers such as go-sql-driver/mysql for MySQL integration
  • Engineers find logging libraries like Zap or Zerolog for production observability

Source: avelino/awesome-go — based on the project README.

Moby Project Updates Container Toolkit with BuildKit and Network Fixes 🔗

Release 29.7.2 patches image pull regressions and upgrades BuildKit to v0.32.2

moby/moby · Go · ▲ 9 in 1d Est. 2013

The Moby Project released docker-v29.7.

2, fixing a regression that caused docker service create to panic when duplicate environment variables were passed. Image pulls now work correctly on older Linux kernels after a fix for file permission handling during device node processing. BuildKit was updated to v0.32.2, improving build performance and reliability. These changes address stability issues introduced in the 29.7.0 release, affecting users relying on automated CI/CD pipelines and containerized workloads. The project maintains its modular design, allowing developers to swap components like the container registry or runtime while keeping secure defaults. The catch: Despite frequent updates, Moby’s developer-focused documentation and lack of end-user tooling create a steep learning curve for teams seeking turnkey container solutions.

Use Cases
  • Developers building custom container platforms
  • Engineers swapping container runtimes in orchestration systems
  • Integrators assembling secure, modular container images at scale

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

Quick Hits

redis Redis empowers developers to build real-time applications with ultra-fast caching, flexible data structures, and advanced vector/document querying for low-latency performance. 76.1k
obs-studio OBS Studio enables high-quality live streaming and screen recording with powerful customization, all free and open source for creators and professionals. 75.4k
PowerToys PowerToys enhances Windows productivity through a suite of lightweight utilities that streamline workflows, customize interfaces, and automate repetitive tasks. 138.1k
linux The Linux kernel provides the foundational, open-source operating system core that powers everything from embedded devices to supercomputers with unmatched stability and scalability. 244.3k
fx fx is a Unix-like coding agent in Zig that lets developers interact with codebases through natural language commands, boosting development speed and clarity. 2.4k

HAL Adds Precise Simulation Controls for FPGA Netlist Debugging 🔗

New timeout property and waveform fixes improve reliability in hardware reverse-engineering workflows

emsec/hal · C++ · 821 stars Est. 2019 · Latest: v4.5.0

The Hardware Analyzer (HAL) framework from emsec/hal has updated its simulation plugin with tighter controls and bug fixes, enhancing its utility for FPGA and ASIC netlist analysis. The release adds a timeout_after_sec property to prevent hanging simulations, resolves issues loading prior simulation data, and refines waveform event handling to block duplicates.

Selection in simulation results now supports range selection with a single click, and external tool integration (e.g., Saleae) for waveform export has been stabilized.

These changes address long-standing friction points in interactive netlist inspection, where engineers isolate modules, trace dataflow, and validate logic behavior. HAL’s C++ core, Python bindings, and plugin system remain central to its appeal, enabling scripted analysis and GUI-driven exploration of complex gate-level designs. The project is actively used in academic settings at Ruhr University Bochum and backed by the Max Planck Institute for Security and Privacy.

The catch: Despite its maturity, HAL still lacks built-in support for modern HDL simulation standards like VHDL-2008 or SystemVerilog assertions, limiting its plug-and-play use in contemporary digital design flows without external tooling.

Previously in The Times “covered” — Aug 24

Use Cases
  • FPGA designers validating netlist behavior post-synthesis
  • Security researchers analyzing IC layouts for side-channel leaks
  • Academia teaching hardware reverse engineering with live netlists

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

More Stories

Microsoft's MakeCode Target Powers DIY Microcontroller Boards 🔗

Enables breadboard-friendly coding for hobbyist hardware projects

microsoft/pxt-maker · TypeScript · 129 stars Est. 2017

Microsoft/pxt-maker provides a TypeScript-based code editor for Maker boards, built on the MakeCode platform. It supports microcontrollers suited for breadboarding, letting users write, build, and flash code locally via Node.

js and Docker. The project includes tooling for CODAL integration and local documentation serving. Recent activity shows ongoing maintenance with commits as recent as yesterday and 47 open issues.
The catch: Setup requires multiple repository clones, dependency linking, and manual environment configuration, posing a steep onboarding barrier for newcomers.

Use Cases
  • Hobbyists programming STM32 or nRF52 boards for sensor projects
  • Educators teaching embedded systems with visual and text code
  • Developers prototyping IoT devices using MakeCode's block-to-text workflow

Source: microsoft/pxt-maker — based on the project README.

C++ Roadmap Project Updates Learning Path for 2026 🔗

Maintains draw.io SVG format with recent commit activity despite aging repo

salmer/CppDeveloperRoadmap · HTML · 3.5k stars Est. 2021

The salmer/CppDeveloperRoadmap project provides a structured learning guide for C++ developers, consolidating core competencies into a single editable draw.io SVG file.

Updated as recently as one day ago, it covers language fundamentals, libraries, AI integration, and coding standards under a CC BY-NC-SA 4.0 license. The roadmap targets beginners to mid-level developers seeking focused study paths without extraneous material.
The catch: The project shows no indication of format modernization beyond draw.io, potentially limiting accessibility for teams standardized on other diagramming tools or version-controlled documentation workflows.

Use Cases
  • Junior developers planning C++ study schedules
  • Teams onboarding engineers to legacy systems
  • Educators designing curriculum for systems programming courses

Source: salmer/CppDeveloperRoadmap — based on the project README.

Blinker Library Simplifies IoT Development Across Embedded Platforms 🔗

Supports Arduino R4, ESP32, and ESP8266 with cloud-based data transmission

blinker-iot/blinker-library · C++ · 5.2k stars Est. 2018

The blinker-iot/blinker-library provides a cross-hardware IoT solution for embedded developers, enabling rapid prototyping of smart home and data monitoring projects. Built in C++, it integrates with Arduino, ESP32, and ESP8266 platforms using public cloud services for data transmission and storage.

The library leverages dependencies like ArduinoJson, Adafruit_MQTT_Library, and painlessMesh to support MQTT, WebSockets, and mesh networking. Recent updates focus on bug fixes and compatibility with ESP8266/Arduino 2.7.4+ and ESP32/Arduino 1.0.5+.
The catch: Despite recent commits, the project shows signs of aging with only six open issues and infrequent major feature updates, raising questions about long-term maintenance and modern IoT protocol support.

Use Cases
  • Developers building smart home sensors with ESP32
  • Engineers creating data loggers using ESP8266 and MQTT
  • Makers prototyping wireless device networks via WebSockets

Source: blinker-iot/blinker-library — based on the README and release notes.

Quick Hits

espectre Detects motion via Wi-Fi signal changes (CSI) using spectral analysis and integrates seamlessly with Home Assistant for smart home automation. 9.1k
OpenAVRc_Dev Advances AVR microcontroller development with enhanced tooling and open-source firmware for next-gen embedded audio and control systems. 70
openflight Parses, manipulates, and validates OpenFlight 3D scene data — an industry-standard format for flight simulation and visual modeling. 962
p3a Plays and renders pixel art animations directly on ESP32-P4 hardware, enabling low-power, real-time retro graphics displays. 88
TuyaOpen Provides a unified AI+IoT framework for Tuya’s T2/T3/T5AI and ESP32 chips, accelerating hardware integration of intelligent agents and fast IoT workflows. 1.8k

Rust GUI Library Egui Adds Drag Detection Fix in Latest Release 🔗

Patch resolves widget interaction bug affecting immediate mode interfaces in games and apps

emilk/egui · Rust · ▲ 10 in 1d Est. 2019 · Latest: 0.36.1

Egui, the Rust-based immediate mode GUI library, released version 0.36.

1 with a targeted fix for drag-and-drop behavior. The update corrects Sense::drag to prevent false drag detections when clicking widgets positioned above interactive elements—a subtle but impactful issue in layered UIs common in game development and complex applications.

Egui remains notable for its portability across web (via WebAssembly) and native platforms, requiring only the ability to draw textured triangles for integration. Its official framework, eframe, supports deployment on Web, Linux, macOS, Windows, and Android. The library emphasizes ease of use, aiming to minimize panics and developer errors while targeting 60 FPS responsiveness even in debug builds.

The fix, contributed by @lucasmerlin and merged in pull request #8396, addresses a regression that could disrupt user input handling in interfaces with overlapping widgets. While egui continues to see active development—with commits as recent as zero days ago and over 1,100 open issues—the project maintains its focus on simplicity and broad compatibility rather than feature bloat.

The catch: Egui’s immediate mode paradigm, while reducing boilerplate, may not suit developers preferring retained-mode GUI architectures or those requiring complex state persistence across frames without manual management.

Previously in The Times “covered” — Aug 20

Use Cases
  • Game developers integrating UI into custom engines
  • Rust builders creating cross-platform desktop tools
  • WebAssembly engineers designing interactive browser apps

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

More Stories

EnTT 4.0.0 drops legacy macros, adds concepts for safer C++ ECS 🔗

Header-only library used in Minecraft and ArcGIS refines type system with C++20 features

skypjack/entt · C++ · 13k stars Est. 2017

EnTT 4.0.

0 removes ENTT_CONSTEXPR, renames exception macros, and drops legacy utilities like popcount and next_power_of_two. It adds C++20 concepts such as enum_bitmask and cvref_unqualified, refines type_info and component_traits to favor concepts over SFINAE, and updates meta::name to return std::string_view. The ECS core sees performance gains in storage and component destruction for trivially types, with view-to-view conversion now supported. Used in production by Mojang and Esri, EnTT remains a dependency-free, header-only solution for game dev and data-oriented design.
The catch: Its deep reliance on advanced C++20 concepts may steepen the learning curve for teams on older toolchains or strict C++17 baselines.

Use Cases
  • Game studios building high-performance entity-component systems
  • Developers needing reflection and serialization in data-oriented applications
  • Teams requiring header-only, zero-dependency C++ libraries for cross-platform deployment

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

Godot Engine 4.7.2 fixes bugs, maintains cross-platform stability 🔗

Maintenance update keeps engine compatible with prior 4.x releases for seamless upgrades

godotengine/godot · C++ · 116.1k stars Est. 2014

Godot 4.7.

2 arrives as a stability-focused maintenance release, patching bugs and usability issues across its 2D and 3D toolkit. Built in C++ and licensed under MIT, the engine continues to support one-click exports to Windows, macOS, Linux, Android, iOS, Web, and consoles. With over 26,000 forks and nearly 19,000 open issues, active community contributions drive its evolution. The update requires no migration steps from earlier 4.x versions, preserving project compatibility.
The catch: Despite frequent commits, the high volume of open issues suggests ongoing challenges in resolving complex engine bugs at scale.

Previously in The Times “covered” — Aug 21

Use Cases
  • Indie developers shipping 2D platformers to mobile and Steam
  • Studios prototyping 3D simulations for architectural visualization
  • Educators teaching game design with a free, open-source engine

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

Quick Hits

o3de Enables free creation of AAA games, cinematic 3D worlds, and high-fidelity simulations using a powerful, open-source, multi-platform 3D engine. 9.6k
cyclopsLevelBuilder Lets you design and block out game levels directly inside the Godot editor with an intuitive plugin workflow. 1.6k
raylib Provides a simple, lightweight C library for easy and enjoyable video game programming and prototyping. 34.5k
comedot Offers a Godot-based template and component framework to streamline 2D game development with reusable systems. 519
func_godot_plugin Adds Quake .map file import and editing support to Godot 4, enabling classic level design in modern workflows. 827
bevy Delivers a refreshingly simple, data-driven game engine in Rust that prioritizes flexibility and performance for developers. 47.8k
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