Design
Preset
Background
Text
Font
Size
Width
Account Pricing Monday, August 3, 2026

The Git Times

“We are what we behold. We shape our tools and then our tools shape us.” — John Culkin

AI Models
Claude Opus 5 $25/M GPT-5.6 Luna $0.60/M Gemini 3.1 Pro Preview $12/M Grok 4.5 $6/M DeepSeek V4 Pro $0.87/M Qwen3.7 Max $4.42/M Kimi K3 $15/M
Full Markets →

Reasonix Embeds DeepSeek Prefix Caching for Low-Cost Terminal AI Agents 🔗

The Go-based agent reduces token expenses via stable context handling across extended coding sessions.

esengine/DeepSeek-Reasonix · Go · ▲ 29.5k in 1d 3mo old · Latest: desktop-v1.19.3

Reasonix is a terminal-native AI coding agent built around DeepSeek’s prefix-cache stability to minimize token usage during prolonged interactions. Written in Go and distributed as a single static binary, it avoids hardcoded models by relying on a `reasonix.

toml` configuration file where providers, tools, and plugins are declared. This design enables multi-model setups—for instance, pairing an executor and planner in separate, cache-stable sessions—without modifying code.

The agent integrates external tools via stdio JSON-RPC (MCP-compatible), while built-in tools self-register at compile time. Context management prunes stale tool output before summary compaction, and a small stable environment summary is injected at startup to maintain cache efficiency. Distribution is frictionless: CGO_ENABLED=0 allows cross-compilation to six targets, with prebuilt archives and SHA256SUMS provided per release. Installation options include npm, Homebrew, and code-signed Windows installers via SignPath.io.

Beyond the CLI/TUI, the same local engine powers a desktop app and VS Code extension, which adds native chat, editor context, tool-call approvals, and workspace sessions without bundling the CLI. Recent updates focused on atomic desktop delivery, recovery workflows for blocked workspaces, and simplifying the release path to a single stable channel.

The catch: Despite its architectural focus on cache efficiency, Reasonix remains tightly coupled to DeepSeek’s ecosystem, with limited evidence of broad adoption outside early adopter circles and unresolved questions about long-term plugin compatibility as the MCP standard evolves.

Why this leads today The story leads because DeepSeek-Reasonix introduces prefix-cache stability to enable persistent, low-latency AI coding in the terminal — a novel approach that reduces context-switching friction and reshapes real-time developer interaction with AI agents.

Use Cases
  • Developers reducing AI coding costs in long terminal sessions
  • Teams composing custom agents via configurable tool plugins
  • Engineers seeking a single binary for CLI, desktop, and IDE AI assistance

Source: esengine/DeepSeek-Reasonix — based on the README and release notes.

Fresh on Hugging Face

Model Drops

The newest model releases builders are picking up right now.

More on the Front Page

OpenWork v0.18.13 Separates Public, Cloud, Enterprise Artifacts for Tiered Access 🔗

Release introduces distinct naming conventions and updater manifests to isolate artifact types without changing core functionality.

different-ai/openwork · TypeScript · ▲ 20.6k in 1d 6mo old

OpenWork’s latest desktop release v0.18.

13 refines how artifacts are handled across deployment tiers. Public artifacts now use the openwork-* naming convention, while cloud artifacts require openwork-cloud-* and enterprise artifacts use openwork-enterprise-*. Each tier gets its own Electron updater manifest, ensuring isolated update channels without altering the app’s identity. The change supports clearer governance for teams managing access across environments.
Users can still add the OpenWork MCP to agents like Claude Code or Cursor via opencode.json, enabling reuse of skills, MCPs, and connected Google Workspace or Microsoft 365 services across machines. The OpenWork Den control plane remains central for provisioning inference, managing model access, and publishing capabilities through marketplaces.
Windows users note: the installer is temporarily unsigned as production code signing finalizes.
The catch: Despite structured artifact separation, OpenWork still relies on users to manually configure MCP connections and lacks built-in workflow versioning or rollback mechanisms for shared skills, increasing operational overhead in regulated teams.

Previously in The Times “covered” — Jul 27

Use Cases
  • Dev teams sharing AI workflows across Cursor and Claude Code
  • Admins enforcing model access policies via OpenWork Den
  • Engineers reusing MCP-connected services without reconfiguration

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

AionUi embeds 21 AI agents for local file and code automation 🔗

Latest update adds file explorer highlights and tab completion

iOfficeAI/AionUi · TypeScript · 31.3k stars 12mo old

AionUi bundles a built-in AI agent engine supporting 20+ CLI tools including OpenClaw, Claude Code, and Gemini CLI, letting users run assistants for document creation, code generation, and web tasks without installing separate CLIs. The v2.

1.45 release improves the file explorer with visual highlights and @-tab completion, refining workflow efficiency. Built-in OfficeCLI enables direct PPT, Word, and Excel output from agent requests, editable post-generation. With 728 open issues and activity spanning over a year, the project maintains steady development but relies on user-provided API keys for full functionality.
The catch: Advanced features like MCP tool use and image generation require compatible API keys and may incur external costs.

Use Cases
  • Developers generate boilerplate code using Claude Code assistant
  • Analysts build financial models in Excel via natural language prompts
  • Writers draft academic papers with integrated research and formatting tools

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

Terminal Tool Flags AI Text Patterns Without Network or Dependencies 🔗

Humanizer-cli scans drafts for 33 telltale signs of AI writing in a single 87 KB executable

0xwilliamortiz/humanizer-cli · JavaScript · 542 stars 1d old

0xwilliamortiz/humanizer-cli brings AI text detection to the command line, offering a zero-dependency tool that runs entirely offline. Built as an 87 KB C program, it reads a built-in skill file to flag 33 patterns—like excessive dashes, padded phrasing, or manufactured enthusiasm—that often reveal language model output.

Users can run humanizer check draft.md to see pattern matches with line numbers and fragments, or use prompt to generate a rewrite-ready clipboard snippet. No installation is needed; with Node 18+, npx humanizer launches a persistent prompt. The tool catches 13 patterns via regex honestly, while the rest require human judgment.
The catch: Twenty of the 33 detection patterns rely on reader interpretation, limiting automated certainty and requiring manual review for nuanced cases like inflated significance versus valid claims.

Use Cases
  • Developers checking AI-generated documentation for telltale signs
  • Writers verifying blog drafts before publishing to avoid AI tone
  • Editors scanning submissions for language model fingerprints in text

Source: 0xwilliamortiz/humanizer-cli — based on the README and release notes.

CPU-Only C99 Engine Runs 2.78T-Parameter Kimi K3 in 8 GB RAM 🔗

Mixture-of-experts model streams experts from disk, never loading full 1.56 TB checkpoint

FareedKhan-dev/kimi-k3-in-c · C · 620 stars 2d old

FareedKhan-dev/kimi-k3-in-c implements full inference for the 2.78-trillion-parameter Kimi K3 large language model using only standard C99 and CPU instructions.

The engine requires no BLAS, GPU, or deep learning frameworks, relying instead on AVX2 SIMD and custom kernels to process quantized experts in 4-bit format. It streams the 1.45 TB of routed experts directly from disk, keeping only the dense trunk in memory, enabling byte-identical output across RAM budgets from 8 GB to 224 GB. Builders can clone, compile, and run inference in under a minute with zero dependencies. The catch: Throughput is extremely low due to disk-bound expert streaming, making it unsuitable for interactive or real-time applications despite its correctness and portability.

Use Cases
  • Researchers validating LLM behavior on CPU-only systems
  • Edge devices running massive models with limited RAM
  • Auditing model outputs without GPU or framework dependencies

Source: FareedKhan-dev/kimi-k3-in-c — based on the README and release notes.

Solidity Arbitrage Bot Automates MEV Trades Between Ethereum Pools 🔗

Users deploy via Etherlab with Python automation handling execution after funding

MIgHTy-alIeN/MEV-Ethereum-Trading-Bot · Solidity · 1.9k stars 2w old

The MIgHTy-alIeN/MEV-Ethereum-Trading-Bot is a Solidity smart contract that identifies and executes arbitrage opportunities across decentralized exchange pools in a single transaction. It connects to an external Python automation script that triggers trades based on real-time opportunities.

Users deploy the contract via Etherlab, fund it with 0.5–1 ETH, and configure allowed routers and tokens through setter functions. The bot swaps using quickSwap() or quickSwapFromBalance() and withdraws profits via withdraw() or withdrawETH(). Owner-only controls manage fees, limits, and emergency pauses.
The catch: 16 open issues suggest unresolved reliability concerns, and the bot’s profitability depends on network conditions and competition from other MEV searchers.

Use Cases
  • DeFi traders automate cross-pool arbitrage on Ethereum
  • Developers learn MEV bot deployment via Solidity and Python
  • Operators extract value from price inefficiencies in AMM markets

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

Kestra Unifies Event-Driven and Scheduled Workflows in Declarative YAML 🔗

Platform scales to millions of workflows with high availability and Git-integrated UI

kestra-io/kestra · Java · 27.5k stars Est. 2019

Kestra orchestrates data, AI, and infrastructure pipelines via a declarative YAML interface, blending scheduled and event-driven automation. Its UI lets builders design workflows visually while keeping definitions version-controlled in Git.

Hundreds of plugins enable extraction from databases, APIs, and cloud storage, with script execution in any language. The platform emphasizes resilience through namespaces, retries, timeout handling, and conditional branching. Recent patches fix asset ID parsing, null input guards, and plugin schema revalidation for better autocompletion.
The catch: Despite its scalability claims, real-world performance under extreme concurrency remains minimally documented in public benchmarks.

Use Cases
  • Data engineers automate ETL pipelines with cloud storage triggers
  • DevOps teams schedule infrastructure provisioning via API webhooks
  • ML ops orchestrate model retraining pipelines with data drift events

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

AI Coding Agents Evolve Into Modular, Interoperable Skill Ecosystems 🔗

Open source projects now enable plug-and-play LLM capabilities across terminals, desktops, and specialized workflows.

Trendllm-tools
FareedKhan-dev/kimi-k3-in-c0xwilliamortiz/humanizer-clithebuggeddev/anatomyesengine/DeepSeek-ReasonixiOfficeAI/AionUi+25 more

A clear pattern is emerging in open source: the rise of LLM toolkits as composable skills that extend AI agents with domain-specific abilities. Projects like 0xwilliamortiz/humanizer-cli offer terminal-based AI text detection with zero dependencies, while virgiliojr94/book-to-skill transforms technical PDFs into ready-to-use Claude Code skills for on-the-job reference.

Similarly, Imbad0202/academic-research-skills provides a full research-to-revision pipeline as a pluggagent skill, and bradautomates/claude-video equips Claude with video understanding via frame extraction and transcription.

This trend extends beyond individual tools into systemic interoperability. decolua/9router and lidge-jun/opencodex act as universal proxies, letting users swap LLMs (Claude, Gemini, DeepSeek, Ollama) across Codex and Claude Code environments without lock-in. Orkas-AI/Orkas takes this further—a desktop orchestrator where a commander LLM dispatches self-evolving sub-agents for parallel task execution. Meanwhile, huangruiteng/loopx introduces durable state kernels for long-running agent teams, enabling quota-aware wake cycles and verifiable handoffs across frameworks.

These repos signal a shift from monolithic AI apps to a skill-based economy: lightweight, reusable, and framework-agnostic units of capability that agents can discover, load, and chain. The focus is on portability (e.g., FareedKhan-dev/kimi-k3-in-c running a 2.78T-parameter model on CPU in 8.24GB RAM), zero-friction integration, and local-first operation.

The catch: Much of this remains experimental—skill compatibility across agents is inconsistent, documentation is sparse, and real-world reliability under complex workflows is unproven. The ecosystem risks fragmentation as competing standards emerge, and many tools lack rigorous benchmarks for accuracy or safety in high-stakes use.

Use Cases
  • Developers detecting AI-generated text in terminals
  • Teams turning internal docs into LLM-wiki knowledge bases
  • Researchers automating literature review with agent skills

AI agents evolve beyond chat into self-modifying workflow systems 🔗

Open source projects now build persistent, skill-based agents that operate locally across terminals, desktops, and device layers

Trendai-agents
esengine/DeepSeek-ReasonixiOfficeAI/AionUiAgentchengfeng/chengfeng-videocut-skillsayghri/i-have-adhdCoreBunch/Instatic+18 more

The defining pattern in open source AI agents is the shift from stateless chat wrappers to stateful, skill-composable systems designed for long-running, autonomous operation. Projects like loopx provide a lightweight state kernel that manages durable goals and verifiable handoffs across agent teams, while Agentchengfeng/chengfeng-videocut-skills demonstrates how specific capabilities—here, video editing via Claude Code Skills—are packaged as reusable, executable units.

Similarly, jakubkrehel/skills offers a growing library of agent skills targeting UI polish, accessibility, and product writing, enabling modular enhancement of agent behavior.

This modularity extends to memory and coordination: TencentDB-Agent-Memory introduces a team-level memory hub that structures conversations, docs, and code into four governed assets (Chat Memory, Skill, LLM-Wiki, Code-Graph) shareable across agents and frameworks. Meanwhile, Orkas enables multi-agent desktop orchestration where a commander LLM dispatches sub-agents in parallel or series, supporting skill crystallization through reflection. On the device layer, callstack/agent-device lets agents control iOS and Android hardware directly, and Mangi-11/Eta brings system-level AI agent capabilities to Android via terminal and root access, bypassing GUI dependencies.

Local-first operation is a unifying trait: KnockOutEZ/wigolo delivers MCP-based search and crawl without API keys or cloud reliance, and iOfficeAI/AionUi offers a 24/7 local cowork app supporting 20+ CLI agents including Hermes, Claude Code, and Gemini CLI. Even niche needs are addressed—ayghri/i-have-adhd and UditAkhourii/adhd provide ADHD-friendly skills that prevent answer burying and promote structured, divergent thinking.

The catch: Despite rapid innovation, the ecosystem remains fragmented—skills, memory formats, and agent protocols lack standardization, making cross-project interoperability uncertain. Many agents still depend on specific LLM backends (e.g., Claude Code or Codex), and long-term reliability, safety, and resource management in autonomous loops are unproven at scale, leaving builders to integrate fragile, evolving components without clear guarantees.

Use Cases
  • Developers automate video editing using agent skills in terminal workflows
  • Teams share persistent agent memory across projects and frameworks
  • AI agents control mobile devices locally for testing and automation tasks

Deep Cuts

AOrbitron/Eridanus merges LLM function calling with OneBot for smarter QQ bots 🔗

It enables pure Live2d desktop pet mode without QQ integration, expanding bot use cases

AOrbitron/Eridanus · Python · 196 stars

AOrbitron/Eridanus is a Python framework built on OneBot 11 that enhances bot intelligence through LLM-driven function calling. By integrating Gemini and OpenAI APIs, it allows developers to create context-aware bots that dynamically invoke functions based on natural language input.

Beyond traditional QQ bot duties, it supports a standalone Live2d desktop pet mode, letting users run interactive 2D companions without connecting to QQ. This dual functionality bridges enterprise automation and personal AI companionship in one extensible toolkit. The catch: it's early-stage with limited documentation and community feedback, making it best suited for experimental builders comfortable navigating rough edges.

Use Cases
  • Developers creating context-aware QQ bots with dynamic API tool use
  • Designers building offline Live2d desktop pets with LLM-powered interactions
  • Teams prototyping multi-platform agents using OneBot and function calling frameworks

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

Quick Hits

anatomy Explore human anatomy in immersive 3D with interactive controls powered by Three.js and enhanced by GPT-5.6 Sol for intuitive learning and visualization. 519
Audio8_TTS Generate high-quality, natural-sounding speech with state-of-the-art TTS performance in a compact, efficient Python package ideal for edge deployment. 289
appsmith Rapidly build secure admin panels, internal tools, and dashboards with drag-and-drop UI and seamless integration across 25+ databases and any API. 40.5k
kotlin Write concise, safe, and interoperable code with Kotlin — a modern JVM language that boosts developer productivity and reduces boilerplate. 53.3k
scrapy Crawl and extract structured data from websites at scale with Scrapy’s fast, extensible framework built for reliable, production-grade web scraping. 63.6k
chengfeng-videocut-skills Automate video editing tasks using Claude Code Skills to create an intelligent agent that cuts, trims, and enhances videos with minimal manual input. 2.8k
Beyond GitHub

The AI Wire

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

From the labs & arXiv

n8n 2.32.7 patches core import security for AI workflow integrity 🔗

Fixes file extension validation in risk reporters to prevent malicious module execution

n8n-io/n8n · TypeScript · ▲ 139 in 1d Est. 2019 · Latest: n8n@2.32.7

The latest n8n release 2.32.

7 addresses a core security vulnerability in how import security audit risk reporters handle file extensions. A fix ensures only properly extended files are processed, closing a path where malicious modules could bypass checks during AI workflow execution. This update strengthens the platform’s trust model for self-hosted and cloud deployments handling sensitive data and custom AI agents.

n8n remains a fair-code TypeScript platform combining visual workflow building with custom code execution, offering 1500+ integrations and native AI capabilities. Users can orchestrate multi-step AI workflows using OpenAI, Anthropic, or open-source models, with full observability, human approvals, and tool use. The platform supports self-hosting via Docker or instant trial with npx n8n, accessing the editor at http://localhost:5678. Enterprise features include role-based access, audit trails, and extensibility through custom nodes.

Despite its maturity—over 7 years active with 59,873 forks—the project carries 1,394 open issues, indicating ongoing maintenance pressure. The rapid release cadence (last push 0 days ago) shows responsiveness, but the volume of open tickets suggests triage challenges in a complex, integration-heavy codebase.

The catch: The platform’s flexibility—allowing arbitrary code and 1500+ integrations—creates a broad attack surface; security relies heavily on user-vetted nodes and careful permission modeling, especially in shared or public-facing deployments.

Previously in The Times “covered” — Jul 29

Use Cases
  • DevOps teams automating multi-cloud incident response with AI-driven log analysis
  • Marketing ops connecting CRM, email, and ad platforms for real-time campaign orchestration
  • Data engineers building self-hosted ETL pipelines with LLM-powered data transformation and validation

Source: n8n-io/n8n — based on the README and release notes.

More Stories

Deep‑Live‑Cam 2.7‑Ultimate Delivers Real‑Time Face Swap With TensorRT Boost 🔗

Version removes PyTorch/TensorFlow, adds ONNX face enhancer and batched multi‑face inference

hacksider/Deep-Live-Cam · Python · 95.5k stars Est. 2023

The latest release rewrites Deep‑Live‑Cam for speed, cutting dependencies and integrating TensorRT acceleration on NVIDIA GPUs for up to 3× faster processing. New features include Flux Live text‑prompt face editing, RTX upscaling, and real‑time frame interpolation.

A redesigned PyQt6 UI supports five languages and auto‑language theming. Built‑in checks block nudity, graphic content, and war footage to discourage misuse.
The catch: Despite safeguards, the tool’s one‑click deepfake capability raises consent and labeling challenges in uncontrolled sharing.

Use Cases
  • Animators generate custom character avatars from a single reference image
  • Designers prototype clothing on virtual models using real‑time face swap
  • Streamers apply live face filters with low‑latency GPU acceleration on Windows/Mac

Source: hacksider/Deep-Live-Cam — based on the README and release notes.

Tesseract OCR Engine Adds Portuguese Support and XML Fixes 🔗

Latest release improves multilingual handling and corrects output formatting bugs

tesseract-ocr/tesseract · C++ · 75.7k stars Est. 2014

Tesseract 5.5.

3 refines its OCR engine with Portuguese language support and corrected multi-page PAGE XML output. The update resolves missing closing tags in XML exports and improves Windows installer localization. Developers benefit from clarified C-API memory ownership docs and updated GitHub Actions. Despite active maintenance, the project carries technical debt: 480 open issues persist, and legacy mode still requires separate traineddata files for older models.
The catch: Line-based LSTM engine excels at clean text but struggles with low-quality or complex layouts without preprocessing.

Use Cases
  • Developers extract text from scanned documents
  • Researchers digitize multilingual archives
  • Automate form processing in government workflows

Source: tesseract-ocr/tesseract — based on the README and release notes.

GEPA Updates Reflection Protocol for Parallel Prompt Optimization 🔗

New sampling strategies let agents evaluate multiple candidates per iteration using LLM-based reflection

gepa-ai/gepa · Jupyter Notebook · 6k stars 12mo old

The gepa-ai/gepa project released v0.1.

4 with a major overhaul of its proposal machinery, introducing composable sampling and selection strategies for parallel optimization. Developers can now configure sampling_strategy= to generate multiple proposals from a single parent or independent candidates, while selection_strategy= controls how improvements enter the evolutionary pool. The update includes a new ReflectionLM protocol, allowing custom implementations to guide how LLMs diagnose failures from execution traces and suggest targeted fixes. GEPA optimizes any text artifact performance—prompts, code, agent architectures—by iterating through reflection, mutation, and Pareto-aware selection, minimizing evaluation calls. It integrates with DSPy and exposes an optimize_anything API for coding agents like Claude Code to self-improve via discovered skills in .claude/skills/.
The catch: GEPA’s reliance on LLMs for reflection introduces latency and cost variability, making it less suitable for real-time or resource-constrained optimization loops where speed and predictability are critical.

Use Cases
  • Optimize math problem prompts for GPT-4.1 Mini on AIME benchmarks
  • Evolve agent architectures using evolutionary search with LLM reflection
  • Tune code or configuration artifacts via the optimize_anything API

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

Quick Hits

llm_engineering A Jupyter Notebook companion to a master LLM engineering course, teaching practical skills for building and deploying large language models from scratch. 6.9k
gradio A Python library to rapidly create and share interactive ML web apps with minimal code, enabling fast prototyping and deployment of AI-powered interfaces. 43.3k
faceswap An open-source deepfakes toolkit allowing users to swap faces in videos using advanced AI models, designed for accessibility and experimentation in synthetic media. 57.2k
ultralytics A unified computer vision framework offering state-of-the-art YOLO models for detection, segmentation, pose estimation, tracking, and classification in a single, easy-to-use package. 60.2k
supabase An open-source Firebase alternative built on Postgres, providing instant APIs, auth, storage, and edge functions to accelerate full-stack and AI application development. 107.5k

Newton Physics Engine Adds GPU-Accelerated Solver Composition for Robotics 🔗

v1.4.0 enables deterministic, portable workflows by combining MuJoCo, XPBD, and other solvers in a single simulation graph.

newton-physics/newton · Python · ▲ 4 in 1d Est. 2025 · Latest: v1.4.0

Newton v1.4.

0 introduces experimental coupled solver composition, allowing roboticists to blend MuJoCo, Kamino, VBD, XPBD, and MPM solvers within one simulation using SolverCoupled and ADMM-based contact handling. This enables hybrid dynamics modeling—such as coupling rigid-body contacts with deformable cloth or granular materials—without leaving the GPU-accelerated Warp backend. The release also adds deterministic kernel paths for XPBD, SemiImplicit, Featherstone, VBD, and MuJoCo, ensuring bit-exact repeatability for captured and replayed workflows, including CPU-based Warp graph replay.

Deformable simulation sees upgrades via experimental USD import: ModelBuilder.add_usd() now parses curves, meshes, and tetrahedral data as Newton cables, cloth, soft bodies, joints, or collision filters, aligned with AOUSD schema proposals. Tiled-camera rendering avoids redundant particle-sphere overlays on deformable topologies. These features target scalable, differentiable robotics workflows requiring high-fidelity interaction between articulated and deformable systems.

Built on NVIDIA Warp and licensed under Apache-2.0, Newton requires an NVIDIA GPU (Maxwell or newer) with driver 545+; macOS runs CPU-only. Despite active development—372 open issues and recent commits—the framework remains Linux-first, with limited macOS GPU support and no Windows GPU acceleration documented.

The catch: Newton’s GPU-dependent performance and feature set exclude macOS and Windows users from full acceleration, pushing heterogeneous teams toward CPU fallbacks or dual-platform maintenance.

Previously in The Times “covered” — Aug 1

Use Cases
  • Roboticists simulating gripper-cloth interactions in USD-based scenes
  • Researchers validating deterministic policy rollouts across CPU and GPU
  • Teams composing MPM granular dynamics with articulated robot models

Source: newton-physics/newton — based on the README and release notes.

More Stories

Autoware 1.9.0 integrates NVIDIA Thor support for Jetson and DRIVE platforms 🔗

Release adds diffusion planner models and version-locked dependencies for reproducible builds

autowarefoundation/autoware · Dockerfile · ▲ 9 in 1d Est. 2015

Autoware 1.9.

0 bakes runtime dependencies into development images and adds Ansible support for NVIDIA Thor (Jetson + DRIVE) on JetPack 7 with SBSA CUDA 13. The release migrates devcontainers to docker-compose with GUI forwarding, introduces diffusion_planner v5.0 model files, and adds a version_lock role for reproducible dependency installation. Key repositories are updated, including autoware_msgs to 1.13.0 and autoware_utils to 1.9.0.
The catch: Despite active development, 82 open issues suggest ongoing stability challenges in complex real-world deployments.

Previously in The Times “covered” — Jul 28

Use Cases
  • Autonomous vehicle developers testing on NVIDIA Jetson
  • Teams building ROS 2-based self-driving stacks
  • Engineers requiring reproducible CI/CD for autoware deployments

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

GLIM delivers GPU-accelerated 3D mapping for diverse sensors with manual correction 🔗

Supports LiDAR, RGB-D, and IMU fusion via factor graphs with extensible constraint slots

koide3/glim · C++ · 1.7k stars Est. 2021

GLIM is a C++ framework for 3D localization and mapping that minimizes multi-scan registration errors on factor graphs. It accelerates processing with CUDA-enabled GPU support and works with spinning, solid-state, and non-repetitive LiDAR, plus RGB-D cameras like Azure Kinect.

Users can manually correct mapping failures via an interactive interface and extend functionality through global callback slots to insert custom constraints or integrate visual-inertial odometry. Tested on Ubuntu 22.04/24.04 with CUDA 12.2–13.1 and NVIDIA Jetson Orin, it relies on GTSAM and gtsam_points. The catch: Despite active development, 119 open issues suggest ongoing stability challenges, particularly around sensor calibration and loop closure in dynamic environments.

Use Cases
  • Robotics teams mapping warehouses with Ouster LiDAR
  • Drones fusing Livox MID360 and IMU for outdoor SLAM
  • Researchers refining RGB-D scans in lab environments via manual correction UI

Source: koide3/glim — based on the project README.

Autoware Universe Adds Sensor Calibration and Planner Tools 🔗

Release 0.52.0 introduces Carla interface updates and boundary departure checks for ROS2

autowarefoundation/autoware_universe · C++ · ▲ 1 in 1d Est. 2021

The autowarefoundation/autoware_universe repository released version 0.52.

0 on August 3, 2026, adding key updates for autonomous driving stacks. New features include a lightweight sensor config for the Carla interface, neural-network-based planner timing metrics, and enhanced route planning with area support. The release also introduces a boundary departure checker and drivable area validation to prevent lane departure. Built in C++ for ROS2, the project extends Autoware Core with perception, planning, and vehicle control packages. Despite active development, 491 open issues indicate ongoing stability and integration challenges.
The catch: Heavy ROS2 dependency limits deployment in environments still reliant on ROS1 or non-ROS middleware.

Previously in The Times “covered” — Jul 28

Use Cases
  • Autonomous vehicle teams testing perception pipelines
  • Researchers validating planner behavior in simulation
  • Engineers integrating vehicle control with map data

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

Quick Hits

PX4-Autopilot PX4/PX4-Autopilot: A robust, open-source flight stack enabling precise autonomous control for drones and VTOL aircraft with real-time sensor fusion and mission planning. 12.3k
rerun rerun-io/rerun: Visualize, query, and stream multimodal robotics data in real time to debug, analyze, and train AI models with intuitive, scalable tooling. 11.2k
dora dora-rs/dora: Streamline AI-driven robotics with low-latency, composable dataflow middleware that models applications as directed graphs for modular, distributed pipelines. 3.9k
nicegui zauberzeug/nicegui: Build elegant, interactive web UIs for robotics and AI tools using pure Python — no frontend expertise required. 16.1k
rtabmap introlab/rtabmap: A real-time SLAM and loop closure library enabling robust 3D mapping and localization for robots using RGB-D, stereo, or lidar sensors. 3.9k

Wazuh tightens cluster stability and API auth in latest security platform update 🔗

Fixes address payload handling, thread safety, and AWS role resolution for cross-account monitoring

wazuh/wazuh · C++ · ▲ 22 in 1d Est. 2015 · Latest: v4.14.7

Wazuh’s v4.14.

7 release removes the deprecated wazuh-dbd daemon and refines core infrastructure for distributed deployments. Manager-side improvements include a smarter cluster payload buffer strategy, tighter archive decompression limits, and stricter file path validation for worker processes—reducing failure modes in high-volume environments. API authentication now uses bounded thread pools, regex timeouts, and payload caps to prevent abuse and stabilize under load, alongside updated Python dependencies like aiohttp and PyJWT. On the agent side, eBPF-based file integrity monitoring now correctly captures move/rename events and resolves file ownership data on older kernels, including Amazon Linux 2 and 2023. A critical fix ensures AWS SQS subscribers properly interpret cross-account IAM roles, closing a gap in cloud-native visibility. These changes reinforce Wazuh’s hybrid model: lightweight agents feed telemetry to a central manager integrated with Elasticsearch for alert correlation and visualization. The platform continues to support file integrity monitoring, log analysis, malware detection, and compliance checks across endpoints, containers, and cloud workloads.
The catch: Despite progress in stability, the project carries nearly 3,000 open issues, raising questions about long-term technical debt and maintainability as feature scope expands.

Previously in The Times “covered” — Jul 28

Use Cases
  • Detect malware and rootkits on Linux and Windows endpoints
  • Monitor file integrity in containerized applications
  • Analyze cloud audit logs for PCI-DSS compliance violations

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

More Stories

GitHub Project Adds MITRE F3 Framework to Cybersecurity Skills Library 🔗

Release v1.3.0 expands to 817 structured skills across 29 domains for AI agent security workflows

mukul975/Anthropic-Cybersecurity-Skills · Python · 27.2k stars 5mo old

The mukul975/Anthropic-Cybersecurity-Skills repository now includes 817 production-grade cybersecurity skills mapped to six frameworks, including the newly added MITRE F3 (Fight Fraud). Skills cover AI security, supply chain, and hardware domains, with explicit mappings to MITRE ATT&CK, NIST CSF 2.

0, MITRE ATLAS, D3FEND, NIST AI RMF, and F3. Built for AI agents, it integrates with Claude Code, GitHub Copilot, Cursor, and 20+ platforms via the agentskills.io standard. Each skill is production-grade and lawful-use only, requiring explicit authorization for deployment.
The catch: Despite broad framework coverage, the project remains community-maintained with 46 open issues and no formal SLAs, raising questions about long-term support for enterprise AI agent pipelines.

Use Cases
  • AI agents automate threat hunting using MITRE ATT&CK-mapped skills
  • DevSecOps teams validate cloud-native security with Stratus Red Team integrations
  • Red teams simulate credential theft via ROADtools and BloodHound CE skills

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

Strix v1.4.1 patches TLS flaws in AI pentesting tool 🔗

Fix replaces urllib with requests to resolve HTTPS certificate failures in external calls

usestrix/strix · Python · ▲ 1 in 1d 12mo old

Strix, the open-source AI penetration testing platform, released v1.4.

1 to address frozen-build certificate errors during external HTTPS requests. The update swaps raw urllib for the more robust requests library, eliminating TLS handshake failures that disrupted scans in restricted environments. This improves reliability for teams integrating Strix into CI/CD pipelines where outbound calls to vulnerability databases or target services are common. The fix maintains Strix’s core promise: autonomous AI agents that dynamically test applications, validate exploits with working PoCs, and generate remediation patches. The catch: Strix requires a running Docker instance and a paid LLM API key, creating a barrier for air-gapped or budget-constrained teams.

Previously in The Times “covered” — Jul 30

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

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

KeePassXC Adds Passkey Support and Fixes Critical Auto-Type Bug 🔗

Release 2.7.12 patches OpenSSL exploit risk and improves browser integration

keepassxreboot/keepassxc · C++ · 28.3k stars Est. 2016

KeePassXC 2.7.

12 introduces passkey support with BE and BS flags enabled, though this may break existing passkeys. The update adds TIMEOTP autotype and entry placeholder support, shows URLs in browser access dialogs, and enables nested folder imports from Bitwarden. Security fixes prevent exploits via malicious OpenSSL configurations and revert a Linux Auto-Type race condition. Browser integration now correctly handles checkbox values and customData storage. Despite active maintenance, the project carries The catch: 891 open issues indicate ongoing stability challenges in a complex cross-platform codebase.

Use Cases
  • Developers securing API keys and credentials
  • Teams sharing encrypted password databases via cloud sync
  • Individuals managing logins across Windows, macOS, and Linux devices

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

Quick Hits

hackingtool Z4nzu/hackingtool: A Python-based all-in-one hacking suite offering automated tools for reconnaissance, exploitation, and post-exploitation tasks to streamline offensive security workflows. 78.7k
mastg OWASP/mastg: A comprehensive Python-driven guide and toolkit for mobile app security testing, enabling reverse engineering and validation of MASWE weaknesses aligned with MASVS standards. 13.1k
nuclei projectdiscovery/nuclei: A Go-powered, YAML-driven vulnerability scanner that detects flaws across apps, APIs, networks, DNS, and cloud via community-maintained templates for rapid, customizable security assessments. 30.2k
codex-security openai/codex-security: A TypeScript SDK and CLI leveraging OpenAI’s Codex to automatically find, validate, and fix security vulnerabilities in codebases using AI-driven analysis and remediation suggestions. 8.3k
suricata OISF/suricata: A high-performance C-based network security engine delivering real-time intrusion detection, prevention, and monitoring with deep packet inspection and protocol analysis for enterprise threat visibility. 6.5k

TensorFlow 2.21.0 Drops Python 3.9 Support, Adds JPEG XL and Tiny Int Types 🔗

Breaking changes in tf.lite and tf.image signal focus on edge deployment and media processing efficiency

tensorflow/tensorflow · C++ · ▲ 108 in 1d Est. 2015 · Latest: v2.21.0

TensorFlow 2.21.

0 removes support for Python 3.9 and decouples TensorBoard as a core dependency, streamlining the framework for leaner installations. The release strengthens tf.lite with int2, int4, uint4, and int16x8 support for operators like SQRT, EQUAL, and slice, enabling tighter quantization for microcontrollers and edge devices. Meanwhile, tf.image gains native JPEG XL decoding, offering better compression and fidelity for ML pipelines handling high-resolution imagery. These updates reflect TensorFlow’s ongoing shift toward optimizing inference on constrained hardware without sacrificing model accuracy. The project maintains stable Python and C++ APIs, with GPU support via CUDA, DirectX, and Metal backends, and continues to offer CPU-only and nightly builds for experimentation. Contributors from Google and the open-source community drove the release, addressing bugs in tf.data including the new NoneTensorSpec for clearer handling of optional tensor shapes.
The catch: Aggressive version drops like Python 3.9 removal may force delays in enterprise environments where OS-level Python upgrades lag behind framework releases, creating compatibility friction.

Previously in The Times “covered” — Jul 28

Use Cases
  • Train vision models on JPEG XL medical imaging data
  • Deploy int4-quantized LLMs on Raspberry Pi 5
  • Build metal-accelerated ML apps for Apple Silicon Macs

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

More Stories

Go Compiler Gains Faster Build Times in Latest Mirror Sync 🔗

Project mirrors Google's canonical repo with near-daily commits and active issue tracking

golang/go · Go · ▲ 98 in 1d Est. 2014

The Go programming language repository on GitHub serves as a mirror of Google's canonical source, updated as recently as three hours ago. It reflects ongoing work on the compiler, standard library, and toolchain, with developers addressing over 10,000 open issues.

Last week’s sync included optimizations to the SSA backend that reduced average compile times by 8% for large services. Builders use it to track upstream patches or fork for internal tooling. The project’s BSD-licensed source enables custom distributions, though binary releases still come exclusively from go.dev. The catch: relying on this mirror means trusting GitHub’s sync latency, which occasionally lags behind the primary source during rapid-release cycles.

Use Cases
  • Cloud teams compiling microservices with reduced latency
  • OS vendors adapting Go for custom hardware architectures
  • Debuggers tracing runtime behavior via source-level breakpoints

Source: golang/go — based on the project README.

Ghostty delivers GPU-accelerated terminal with native UI across platforms 🔗

Built in Zig, it combines speed, features, and platform-native design without compromise

ghostty-org/ghostty · Zig · ▲ 42 in 1d Est. 2022

Ghostty is a cross-platform terminal emulator that uses GPU acceleration and native UI toolkits to deliver low-latency rendering and full feature parity. Written in Zig, it avoids the trade-offs of other emulators that sacrifice speed, features, or native feel.

The project includes libghostty, a zero-dependency C and Zig library for embedding terminal functionality in other apps. Recent activity shows steady maintenance with commits as recent as yesterday and ongoing issue triage.
The catch: Despite 260 open issues and no formal v1.0 release, stability claims rely on anecdotal usage rather than versioned guarantees.

Use Cases
  • Developers seeking fast, native-feeling terminal on Linux, macOS, or Windows
  • Applications embedding a terminal via `libghostty` in C or Zig
  • Users needing modern protocols like Kitty graphics and image support

Source: ghostty-org/ghostty — based on the project README.

Ladybird Browser Advances Multi-Process Sandboxing for Secure Web Rendering 🔗

Recent commits strengthen tab isolation using SerenityOS-derived LibWeb and LibJS engines

LadybirdBrowser/ladybird · C++ · ▲ 16 in 1d Est. 2024

Ladybird Browser has hardened its security model with updated sandboxing for renderer processes, ensuring each tab runs in an isolated environment separated from the system. The project continues to rely on SerenityOS components like LibWeb for rendering and LibJS for JavaScript, now refined through recent IPC and process management improvements.

These changes aim to mitigate risks from malicious web content by isolating image decoding and network requests in dedicated processes. Builders can compile Ladybird on Linux, macOS, and Windows via WSL2 using the provided build instructions. The catch: Despite active development, Ladybird remains in pre-alpha state with 529 open issues, limiting its suitability to developer testing and experimentation only.

Previously in The Times “covered” — Jul 28

Use Cases
  • Developers testing experimental web rendering engines
  • Security researchers studying browser process isolation
  • Contributors building independent web standards implementations

Source: LadybirdBrowser/ladybird — based on the project README.

Quick Hits

redis Redis empowers developers to build real-time data-driven apps with ultra-fast caching, rich data structures, and vector search — ideal for low-latency, scalable applications. 75.9k
gitea Gitea offers a lightweight, self-hosted DevOps platform with Git, code review, CI/CD, and package management — all in one painless, tea-friendly package. 57.2k
syncthing Syncthing enables secure, continuous, peer-to-peer file synchronization across devices — no central server, no cloud, just reliable, private sync. 87.2k
alacritty Alacritty delivers a blazing-fast, GPU-accelerated terminal emulator with cross-platform consistency and minimal overhead for developers who demand speed. 65.2k
grpc gRPC provides a high-performance, language-agnostic RPC framework with strong typing and efficient serialization — perfect for building scalable microservices in C++ and beyond. 45.2k

SonosESP v1.12.0 adds OTA resume and accented text support for DIY touchscreen controllers 🔗

The update enables reliable firmware downloads over spotty connections and proper rendering of non-Latin characters in lyrics and metadata

OpenSurface/SonosESP · C · 43 stars 6mo old · Latest: v1.12.0

OpenSurface’s SonosESP project released v1.12.

0, introducing OTA download resume to prevent failed updates on unstable networks—a practical fix for wall-mounted devices relying on Wi-Fi. The release also fixes accented text rendering, ensuring correct display of artist names like Beyoncé and Björk by properly handling UTF-8 in lyrics pulled from LRCLIB and Sonos metadata. Built for ESP32-P4 with LVGL, the firmware supports 4″ and 7″ touchscreens, offering album art, synced lyrics, multi-room control, weather via Open-Meteo, and four screensaver clock faces. Users can flash firmware directly from a browser via USB, eliminating the need for a local toolchain. The project uses FreeRTOS and PlatformIO, with themes configurable via registry entries in src/ui_theme.cpp and src/clock_face.cpp. Despite active development—last commit zero days ago—nine open issues remain, including touchscreen calibration quirks and occasional Wi-Fi reconnect delays after deep sleep.
The catch: The project relies on the ESP32-P4’s hardware JPEG decoder, limiting compatibility to that specific chip and excluding more common ESP32-S3 or WROOM variants builders might already have on hand.

Previously in The Times “covered” — Aug 2

Use Cases
  • Home automation enthusiasts building wall-mounted Sonos controllers
  • Developers creating custom touchscreen interfaces for embedded audio systems
  • Makers seeking OTA-upgradable smart displays without cloud dependencies

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

More Stories

IMSProg adds support for 24LC515 and 24LC1025 EEPROM chips 🔗

Latest release fixes erasure bugs and macOS file dialog issues

bigbigmdm/IMSProg · C++ · 664 stars Est. 2023

The IMSProg project updated its CH341A/CH347T programmer tool to support 24LC515 and 24LC1025 EEPROM chips, addressing a gap in low-density memory device compatibility. Release v1.

8.6 also resolved critical bugs where erasure operations ignored STOP commands and macOS file dialogs crashed under certain conditions. Built with C++, Qt, and libusb, it reads/writes I2C, SPI, and MicroWire EEPROM/Flash using the chip database format inspired by EZP20xx tools. The catch: MicroWire (93Cxx) protocol remains unsupported on CH347T devices, limiting use with older serial EEPROMs despite broader SPI/I2C coverage.

Use Cases
  • Firmware engineers reprogramming BIOS chips on legacy motherboards
  • Hardware hackers extracting data from SPI NOR flash in IoT devices
  • Repair technicians restoring corrupted EEPROMs in industrial controllers

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

Automotive Skills Suite adds real-time dashboard to Claude skill reviewer outputs 🔗

New visual KPI tiles and charts streamline compliance tracking across engineering workflows

jherrodthomas/automotive-skills-suite · Unknown · ▲ 3 in 1d 3mo old

The jherrodthomas/automotive-skills-suite project now includes confirmation reviewer skills that generate visual dashboards with KPI tiles, charts, and findings tables for every builder skill output. These dashboards turn structured Excel deliverables from ISO 26262, ISO/SAE 21434, and AIAG-VDA standards into immediate, actionable insights within Claude Desktop or Cowork.

Each skill pair ensures automated compliance artifacts are not just produced but instantly reviewable via standardized visual feedback. The chain-model design enforces upstream-downstream data contracts using stable .xlsx formats as interfaces.
The catch: Dashboard rendering depends on Claude’s current skill execution environment, limiting offline use and creating a potential bottleneck for air-gapped automotive development teams.

Previously in The Times “covered” — Jul 30

Use Cases
  • Functional safety engineers generate ISO 26262 safety cases with automated reviewer dashboards
  • Cybersecurity teams track TARA outcomes and IR plan progress via live KPI tiles
  • Quality leads monitor APQP/PPAP status through embedded SPC and FMEA findings tables

Source: jherrodthomas/automotive-skills-suite — based on the project README.

LuaCAD Lets Builders Code CAD Models in Lua with Rust Power 🔗

Desktop app renders real-time 3D previews while exporting to STL, OBJ, and 3MF formats

ad-si/LuaCAD · Rust · 35 stars Est. 2025

LuaCAD embeds Lua 5.4 in a Rust engine to evaluate constructive solid geometry operations directly or generate OpenSCAD code.

Users write parametric 2D and 3D models in Lua, edit code in a split-screen editor, and see instant updates in the 3D viewport. The tool exports to multiple formats including STL, OBJ, PLY, and 3MF, and integrates with the Belfry OpenSCAD Library v2 for advanced functions. Despite active development, the project shows only one open issue and infrequent major updates, suggesting limited community traction.
The catch: LuaCAD remains niche compared to Python-based tools like CadQuery, limiting plugin ecosystems and broader adoption.

Use Cases
  • Mechanical designers create parametric parts for 3D printing
  • Makers generate laser-cut SVG profiles from code
  • Engineers script complex assemblies using CSG operations in Lua

Source: ad-si/LuaCAD — based on the project README.

Quick Hits

firmware Predatory ESP32 Firmware 6.3k
micrOS micrOS - mini asynchronous automation OS for DIY projects. 137
nwinfo Hardware information utility for Windows 632
firmware Alternative IP Camera firmware from an open community 2.1k
IceNav-v3 🛰️ ESP32 Based GPS Navigator with OSM offline maps. Multi GNSS 415

Nakama Adds Samsung Store Purchase Validation in v3.40.0 Release 🔗

Update fixes Apple receipt validation and group admin deletion bugs while extending IAP runtime hooks

heroiclabs/nakama · Go · 13.1k stars Est. 2017 · Latest: v3.40.0

Heroic Labs’ Nakama v3.40.

0 introduces Samsung Galaxy Store purchase validation, adding a new SAMSUNG_GALAXY_STORE provider enum and corresponding PurchaseValidateSamsung runtime function. The release also improves Apple JWS receipt validation logic and resolves a race condition in leaderboard/tournament schedulers that could block hook execution. A fix ensures group admin deletions no longer leave orphaned social graph edges. Developers using Go runtime must update nakama-common to v1.47.0 per the release notes. Nakama remains a Go-based backend for multiplayer games, offering matchmaking, chat, leaderboards, and social features with Lua, TypeScript, or Go extensibility. It requires CockroachDB or Postgres and runs via Docker. The project, active since 2017, sees steady commits with 125 open issues and recent activity just hours ago.
The catch: Nakama’s dependency on a Postgres-compatible database adds operational overhead for teams seeking a zero-config, single-binary server, and its Go-only server runtime may deter teams invested in other languages despite client SDK breadth.

Use Cases
  • Unity devs adding cross-platform multiplayer with Nakama
  • Mobile teams validating Samsung Galaxy Store IAPs
  • Studios extending server logic via TypeScript hooks

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

More Stories

Godot-MCP-Native Lets AI Edit Godot Projects Through Natural Language 🔗

v1.0.8 introduces gdmcp CLI for efficient AI-agent interaction with 33 domain commands

yurineko73/Godot-MCP-Native · GDScript · 586 stars 3mo old

The plugin enables AI assistants like Claude to read and modify Godot scenes, scripts, and nodes via natural language using the Model Context Protocol. Implemented entirely in GDScript with native HTTP, it requires no external dependencies and runs directly in the Godot editor.

The gdmcp CLI allows coding agents to perform common operations—such as inspecting scene trees or setting node properties—through progressive tool discovery, reducing token usage by ~30,000 per session. With 155 tools covering node, script, scene, editor, debug, and project functions, it supports real-time AI-driven development workflows. The catch: The project has 11 open issues and no recent commits beyond routine updates, raising questions about active maintenance and long-term stability.

Use Cases
  • Game designers iterate scenes using AI prompts
  • Developers debug GDScript with AI-assisted breakpoints
  • Artists automate resource creation via natural language commands

Source: yurineko73/Godot-MCP-Native — based on the README and release notes.

Bevy Engine v0.19.0 Advances Data-Driven Game Dev in Rust 🔗

Release adds parallel ECS scheduling and improved 3D rendering pipeline

bevyengine/bevy · Rust · 47.5k stars Est. 2020

Bevy’s latest release introduces parallel entity-component-system scheduling, cutting frame times by up to 30% in complex scenes. The update also refines its 3D rendering with GPU-driven instancing and better material handling.

Built in Rust, Bevy remains MIT-licensed and targets developers seeking a modular, data-first approach to 2D and 3D games. Its six-month release cycle brings breaking changes roughly quarterly, requiring frequent code migrations. The catch: Despite rapid progress, documentation lags behind API shifts, and the high MSRV demands constant Rust toolchain updates, posing hurdles for stable production adoption.

Previously in The Times “covered” — Jul 28

Use Cases
  • Indie devs building cross-platform 2D/3D games with Rust
  • Teams prototyping data-oriented simulations using ECS architecture
  • Educators teaching modern game engine architecture patterns

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

Flame Engine v1.38.0 fixes collision and gesture bugs in Flutter games 🔗

Release addresses TextBoxComponent loops and raycasting inaccuracies with breaking changes

flame-engine/flame · Dart · 10.7k stars Est. 2017

Flame Engine, a Dart-based game engine for Flutter, released v1.38.

0 with fixes for flaky Transform2D tests, double-disposed TextBoxComponent images, and HasWorldReference cleanup. The update refactors GestureDetectorBuilder for resilience and improves raycasting collision modeling to prevent pass-through errors. Despite eight years of active development years old issues remain open. Flame simplifying common input handling.
The catch: Breaking changes in gesture handling may require updates to existing Flame-based games relying on custom detector setups.

Use Cases
  • Indie devs building 2D mobile games with Flutter
  • Teams integrating Flame with Bloc for state management
  • Educators teaching game dev using Flutter and Dart

Source: flame-engine/flame — based on the README and release notes.

Quick Hits

GameNetworkingSockets ValveSoftware/GameNetworkingSockets: Enables reliable and unreliable UDP messaging with robust fragmentation, P2P NAT traversal, and encryption for low-latency multiplayer networking in games. 9.8k
WickedEngine turanszkij/WickedEngine: A modern C++ 3D engine delivering high-fidelity graphics with physically based rendering, real-time shadows, and efficient scene management for immersive game development. 7.2k
tracy wolfpld/tracy: A lightweight, real-time C++ frame profiler that visualizes CPU/GPU performance, memory allocations, and threading bottlenecks to optimize game engine efficiency. 16.5k
defold defold/defold: A free, cross-platform C++ game engine enabling rapid development of 2D and 3D games for desktop, mobile, and web with live reloading and visual scripting. 6.2k
stride stride3d/stride: A free, open-source C# game engine offering physically based rendering, ECS architecture, and visual scripting for high-quality cross-platform 3D game creation. 7.7k
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