Tech Updates 28-08-2025

AGENTS.md

https://camo.githubusercontent.com/0441657806fba5628054f7f2a01cdfa42acceb81913b6b66e66cafd004caa840/68747470733a2f2f6167656e74732e6d642f6f672e706e67

AGENTS.md is a dedicated markdown file designed to guide AI coding agents working on your project. It contains setup commands, code style rules, testing instructions, and workflow guidelines—information that may be too detailed or agent-specific for README.md. By keeping these instructions separate, AGENTS.md ensures automation is predictable and agents can contribute efficiently. The format is open, flexible, and widely supported by popular agent tools, making it easy to maintain clear, up-to-date guidance for both monorepos and single-package projects.

https://github.com/openai/agents.md

https://agents.md/

DinoV3

DINOv3 is Meta AI’s third-generation self-supervised vision foundation model. It produces high-quality dense and global image features that serve as a universal, frozen backbone for a wide range of vision tasks—like classification, detection, segmentation, and depth estimation—without needing task-specific fine-tuning.

https://github.com/facebookresearch/dinov3

SpatialLM

SpatialLM is a large language model fine-tuned to interpret 3D point cloud data and output structured descriptions of indoor scenes—identifying elements like walls, doors, windows, and object bounding boxes with semantic labels. It can handle diverse input types such as monocular video reconstructions, RGB-D captures, or LiDAR scans, making it useful for tasks like layout extraction, robotics, autonomous navigation, and augmented reality. The output is in a human-readable, script-like format that can be easily processed or extended.

https://github.com/manycore-research/SpatialLM

LangExtract

LangExtract is an open-source Python library from Google that uses large language models (LLMs) to turn unstructured text into structured, trustworthy data. It ensures outputs are grounded—traceable back to the exact source text—and enforces output schemas via few-shot instructions. It’s optimized to handle long documents efficiently through chunking and parallel processing, and even generates interactive HTML visualizations for review and auditing.

https://github.com/google/langextract

Embedding Atlas

https://github.com/apple/embedding-atlas/raw/main/packages/docs/assets/embedding-atlas-dark.png

Embedding Atlas is an open-source tool from Apple that provides interactive, real-time visualizations for large embedding datasets—letting you explore, search, and cross-filter embeddings alongside metadata.

  • Offers automatic clustering & labeling, density contours, nearest-neighbor search, and smooth WebGPU-based rendering (fallback to WebGL-2).
  • Available as a Python CLI, Jupyter widget, Streamlit component, and an npm/React/Svelte UI library for embedding in custom applications.

https://github.com/apple/embedding-atlas
https://apple.github.io/embedding-atlas/

POML

https://camo.githubusercontent.com/91c3ae3df86c06d8063c50e7672eda30c45e0eda1faa1dcc404c2271af82b0b6/68747470733a2f2f69332e7974696d672e636f6d2f76692f623957446346734b69786f2f6d617872657364656661756c742e6a7067

POML (Prompt Orchestration Markup Language) is an open-source, HTML-style markup language created by Microsoft to bring structure and modularity to prompt engineering for large language models.

  • Lets you define prompts with semantic tags like <role>, <task>, <example>, making them more organized and reusable.
  • Supports embedding rich data—documents, tables, images—and includes templating logic like variables, loops, and conditionals.
  • Comes with a developer toolkit: a VS Code extension for syntax highlighting, previews, and diagnostics, plus SDKs for Python and Node.js.

https://github.com/microsoft/poml

Ubicloud

https://cdn.prod.website-files.com/64f9d9b4e737e7b37d4e39a4/64fe47713ba4d520dd418c4c_hero%20image.png

Ubicloud is an open-source alternative to major cloud providers like AWS—offering core IaaS services such as elastic compute, block storage, firewalls, load balancing, managed PostgreSQL, Kubernetes orchestration, AI inference, and IAM.

  • You can deploy it yourself on bare-metal hosts (e.g., Hetzner, Leaseweb, AWS Bare Metal), or use Ubicloud’s managed offering for quick setup and lower cost.
  • It provides a control plane (built in Ruby with a Postgres backend) to manage data plane services using technologies like Cloud Hypervisor, IPsec, SPDK, and ABAC for access control.

https://github.com/ubicloud/ubicloud

https://ubicloud.com/

Dawarich

https://github.com/Freika/dawarich/raw/master/screenshots/map.jpeg

Dawarich is a self-hostable web application that replaces Google Timeline (Location History), giving you full control over your location data.

  • Visualize your location history on interactive maps, view heatmaps, travel routes, and travel statistics like distance, cities, and countries visited. :contentReference[oaicite:1]{index=1}
  • Import location data from sources like Google Maps Timeline, OwnTracks, GPX/GeoJSON files, and photo EXIF geotags; export data as GeoJSON or GPX. :contentReference[oaicite:2]{index=2}

https://github.com/Freika/dawarich

Microsandbox

https://github.com/microsandbox/microsandbox/raw/main/assets/microsandbox-banner-xl-dark.png

Microsandbox is a self-hosted platform that runs untrusted AI-generated or user code in lightweight, hardware-isolated micro-virtual machines (microVMs), offering the speed of containers with the security of VMs.

  • Provides true VM-level isolation using libkrun, with startup times typically under 200 milliseconds.
  • Includes a CLI and SDKs (Python, JavaScript, Rust) to manage sandbox environments via a project-based workflow using a Sandboxfile, with persistent states across sessions .
  • Ideal for safely executing untrusted or AI-generated code (e.g., for development, data analysis, web automation), while giving you full control over infrastructure since it’s self-hosted and licensed under Apache 2.0.

https://github.com/microsandbox/microsandbox

ArchiveBox

https://camo.githubusercontent.com/e5460e7da84bb8afb537fc63ae4a984fd33f8065769ef4d54bceaadf67021da3/68747470733a2f2f61726368697665626f782e696f2f69636f6e2e706e67

ArchiveBox is a powerful self-hosted internet archiving solution for saving and preserving web pages.
It can ingest bookmarks, RSS feeds, or entire browsing histories and archive them in multiple formats (HTML, PDF, screenshots, WARC, etc.).

  • Lets you build your own “Wayback Machine” with full-text search.
  • Supports scheduled imports from browsers, services, and APIs.
  • Stores data locally for privacy and long-term control.

https://github.com/ArchiveBox/ArchiveBox

Cap

https://github.com/tiagorangel1/cap/raw/main/assets/banner.png

Cap is a lightweight, modern open-source CAPTCHA alternative that uses SHA-256 proof-of-work instead of tracking or puzzles—making it fast, private, and easy to integrate.

  • The frontend @cap.js/widget is ultra-compact (~12 KB minified & Brotli-compressed), running in web workers and optionally WASM-accelerated.
  • A simple, zero-dependency @cap.js/server handles challenge generation and validation, and there’s a standalone Docker mode plus CLI and solver tools for robust flexibility.
  • Fully self-hostable, highly customizable, and privacy-focused—no data collection or fingerprinting required.

https://github.com/tiagorangel1/cap

Sniffnet

https://raw.githubusercontent.com/GyulyVGC/sniffnet/main/resources/repository/pages/overview.png

Sniffnet is a cross-platform, Rust-based network monitoring app that lets you comfortably monitor your internet traffic with privacy and ease.

  • Offers real-time charts, filtering, PCAP import/export, service/protocol detection, domain/ASN lookup, geolocation, notifications, theming, and more.
  • Built with usability in mind, it’s free, open-source (MIT/Apache-2.0), translated into many languages, and designed for secure performance.
  • Actively developed and widely adopted—recent major release v1.4.0 (June 27, 2025), and upcoming features like per-process connection identification are in the works.

https://github.com/GyulyVGC/sniffnet

A-Frame

A-Frame Example

A-Frame is an open-source web framework built on top of three.js that lets you create VR, AR, and 3D experiences using simple, declarative HTML tags.

  • Uses an entity-component architecture so you can compose 3D scenes with familiar HTML (e.g., <a-scene>, <a-box>).
  • Works across platforms—desktop, mobile, and WebXR-capable headsets—and is lightweight, performant, and easy to get started with.

https://github.com/aframevr/aframe
https://aframe.io/

ConvertX

https://github.com/C4illin/ConvertX/raw/main/images/logo.png

ConvertX is a self-hosted online file converter built with TypeScript, Bun, and Elysia. It supports over 1,000 file formats, letting you convert documents, images, audio, video, 3D assets, and more—right on your own infrastructure.

  • Supports batch conversions, password-protected access, and multi-account management.
  • Leverages tools like FFmpeg, Pandoc, GraphicsMagick, Inkscape, and others to offer wide-ranging format support.
  • Easily deployable via Docker (e.g., with docker-compose) for quick setup and integration.

https://github.com/C4illin/ConvertX

pad.ws

https://github.com/coderamp-labs/pad.ws/raw/main/docs/canvas_ide.png

pad.ws is a browser-based “whiteboard IDE” that blends visual sketching and coding in one seamless workspace.

  • Offers an infinite drawing canvas (via Excalidraw) alongside a full-featured IDE—you can open terminals or VS Code right beside your sketches.
  • Supports real-time collaboration (think Figma-style multiplayer cursors and live syncing), making it ideal for brainstorming, teaching, or collaborative architecture/design work.
  • Accessible both via a hosted version (pad.ws) and self-hostable via Docker, with persistence via PostgreSQL/Redis and optional Keycloak for user management.

https://github.com/coderamp-labs/pad.ws
https://pad.ws

FossFLOW

FossFLOW is a free, open-source Progressive Web App (PWA) for creating stunning isometric infrastructure diagrams—think 3D-style visuals but in your browser.

  • Built with React and TypeScript (via the Isoflow library), it runs entirely client-side with offline support and auto-saving.
  • Offers drag-and-drop components, extensible icons (including AWS, Azure, GCP, or your own), JSON import/export, and optional Docker deployment for persistent server-side storage.

https://github.com/stan-smith/FossFLOW

Nano Banana (Gemini 2.5 Flash Image)

Nano Banana is Google’s latest image generation and editing model, now part of its Gemini app.

  • Lets you make multi-step edits—like changing backgrounds, blending photos, or styling objects—while preserving the look of people, pets, or other key details.
  • Available to both free and premium users globally, all edits include a visible watermark plus an invisible SynthID tag.

https://blog.google/intl/en-mena/product-updates/explore-get-answers/nano-banana-image-editing-in-gemini-just-got-a-major-upgrade/