GitHub Trending Today #54: scout, freellmapi, AnyPS5, desigma, yobiru, sgrep, indra, BRUTAL, armada Transcript, AI Summary & Key Points

agentplay is an open-source command-line tool and local web viewer maintained by Entelligence AI for replaying coding-agent sessions as a map and timeline. It discovers session data from Claude Code, Codex, Cursor, GitHub Copilot, OpenCode, and Kiro, parses transcripts or local database stores into redacted agentplay/v1 JSON maps, and serves them from a local viewer. The session map records files and repositories touched, shell verbs, tools, MCP servers, models, subagents, errors, retries, delays, outputs, token counts, and timing without storing prompt or assistant text, tool inputs or outputs, file contents, or command arguments. The viewer provides a radial session map, playback with step or wall-clock time, error and retry navigation, task and subagent scopes, and import/export of map files. It runs locally without an account or API key; session data remains on the device, with an optional email-signup network request that can be disabled. The CLI requires Python 3.11 or newer, has no runtime dependencies beyond the standard library, and is distributed under the Apache License 2.0.

Anima is a Claude Code skill that turns biology prompts into interactive, sourced, isometric figures delivered as standalone HTML pages. It collects sources before drawing, uses an isometric drawing kit with biology primitives, builds a mechanism that can be toggled or otherwise manipulated, and uses one small model to drive both the animation and a live readout. Figures cite every displayed number, distinguish established findings from leading models and animal from human evidence, and include hover explanations, source links, simplification notes, and an educational-use disclaimer. The repository includes example figures covering subjects such as dopamine receptors, stimulants, reward circuits, semaglutide and GLP-1, estrogen, and myelinated neurons, along with scripts for building, capturing, checking, and recording figures. It can be installed with `npx skills add Garinmckayl/anima -g -a claude-code`, requires Claude Code and Node.js 20.10 or later, uses no runtime dependencies, and is released under the MIT license.

armada is a command-line tool and TypeScript/Python SDK for running shell commands or functions over many inputs concurrently on Cloudflare Containers in the user's own account. It prepares each execution environment once, snapshots it, and has containers repeatedly pull tasks from a shared queue; results stream back as they finish, while large outputs, logs, and artifacts are stored in R2. The deployment consists of a Worker, Durable Objects, and an R2 bucket, and supports retries, weighted tasks, caching, resource-aware packing, multi-container gangs, and optional CPU-side reinforcement-learning workloads such as grading generated code. It also provides matrix-based CI runs, GitHub webhooks, verdict reports, a dashboard, secrets, and separate local execution; Cloudflare's Workers Paid plan is required.

BRUTAL is an open-source C# framework for making games or building a separate game engine. It provides a custom interop layer for direct access to native APIs, along with low-level libraries for unmanaged memory, collections, strings, mathematics, concurrency, logging, localization, and bindings for graphics, windowing, audio, networking, physics, and other native components. BRUTAL deliberately has no editor, scene graph, renderer, or general-purpose engine architecture. Developers assemble their own stack from separate NuGet packages, such as packages for Vulkan, GLFW, Dear ImGui, FMOD, RakNet, Jolt, and Box2D; the framework uses unmanaged memory, unsafe code, and allocation-avoiding components where appropriate. The repository contains the source for its published packages. BRUTAL is developed by Ahwoo Limited and is free to use, modify, and distribute, including in commercial games, under a licence based on MIT with attribution requirements.

cc-stretch is a Claude Code plugin that presents replies as stretchtext: responses open in a short form, and clicking a linked phrase expands that phrase in place while keeping previously read words stationary. Claude is told which stretches were opened, and stretches can nest; during streaming, expanded content remains hidden until selected. The plugin also folds completed groups of Claude Code tool calls into a one-line summary while leaving running and errored calls visible. It requires Claude Code's fullscreen renderer in the terminal or the Claude desktop app, and it uses Claude's function-hook plugin API. Claude generates all expansion levels up front, so the plugin reduces the amount read rather than the token cost; unsupported interfaces display the markup fully open, and Windows clicking is reported as not yet supported.

Claudemy is an open-source Claude Code skill that turns a topic into a short audio course. It maps the topic into subtopics, researches roughly 20–40 web pages while limiting sources from any one site, compares cited notes for agreement and disagreement, and plans the material in lesson arcs. It then creates about 12 two-minute lessons with two-host audio, a lesson card, one small exercise, source links, a playbook, and a lesson plan; a sample lesson is produced first for review before the rest is written and recorded. Audio can be generated with ElevenLabs or, on macOS, built-in voices. The skill requires Claude Code with web search, Python with BeautifulSoup, and ffmpeg; pandoc is optional for Word copies of the lesson cards. It is free and MIT-licensed, with ElevenLabs credits and the user's Claude plan as possible costs.

Clint Gecko is an open-source Android browser based on Firefox's GeckoView engine. It supports installing Firefox extensions from the add-on store or XPI files, background audio and video playback, reader mode, incognito browsing, and separate profiles with independent cookies, logins, and site data. Its built-in Advanced Download Manager provides multi-threaded and multi-part downloads, pausing and resuming, scheduling, speed limits, and custom download locations. A media-capture feature detects video, audio, and subtitle streams, including HLS/M3U8 and DASH streams, and can download detected media. The app is built with Kotlin, Jetpack Compose, and Material 3. The project says it opts out of GeckoView and Mozilla telemetry and has no backend, analytics, or data-collection servers. It requires Android 8.0 or later, is distributed through APK releases, and its source code is licensed under GPLv3.

decomp.wtf is a locally running AI-assisted game decompilation tool for games the user owns. It accepts ROMs, disc images, or program files; identifies functions and data with tools such as splat, spimdisasm, decomp-toolkit, and ds-decomp; and gives each function to an AI agent with its assembly, related data, call context, look-alike code, and a draft implementation. Each agent attempt is compiled with the game's original or project-specific compiler, linked at the function's original address, and compared byte for byte with the source game. Mismatches return an instruction diff for another attempt, while matches are saved and shared with other agents. The tool also supports parallel helper agents, background permutation of near misses, public decompilation projects, rebuildable exports, and community rooms where participants work from their own copies of the game. It supports Linux and Windows through a WSL 2 distribution; macOS is not supported. GameCube, Wii, and Nintendo 64 are the recommended platforms, while PlayStation, PlayStation 2, Game Boy Advance, and Nintendo DS support is described as beta. The application runs locally without an account, telemetry, or uploading game data, and is released under the MIT license.

desigma is a free, offline, open-source interface-design and whiteboard application written in Rust, with an egui desktop editor and a deterministic CPU renderer based on tiny-skia. It supports frames, vector editing, auto layout, components and variants, variables, prototyping flows, presentations, shared libraries, and whiteboard boards with stickies, shapes, and connectors. It imports Figma REST JSON and .fig files, including components, styles, overrides, images, and prototype interactions, and exports PNG, JPG, SVG, Figma REST JSON, CSS, and code for Tailwind, SwiftUI, Jetpack Compose, and React. The repository also includes an optional local MCP server for inspecting and editing design files and generating code. desigma is in early alpha, with stated gaps including collaboration, plugins, some .fig features, and code-signed releases. It is licensed under MIT or Apache-2.0.

DEsk is a beta Wayland desktop environment for Linux built on wlroots. It presents the desktop as a realistic skeuomorphic desk: application windows appear as paper sheets that can be folded or discarded, the dock is a drawer, applications are organized in binders on a bookshelf, notifications arrive in a post box, and system controls are built into the desk. It includes a compositor, session manager, launcher, notification centre, lock screen, settings application, calendar, memo pad, multi-monitor support, Xwayland support, and standard desktop integrations. The room is ray-traced with GLES 3 on the GPU when available, with a CPU renderer and a 2D fallback for systems without suitable graphics support. DEsk can be installed as a login session on Debian, Ubuntu, Arch Linux, Fedora and derivatives, or built with Meson; its installers can also install the recommended desktop programs. The project is distributed under the MIT license and is described as an early release with expected rough edges and ongoing changes.

diff-tour is a read-only terminal user interface (TUI) that turns a Git change into a guided code tour for understanding changes such as features, refactors, and AI-written code. It uses Claude Code, Codex CLI, Gemini CLI, OpenCode, or Copilot CLI to read the repository's commits, diff, and relevant files, then write a tour that links changed lines to surrounding code, explains design judgments, and pauses for questions about failure cases, repeated use, rollout, inputs, scale, and tests. The tour can show code and diffs from the appropriate Git commit, example code, flow and timeline diagrams, database schemas, and example data; follow-up questions can be asked from the terminal. Before displaying a tour, diff-tour checks the generated JSON's file and line references and asks the selected LLM to correct invalid references up to two times. The LLM operates on an exported repository snapshot with read-only permissions; diff-tour itself makes no network requests and collects no telemetry. Tours are saved locally for later viewing, and the project is distributed under the MIT license.

FreeLLMAPI is an open-source proxy that aggregates free tiers from multiple LLM providers behind a single OpenAI-compatible `/v1` endpoint. It can also connect custom OpenAI-compatible chat, embedding, image, and audio endpoints, including local servers and remote gateways. Its router selects an available model, fails over when a provider is rate-limited, stores provider keys encrypted, and tracks per-key usage against free-tier limits. The router refreshes its model catalog from a signed feed; free installations receive periodic snapshots, while the hosted premium offering provides the live catalog. The project is intended for personal experimentation and is available as a self-hosted install with a hosted service at freellmapi.co.

Glide is an open-source, native Rust screen recorder and editor for macOS that produces polished product videos. It records displays, windows, selected areas, and iPhone or iPad screens over USB, with separate system-audio and microphone tracks and optional camera capture. Its editor logs input and re-renders the cursor at high resolution, automatically targets click clusters with spring-animated zooms and edge snapping, and supports manual or mouse-follow zooms. It provides a timeline for clips, zooms, masks, camera layouts, speed changes, and captions generated on-device through Whisper or Apple Speech; styling includes generated backgrounds, blur, shadows, rounded corners, padding, aspect ratios, and device frames. Projects export as MP4 through VideoToolbox or as GIF, and the command-line interface can record, export, inspect devices, transcribe, check masks, and generate synthetic projects. Glide requires macOS 15 or later and uses Rust, Xcode, CMake, ScreenCaptureKit, AVFoundation, wgpu, and VideoToolbox. The repository states that it was reverse-engineered from Screen Studio, is unaffiliated with Screen Studio, and contains none of its code or assets. It is distributed under the MIT OR Apache-2.0 license.

Gloss is an open-source AI reader for articles and PDFs. A reader selects a line, asks a question, and receives a cited answer in the margin beside the source; the video also describes keeping notes alongside passages and exporting selections to Markdown for Obsidian. The app uses Together AI for LLM inference, with DeepSeek V4.1 Flash and Kimi K3 listed for language-model tasks. Exa provides optional web search, while LiteParse handles in-browser PDF parsing and Qwen on Together processes scanned pages. Optional Clerk authentication and Neon database syncing support a shared reading library.

GPT with Taste is a behavioral recipe and system-instruction package for GPT-based coding harnesses, developed by NB-Corp. It provides a single English authoring recipe that compiles into host-specific instructions and event steering for OpenAI Codex, Claude Code, Kiki, Pi, Oh My Pi, and DeepSeek Harness. The instructions promote sensible defaults, continued execution, synthesis of delegated-agent results, and attention to engineering and visual details. For Codex, the installer writes managed prompt files and can add hooks that refresh scoped reminders when users submit work, child agents start, or supported wait results arrive. It supports a prompt-only installation or a Node.js-based installation with hooks, requires no npm packages, and leaves the model, provider, authentication, permissions, MCP servers, roles, and AGENTS.md unchanged. The repository includes harness porting, authoring, recovery, research, and comparison documentation and is released under the MIT License.

Graphicle is an open-source code graph and project-memory tool for coding agents, distributed as a single Rust program by KarLex AI. It parses repositories with tree-sitter to map files, functions, classes, calls, and imports; indexes project documents, decisions, procedures, and captured agent sessions; and returns cited excerpts with exact files and lines through its CLI and MCP server. Reviewed memories are linked to the functions they describe and become stale when those functions change, while synchronization keeps the graph current after edits and Git operations. The setup command connects supported MCP clients such as Claude Code, Cursor, Codex, VS Code, and Claude Desktop, and the app provides a browser interface for browsing sources and approving memories. It runs locally without an account, API key, telemetry, Python, Docker, or a database server for most features, stores data in SQLite, and is licensed under Apache-2.0.

Handwired is a JavaScript tool that transforms ordinary programs containing conditionals, loops, and state into differentiable neural-network-style computations. It rewrites control flow into arithmetic using multiplication, addition, and ReLU, then collects those operations into a large matrix multiplication; the resulting program is designed to produce the same outputs without training. The computation can run in JavaScript or WebGPU, and the manually written weights remain differentiable for experiments such as training them or connecting the frozen program to another neural network. The repository includes live demos and can be run locally with Bun.

House Control is an open-source smart-home app for macOS and iPhone that renders a home as a live 3D model in the Mac menu bar, around the MacBook notch, on the desktop, and in iPhone widgets. A JSON floor plan or the visual editor defines rooms, walls, doors, windows, furniture, lamps, and air conditioners; each device can be linked to a real device, with controls for lighting, brightness, color temperature, scenes, power, target temperature, operating mode, and fan speed. It connects Smart Life/Tuya, Home Assistant, Philips Hue, and Apple Home in one interface. The Mac and iPhone apps use native Swift shells around a shared Three.js web UI and Swift package, while optional pairing synchronizes plans and connections directly over the local network using Bonjour and TLS rather than a server. Devices can also be simulated with bundled demo plans and testing tools. House Control includes an optional AI Access gateway that exposes selected home controls and plan-editing tools through HTTP/JSON, OpenAPI, or MCP for assistants such as ChatGPT and Claude. The gateway runs on the Mac, binds locally, and can use a temporary Cloudflare tunnel; it can be stopped or automatically expired. The Mac app supports macOS 13 or newer and is distributed as an ad-hoc-signed DMG; the iPhone app requires iOS 17 or newer and is not yet on the App Store according to the repository. The project is licensed under MIT.

Indra OCR is a CPU-oriented document OCR and parsing tool developed by Datalab. It converts PDFs and images into Markdown, plain text, or JSON containing recognized lines, bounding boxes, layout blocks, labels, reading order, confidences, rotation, and skew information. It can use a digital PDF's text layer where it passes checks and OCR only missing content, or OCR every page. The pipeline uses ONNX Runtime models without PyTorch to detect lines, recognize text, classify blocks such as headings, lists, tables, captions, footnotes, headers, and forms, and order them for reading. Its local viewer displays page overlays, Markdown, source lines, JSON, confidences, and per-stage timings. An optional math mode recognizes mathematical lines and emits inline LaTeX. The Python package and CLI are distributed under Apache 2.0, while the model weights use a modified OpenRAIL-M license with additional commercial-use conditions. The models are downloaded from Hugging Face on first use; the repository reports a model size of about 18 MB. OCR support covers Latin, Cyrillic, Greek, Devanagari, Bengali, and Chinese scripts, while other scripts can be processed through digital PDF text layers. The project documents limitations for handwriting, degraded scans, complex tables, vertical text, and some scripts in scanned images.

Libre Calculator is a free, open-source collection of web and desktop calculators for finance, construction, cooking, physics, education, health, geometry, statistics, unit conversion, and other domains. The web version uses Astro and React, while the desktop app uses Tauri; calculators are organized into optional packs, and each calculator has its own web page with offline caching for active packs. Each result is generated from the user's inputs and includes the calculation steps, a diagram, and a short lesson explaining the formula. Its calculation engines are pure functions with written numeric contracts and fixture tests; arithmetic uses 50-digit decimals and rounds only for display, while invalid input produces an explicit failure instead of a guessed result. The desktop app provides a quick panel opened with Cmd/Ctrl+Shift+C. The project is distributed as open-source software through its website and GitHub repository, with desktop builds for macOS, Windows, and Linux.

LumaBrowser is a free, open-source desktop browser and local AI application maintained by Lumabyte, LLC. It combines a Chromium-based browser with GPU-aware local LLM runtimes, chat and agents, image, video, music and voice generation, browser automation, and an optional local API. The application reads the computer's hardware and estimates which models fit before download, selecting full, partial, or CPU offload and supporting model sources and runtimes such as llama.cpp, MLX, Ollama, LM Studio, and Hugging Face caches. Its agents can browse in separate tabs, read pages, use files and images, run document retrieval, and connect to external MCP servers. Claude Code, Codex, Cline, OpenCode, Selenium, Puppeteer, and Playwright can control it through MCP, WebDriver, or CDP, while a CLI and IDE plugins provide coding-agent access. It runs on Windows, Linux, and Apple Silicon macOS, with a Docker mode for headless server use. The software is licensed under AGPL-3.0-or-later; official unmodified builds can be used without a commercial license, while private modifications, proprietary combinations, or private extensions require licensing from Lumabyte, LLC.

Machine Skill is an open-source Agent Skill for Claude Code that turns a one-line description of a machine, mechanism, engine, instrument, or device into an interactive 3D simulation in a single HTML file. It researches the real machine, generates procedural three.js geometry with real proportions and named parts, assigns materials, and uses a small machine-specific physics solver—such as a Brayton cycle, rocket chamber and nozzle model, pump curve, or torsional oscillator—to drive motions and displayed values. The resulting pages support drag interaction, explode views, solid-face cutaways, X-ray mode, live labels, slow motion, and a computed “break it” scenario; the repository states that its physics is an explanatory model rather than CFD or FEA and that geometry is proportionally accurate rather than manufacturing-tolerance accurate. The skill also provides a deterministic browser film mode and Node-based scripts for recording simulations to video or stills, including optional music, sizing, supersampling, motion blur, and frame validation. Its runtime is an inline ES module based on three.js, with templates, solver references, example machines, and scripts for screenshots, recording, pink-frame detection, and inlining. It can be installed as a Claude Code skill with the Skills CLI or Git, and its pages run in browsers with WebGL 2; recording requires Node, Playwright, and ffmpeg. The repository is licensed under MIT.

Pullboard is a local-first coordination tool for developing software with coding agents. It keeps the project’s approved requirements in SPEC.md, its working rules in DOCTRINE.md, and its activity in a board stored inside the Git repository, so decisions and progress remain recorded without an account or hosted service. Agents work on items in separate Git worktrees and lanes. Each item has a requirement and a gate; a submit runs the configured test command, and a different agent then verifies the exact submitted commit against the requirement, accepting it with evidence or rejecting it with a reason. Hooks enforce lane ownership and other machine-checkable rules, while the coordinator routes questions and only escalates decisions to the human. The package provides a CLI and a local view for approving requirements, answering questions, monitoring activity, managing lanes, and exporting a replayable board snapshot. It requires Git and Node.js 22.13 or newer, supports configurable gates such as npm test, pytest, and go test ./..., and is released under the MIT License. The project describes worktree identity as a same-machine coordination mechanism rather than a security boundary, and says Windows is not yet tested.

Remory is an open-source system for adding learned residual memory to large-language-model agent context compaction. It supplements an agent’s generated summary with soft memory conditioned on that summary, allowing a frozen model to continue work after earlier conversation history is discarded. During compaction, its hook reads the actor model’s hidden states and runs a memory network; during continuation, the saved soft tokens are inserted after the summary. The repository provides SGLang and Transformers deployment, model checkpoints, a Python Harness and Client API, a Codex integration, and training procedures for reconstruction and summary-conditioned residual-memory networks. It is released under the MIT License.

ResiMind is an open-source neuro-symbolic agent framework with independently verified memory. A proposer—such as a model, rule, or symbolic tool—submits a step with evidence references; domain verifiers accept, reject, defer, or interrupt it, and rejected steps become residual feedback for the next attempt. Accepted facts are committed and remaining obligations are rebuilt. Completed derivations can produce candidate rules that undergo independent admission, persistence, re-verification on reload, and verification again on each future application, allowing checked knowledge to be reused without changing model weights. The project includes bounded mathematics and algebra workflows, MCP tools, an algebra repair Skill, optional Lean proof feedback, and adapters for planning, customer support, constrained optimization, bridge analysis, linear equations, and axial bars. Supported math routes can run offline without model calls, while optional DeepSeek, OpenAI, and LangGraph integrations provide model callbacks or an outer connector. The core requires Python 3.10 or later and has no runtime dependencies; optional MCP support requires Python 3.11 or later. It is distributed under the MIT License.

Scout is an open-source TypeScript browser agent made by Outcrawl. It accepts a task in plain language, drives a real Chrome or Chromium browser over the DevTools Protocol, reads pages, clicks, types, scrolls, checks the results, and returns an answer or a structured response. Its runtime repeats the observe–choose action–act–check loop, while separate optional model jobs can handle planning (brain), screenshot interpretation (eyes), and answer extraction (extract); the required hands model performs page reading and interaction. Scout can use hosted or self-hosted models through chat-completions-compatible endpoints, run in Outcrawl's cloud or on a local or already-running browser, and return status, steps, errors, costs, and a cloud replay URL. The repository contains the Scout API, runtime, model layer, shared types, CDP browser driver, and agenthands input planner. It is licensed under Apache-2.0, with the agenthands package under its own MIT license.

sgrep is a Rust command-line tool from Context.dev for searching local code and documents with plain-English queries. It combines ripgrep file discovery with lexical BM25 matching and semantic matching from small static embedding models, returning source passages with file paths and line ranges rather than generated answers. It can also rank ripgrep results supplied through standard input or JSON. The tool splits source into passages using Chonkie and Tree-sitter, using syntax-aware chunks for supported programming languages and text chunks for prose. Its code and text modes use local Potion Code and Potion Base models; embeddings are created by averaging fixed token vectors into 256-number representations, then combining semantic and lexical scores with exact identifier matches prioritized. The first search downloads the models and required parsers, after which cached searches can run offline without an account, hosted index, model server, or per-query API call. It requires Rust and ripgrep and is released under the MIT license.

smlibpy is a Python library for 3D solid and surface modeling on the SMLib NURBS kernel. Its value-based API follows build123d's algebra mode: primitives and sketches produce parts, operations return new parts, and Boolean operations can be written with +, - and &. Sketches can be extruded, revolved, lofted or swept, while features include holes, pockets, fillets, chamfers, shelling, sheet-metal bends, deformation and draft operations. Parts use exact B-rep surfaces, including planes, cylinders, cones, spheres, tori and NURBS; volumes, areas and mass properties are computed from those surfaces, and the library supports geometry-aware selection of faces and edges. It reads and writes formats including STEP, USD BrepArray bodies, DXF, STL, glTF/GLB and OBJ, and provides rendering, technical drawings, sections and model inspection. The package includes an MCP server for agent-driven modeling, with function guides, scripted modeling steps, snapshots and validation loops. The same API is available in the browser and Node through the related smlibwasm WebAssembly package. smlibpy is distributed through pip with compiled wheels for common platforms and Python 3.10–3.14; the base package uses NumPy as its stated dependency. The project is MIT licensed and bundles the SMLib kernel from NVIDIA-Omniverse/usd-brep under the Apache License 2.0.

SwiftVM is a native macOS application for running macOS, Linux, and Windows 11 for Arm virtual machines on Apple silicon. It automates guest setup, including macOS installation, Linux cloud-image and desktop installation, and Windows 11 installation with drivers; setup scripts can configure additional software and machine state. It supports SSH access, shared folders and clipboard, port forwarding, USB devices, disk images, snapshots, suspend and resume, and APFS-backed duplicates that share unchanged disk blocks. Machines can be saved as templates for creating new virtual machines. The accompanying `swiftvm` command-line tool can run guest commands, copy files, view the guest screen, and operate its mouse and keyboard for scripts and AI-agent automation. The project is distributed for macOS 27 or newer on Apple silicon under the Apache-2.0 license.

terminal-rs is a native Rust, keyboard-driven market terminal with a Bloomberg-style tiled workspace and GPU-rendered charts for stocks, ETFs, indices, foreign exchange, crypto, Treasuries, futures, commodities, and options. Its chart renderer uses level-of-detail bar pyramids and generates geometry in the vertex shader; it supports multiple chart types, indicators, drawings, comparisons, live crypto feeds, cached history, and offline sample data. The terminal ingests Senate eFD and House Clerk STOCK Act disclosures, normalizes politicians, transactions, dates, tickers, and amount ranges into SQLite, and provides congressional-trade feeds, politician profiles, chart markers, tracking alerts, and per-security queries. Panels can be docked, split, linked by security, popped into native windows, and saved as layouts; themes and keyboard shortcuts are configurable. Brokerage accounts connect through SnapTrade and are read-only by default. A local paper-trading account is included, while live order entry is opt-in and uses SnapTrade previews and confirmation. The repository states that macOS is the primary target, with Linux support, and that scanned filings are recorded but not OCR'd or itemized.

VyManager is an open-source, multi-tenant control plane for managing VyOS routers across multiple sites from a web interface. It organizes routers into sites, supports OWNER, ADMIN, and VIEWER permissions per site, and provides configuration deployment, live interface counters, system information, network graphs, and WireGuard peer monitoring through the VyOS REST and GraphQL APIs. The platform uses a version-aware backend for VyOS 1.4, 1.5, and rolling releases. Its three-layer architecture separates router API endpoints, batch-operation builders, and version-specific command mappers; capability endpoints let the frontend show features supported by the connected VyOS version. The frontend uses Next.js and the backend uses FastAPI with PostgreSQL, and deployments run through Docker Compose or an on-box VyOS installation. VyManager stores SSH credentials encrypted, uses API-key authentication, and supports HTTPS, custom CA certificates, WebSocket-based monitoring and console access, and configurable API timeouts. Its related VyMCP integration exposes the VyManager API to AI agents with inherited authentication, role-based access control, audit logging, token scoping, and an opt-in write mode.

wifi-sensing-kit is an open-source local Wi-Fi sensing kit that uses three ESP32-S3 boards to recognize activities such as sitting, squatting, walking, jumping, or an empty room from changes in Wi-Fi channel state information (CSI). During training, a webcam uses a pose model to label movements; the resulting movement model can then classify activity from Wi-Fi while the camera is covered. The kit provides dashboards for live CSI signals, a room map showing sensing links between boards and the router, and movement-model training, testing, and evaluation. It is built on RuView's open-source CSI firmware, with a patch that gives each board a distinct light color and lets boards measure one another as well as the router. Processing is local: webcam video stays in the browser and only movement labels are sent to the local server. The system must be retrained when the person, room, or board positions change, and it does not produce a skeleton or exact position. The repository includes firmware build and flashing scripts, a local Python server, configuration files, and evaluation tools; it requires three ESP32-S3 boards with at least 8 MB of flash, a 2.4 GHz Wi-Fi network, and a macOS or Linux computer for setup. The kit's code is MIT-licensed, while its bundled firmware patch, pose model, and fonts retain their stated upstream licenses.

Yobiru is a communication layer for connecting personal AI agents across computers and cloud workspaces over Tailscale. Each participating machine runs a receiver that authenticates the connecting Tailscale machine with `tailscale whois`, applies local access rules, and stores accepted messages in a local inbox. Agents can send text, JSON, reports, logs, and files; reply to messages; and create tracked requests with receipts, progress states, final outcomes, deadlines, and timeout status. Attachments are fetched from the sender over Tailscale and verified by size and SHA-256. Wake requests are handled by a receiver-configured local executable or webhook, with sender and agent limits, quiet hours, and rate limits; the receiving device makes the final wake-up decision. Yobiru provides a Python standard-library CLI and plugin package for supported agent environments, but it does not provide a hosted messaging service, shared chat-room interface, scheduler, autonomous task manager, or exactly-once task execution. It requires Python 3.8 or later, an installed and reachable Tailscale network, and a persistent receiver for receiving messages and files. The project is maintained under Chekusu and released under the MIT License.