GitHub Trending Weekly #43: Cloudflare Computer, waku-agent, querysplat, findphone, Soup, morphicons Transcript, AI Summary & Key Points

anydoc is an open-source Rust library from Firecrawl that converts Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV, and PDF files into GitHub-Flavored Markdown. It uses a shared document model and provides Node.js, Python, browser WebAssembly, and Rust interfaces, along with a command-line interface that reads files or standard input and writes Markdown to standard output or a file. The library can return either Markdown or its intermediate document model, which carries embedded assets. A browser demo runs the WebAssembly build locally, while hosted OCR through Firecrawl Parse can process scanned PDF pages that anydoc cannot read by itself. It is also distributed as an agent skill compatible with agent tools such as Claude Code, Codex, Cursor, and OpenCode.

Backchannel is an open-source userscript and renamed continuation of HNewhere that brings discussions from Hacker News, Bluesky, Reddit, Lemmy, Lobsters, and other selectable sources into a sidebar beside the page or PDF being read. It checks visited pages for matching threads, merges enabled sources into one conversation with source labels, and matches quoted comments to article passages as clickable annotations; enhanced PDF highlighting is supported except in Firefox. Users can vote, reply, and submit through popups on supported source sites, add locally stored notes to pages and PDFs, and use a customizable front page combining content from selected sources. The single-file script runs without a build step, backend, analytics, or telemetry, and is distributed through userscript managers such as Tampermonkey, Violentmonkey, and Userscripts.

Bindwidth is a browser-based calculator for sizing on-premises LLM inference infrastructure and estimating total cost of ownership. It evaluates three constraints for a resident text model—KV-cache memory, combined prefill and decode serving capacity, and the runtime session ceiling—and identifies which constraint binds for a workload. The tool accounts for interactive users and autonomous agents as different load types, then compares owned hardware, eligible sovereign rentals, and enterprise subscriptions with per-token APIs for agents. It distinguishes measured from estimated hardware and model data through confidence levels and profile-specific overrides, flags domain and model-capability limits, and exports a decision record as Markdown or the full scenario as JSON. The repository describes it as a directional sensitivity estimator rather than a procurement quote. It is a frontend-only application with no build step, server, database, account, or telemetry; calculations run in the browser and entered data is not transmitted.

Can I Vibecode It? is an open-source catalog that evaluates whether AI coding agents such as Claude Code, Codex, or Cursor can build a usable personal replacement for a paid SaaS application. Each app entry provides a YES, KINDA, or NOT REALLY verdict, the exact prompt for attempting the replacement, and the tradeoffs of leaving the original service, including lost network effects, data, or infrastructure. App entries are contributed as one JSON file per app, with the schema and verdict criteria documented in the repository. The site uses Astro server output with a Node adapter to render fully server-rendered HTML, SQLite via better-sqlite3 for vote counters and waitlist data, and vanilla JavaScript and CSS for interactions. Satori and resvg generate Open Graph images at build time. It runs locally with npm install, npm run dev, npm run build, and npm start, requires no environment variables for local development, and supports optional analytics, sign-in, payments, email, and media-storage configuration. It can be deployed on a VPS behind a reverse proxy, with DATA_DIR used to keep user data outside the repository. The project is released under the MIT License, and its prompts are intended to remain free.

Capptivo is a free, open-source screen recorder and editor for creating demo videos on macOS, Windows, and Linux. It tracks cursor movement and clicks to apply follow-cursor and click-based automatic zooms, and provides editor presets, annotations, facecam, backgrounds, captions, and MP4, WebM, or GIF export. Captions use a system whisper.cpp command-line binary, with model weights downloaded on first use. The application captures macOS screens through ScreenCaptureKit with VideoToolbox H.264 encoding, Windows screens through Windows.Graphics.Capture with hardware-encoding support, and Linux screens through xdg-desktop-portal and PipeWire. It is distributed through platform installers, including DMG or app archives for macOS, MSI or setup executables for Windows, and DEB, AppImage, or RPM packages for Linux. The repository releases it under the MIT license.

cargo-frisk is a Rust command-line tool that checks the files a Cargo package will ship. It builds the .crate archive, compares its contents with `git ls-files`, and reports files that are shipped but not tracked, files tracked but not shipped, and the expected shipped-and-tracked files. Cargo-generated files such as `Cargo.toml.orig`, `.cargo_vcs_info.json`, and `Cargo.lock` are listed separately. It scans packaged files for provider tokens, private keys, `.env` files, credentials files, editor backups, and other patterns adapted from gitleaks, with additional rules for Cargo packaging. Findings can be adjusted using severity thresholds, entropy checks, path-based severity reductions, inline suppressions, ignored paths, ignored rules, and per-file size limits. Binary or oversized files are reported as not scanned rather than silently skipped. The tool is installed with `cargo install cargo-frisk` or `cargo binstall cargo-frisk`, then run with `cargo frisk` from a package. Its exit codes distinguish no findings, findings at or above the configured threshold, and tool errors such as a failed `cargo package` operation or malformed archive.

A Rust library for creating and reading specially crafted .tar.gz archives with embedded chapter-boundary information. A chapter contains zero or more consecutive tar entries; chapter markers are encoded as valid empty gzip blocks with distinctive Huffman code alphabets, allowing readers to skip groups of entries in constant time without decompressing intervening data and to decompress separate chapters concurrently. Archives remain readable by existing tools as ordinary tar.gz files, and archives without chapter metadata are treated as one chapter.

Claude Code is Anthropic's agentic coding tool for the terminal, IDEs, and GitHub. It uses natural-language commands to understand a codebase, create and read files, execute commands, run tests, explain code, manage Git workflows, and handle routine development tasks. It can also load persistent project context, run custom slash commands, use plugins with custom commands and agents, and operate with configurable autonomy while leaving actions such as final pull-request merging to a human. The official repository documents installation for macOS, Linux, and Windows, and identifies npm installation as deprecated.

Claude Desktop is Anthropic's desktop application for interacting with Claude, its AI assistant. It can host extensions such as Token-Saver, which sends selected passages from a PDF instead of the full document.

Cloudflare Computer is an open-source library that provides AI agents with a virtual filesystem backed by a Cloudflare Durable Object. The Durable Object stores the authoritative state in SQLite, while the workspace exposes a single execution interface through `workspace.runtime.exec(source, { backend })`. It supports three execution backends: Container projects the SQLite state into a sandbox container through a FUSE mount and synchronizes changes over capnweb RPC; Isolate shell runs `just-bash` in a Dynamic Worker; and Isolate JavaScript runs an ECMAScript module in a fresh Dynamic Worker with structured inputs and results, durable relative imports, configured libraries, workspace-backed `node:fs/promises`, and trusted `ws:git` and `ws:artifacts` modules. Workspaces can register multiple backends under stable identifiers, connect to them lazily, or be used solely as a filesystem without an execution backend. The repository includes examples for container, Worker shell, Worker JavaScript, egress policies, MCP, agent workspaces, tutorials, artifacts, and assets. The package is marked preview-only: its APIs are unstable and the repository says it is intended for experiments, exploration, and prototypes rather than production use.

A ComfyUI acceleration extension for the native MiniMax H3 audio-video model. It reduces expensive transformer evaluations during sampling by retaining the packed hidden state after actual transformer steps, then using Chebyshev ridge regression to forecast that state on selected forecast steps. Forecast steps skip the H3 transformer blocks and continue through the native output and sampler path, with adaptive scheduling, sampler-aware safeguards, CPU/VRAM history storage, and native fallbacks. Spectrum is a training-free approximate accelerator: forecasted states alter the denoising trajectory, so outputs can differ from native H3 even with the same seed and workflow.

CommerceAgentBench is a benchmark developed and maintained by the Accio team at Alibaba International for evaluating whether AI agents can complete long-horizon business workflows in high-fidelity, stateful, reproducible replicas of online services. Its tasks cover browser operations, native-style command-line tools, file and spreadsheet production, API and MCP workflows, public-web research, supplier analysis, product publishing, logistics, storefront configuration, and other commerce operations. Each task runs in a fresh container and is graded by a deterministic or LLM-assisted verifier. The benchmark uses local mock services representing commerce, SaaS, messaging, document, and operational systems; runs preserve the resolved configuration, agent trajectory, verifier result, artifacts, logs, and container metadata for auditing and reproducibility. The repository describes 107 tasks spanning CLI, browser, file, and API/MCP interfaces, with text-only, browser-text-capable, and vision-required capability slices. The project was previously called RealReplicaBench.

creakwork12 is a Rust utility that reads a Framework 12 laptop's hinge position and movement, turning the hinge into a creaky wooden-door sound. The sound pitch depends on the hinge position, while volume depends on how quickly the hinge is adjusted; its filtering is optimized for the Framework 12's hinge stiffness. It is distributed as an executable in the GitHub releases or can be built with Cargo, requiring the system's IIO and ALSA development libraries. The project was built on Fedora and may have dynamic-linking issues on other Linux distributions.

da-cli is a zero-dependency Python command-line tool for syncing DeviantArt galleries to local folders as a backup of watched artwork. It uses the DeviantArt API with OAuth 2.1 PKCE, stores a local SQLite index so repeat runs can avoid downloading unchanged content, and provides commands for syncing feeds, individual artists, or watched users, along with search and browsing helpers for tags, topics, users, deviations, and daily deviations. The distribution package is named da-sync and installs the da command. It supports unattended scheduling through launchd on macOS or a systemd timer on Linux; macOS secrets are stored in the Keychain, while Linux Secret Service support is planned. The project describes its sync and search flows as beta, with 868 tests reported in the README.

Desko is an open-source desk dashboard that turns a spare phone into an always-on display. A lightweight Python process on a PC collects data and pushes it over WebSocket to a browser tab on a phone on the same Wi-Fi network, with no phone app, account, cloud service, or build tooling required. The dashboard cycles through five scenes: an idle view with clocks, weather, battery, latency, and PC uptime; a music view with album art, synchronized karaoke lyrics, and PC playback controls; a statistics view with CPU and GPU load and temperatures, memory, network activity, sparklines, and an optional running game; a development view with workspace, Git branch status, changed files, cursor position, commits, and lines, sourced from VS Code or Git; and a focus view with a server-side Pomodoro timer. Scenes can be manually selected or pinned, and widgets hide when their underlying data sources are unavailable.

diri is a native desktop orchestrator for coding agents on macOS and Linux. It runs Claude Code, Codex, Cursor, Gemini, and ordinary shells in parallel, using separate Git worktrees or remote hosts over SSH and tmux. Each session is a real terminal with a persistent PTY managed by a background daemon, so closing or reopening the app does not terminate sessions; the daemon restores their output and state. The app displays working, needs-you, and done status, while its MCP server allows one running agent to spawn, monitor, read, and respond to another. Claude Code and Codex have first-class status detection and resume support, while Cursor and Gemini have partial support.

doc7 is an open-source document-to-Markdown tool that converts PDFs, Office files, scans, screenshots, charts, formulas, and diagrams into AI-ready Markdown using an OpenAI-compatible multimodal vision model. It renders and understands complete pages rather than relying on character extraction, preserving text, tables, formulas, chart and diagram relationships, image meaning, and visible UI state; the video also describes page-level retry and resume. The model can run locally through LM Studio or Ollama or through a remote endpoint, with no required OCR stack or document-processing service. The same binary provides an interactive CLI, batch processing, model checks, MCP support, a Go SDK, and an asynchronous HTTP service, with installers for macOS, Linux, and Windows.

Falco is a tiny browser engine written from scratch in Rust. It parses HTML, applies CSS, executes JavaScript in a custom virtual machine with a baseline JIT compiler, loads images, computes layout, and paints the result to a canvas that can be exported as PNG or displayed in a live interactive window. Its rendering pipeline runs from HTML through the DOM, style and layout trees, paint commands, and a canvas; the interactive mode supports scrolling, link navigation, form filling, and browsing history. The project includes its own HTML and CSS parsers, layout engine, SVG renderer, image loader, PNG encoder, WebAssembly interpreter, networking layer, and JavaScript DOM bindings, rather than wrapping WebKit, Gecko, or Chromium.

findphone is a macOS command-line tool for locating a nearby Bluetooth device by tracking signal strength when Find My is unavailable. It provides a survey mode for nearby Apple handhelds, a hunt mode for following a named device, and a listing mode for paired devices. An optional sound mode emits clicks that become more frequent as the smoothed signal strengthens, allowing a user to search by ear. The tool reads measurements from a connected BLE GATT link, passively observed BLE advertisements, or macOS Bluetooth data for paired devices, and reports signal strength in dBm as a coarse distance indicator. It is distributed as an unsigned universal arm64/x86_64 binary or can be built from Swift, requires macOS 13 or later, and has no dependencies.

Finger Frame AI is a browser app and Python CLI pipeline by Sophia Yang that places an AI-generated world inside a two-hand finger-frame gesture. It sends an uploaded video to Gemini Omni Flash for video-to-video restyling using a selected style or custom prompt, with alignment instructions intended to preserve framing, facial positions, expressions, blinks, gaze, and motion. MediaPipe Hand Landmarker then tracks both hands and the finger-frame quadrilateral frame by frame; the tracking pipeline uses anatomical corner ordering, spread and area gates with hysteresis, teleport rejection, velocity-adaptive smoothing, dropout holding, and presence fading. The compositor reveals the restyled video through the tracked quadrilateral and adds a dashed outline with pulsing corner markers. The browser app exports MP4 where browser recording supports it and otherwise WebM. The CLI separates restyling and compositing into `stylize.py` and `composite.py`, accepts formats readable by FFmpeg or OpenCV, and can produce H.264 MP4 while retaining the original audio track when present. Users provide their own Gemini API key; the app keeps it in the browser and sends it only to Google's API. Generation is billed per video, and the app recommends keeping uploads under about 15 MB. A no-key placeholder mode runs the tracking, compositing, and export pipeline with a hue-shifted stand-in.

GenOffice is a free, open-source AI office suite from Genspark AI for macOS, Windows, and Linux. It consists of six Electron applications sharing an engine layer: Docs for .docx files, Sheets for .xlsx spreadsheets, Slides for .pptx presentations, PDF for viewing and editing PDFs, Markdown for plain Markdown files, and a shell application that hosts the editors. It opens and saves Microsoft Office formats, and includes light, dark, and system themes. Its AI panel supports document-aware, block-level edits with version snapshots and diffs; in spreadsheets, presentations, and PDFs it provides a tool-calling agent over document state. Built-in tools include web search, image search, image generation, and media analysis. Users can sign in through Genspark or supply their own keys for providers including Claude, OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen, Doubao, MiniMax, Grok, Mistral, and OpenRouter, as well as custom OpenAI-compatible endpoints and local servers. GenOffice Docs uses byte-preserving .docx editing in which only modified paragraphs are regenerated, with Word-compatible pagination, tracked changes, comments, styles, equations, and ink. GenOffice Sheets uses the open-source Univer core with in-house extensions and a Rust .xlsx import/export sidecar, plus charts, pivot tables, slicers, conditional formatting, and formula tracing. GenOffice Slides has an in-house .pptx parsing, rendering, and editing engine with masters, layouts, charts, cropping, ink, and text shaping. GenOffice PDF supports annotations, forms, outlines, stamps, signatures, page operations, printing, direct text reflow and image editing, and local conversion to editable Word, PowerPoint, or Excel files; scanned PDFs can use system OCR on macOS and Windows. GenOffice Markdown is a Tiptap block editor that saves headings, lists, tables, images, and code blocks back to plain Markdown. The suite is distributed under the Apache-2.0 license with installers for the three supported desktop platforms.

Gmail is an email service developed and operated by Google that provides web-based email, POP/IMAP access, storage, spam filtering, and integration with other Google services. It was launched in 2004 and is available for individual and enterprise (Google Workspace) users.

Jungle Trail is a first-person procedural jungle walkthrough built with Three.js, taking the player along a winding trail toward overgrown stone ruins and a waterfall. It uses no external art assets: textures, meshes, materials, character elements, and sound are generated in code, including GLSL-based procedural textures and normal maps. The scene is assembled from procedural terrain, plants, stone blocks, and a character, with synthesized audio generated at load time. The project is a plain ES-module web application with no build step and can be run through a static file server. It provides desktop mouse-and-keyboard controls, touch controls, and debug and teleport options; the README states that it is intended for a desktop GPU and has no reduced-quality mobile version.

King's Gambit is a browser-based 3D chess game in which sculpted, rigged character pieces from the Ivory Kingdom, Sun Empire, or Grande Armée march, attack, and dissolve during cinematic captures. It supports standard chess rules—including castling, en passant, promotion, checkmate, stalemate, repetition, the fifty-move rule, and insufficient-material draws—through chess.js, while a Web Worker search engine provides the computer opponent. The game is built with Vite, React, TypeScript, and three.js; it is a static site with no backend or account requirement.

MAGI-2 Preview is the inference implementation for SandAI’s unified audio-video generation model. The 114-billion-parameter architecture, built on MagiMoE, activates about 6 billion parameters per token and generates 10-second clips from text prompts (T2V) or a prompt plus a still image (I2V), with sound generated alongside the video and muxed into the output file. Generation runs in two stages: `magi2_preview` denoises the clip at low resolution, after which `magi2_refiner` increases it to 1080p. The current base release is not step-distilled, so denoising steps account for most of the generation time; a distilled release is described as forthcoming. The repository contains inference code, while the model weights are downloaded separately from the `sand-ai/MAGI-2-preview` Hugging Face repository and occupy hundreds of gigabytes in total. An optional prompt-enhancement stage sends the input to an OpenAI-compatible instruction-following LLM, which rewrites it as a structured JSON caption for the 10-second clip, renders that result as Markdown, and passes it to the generation pipeline. It can be disabled to use the raw prompt. The documented runtime requires eight NVIDIA Hopper GPUs, Python 3.12, a recent CUDA toolkit, and FFmpeg for audio-video muxing; Docker images and a source installation are provided.

Microsoft Outlook is an email and calendar client developed by Microsoft for consumer and enterprise use. It is available as a desktop application for Windows and macOS, mobile apps for iOS and Android, and a web app at outlook.office.com. Outlook connects to Exchange Server and Microsoft 365 accounts and supports IMAP/POP/SMTP, calendar scheduling, contact management, rules and add-ins; features include Focused Inbox, Teams and meeting integration, and support for PST/OST data files and enterprise deployment policies. It is distributed as part of Microsoft 365 subscriptions, as a standalone product, and as the free Outlook.com web service for consumers.

morphicons is an open-source, zero-dependency JavaScript library for morphing one stroke-based SVG icon into another with spring physics. It accepts icon data such as IconNode objects or raw path strings, supports icon sets including Lucide, Tabler, Heroicons, and Iconoir, and can use custom paths. Its morphing algorithm solves the optimal similarity between shapes using 2D Procrustes analysis and polar interpolation, allowing rotation and scale to emerge mathematically rather than being declared for each icon pair. The library provides React, Vue, Svelte, React Native, vanilla custom-element, Astro, and core usage options, including uncontrolled prop-driven animation, controlled progress, and imperative morphing. It is distributed as an ESM package with optional framework peer dependencies.

Node.js is an open-source, cross-platform JavaScript runtime environment used to run server-side and other JavaScript applications. The videos describe it as a common runtime and application stack for web applications, including full-stack TypeScript projects, while the official project repository documents Current, Long Term Support (LTS), and nightly release channels. The project uses an open governance model and is supported by the OpenJS Foundation; it is distributed under the MIT license.

OpenAI Codex is an AI coding agent from OpenAI available as a command-line tool (Codex CLI) that helps developers produce software. It can be used alongside Gemini for adversarial audits of software requirements and implementation plans, listed as a supported coding-agent or model option in several projects, and its logs can be joined with task and test evidence. The Codex CLI can also receive and answer requests from the Penako canvas.

OpenEdit is an open-source, agent-driven video-editing pipeline from VEED. It has no graphical interface or timeline; an agent-agnostic skill and repository guide drive it through coding agents such as Claude Code, Codex, and Gemini. The pipeline can edit, cut, and reframe footage; create burned-in subtitles, motion graphics, charts, and visual elements; assemble slides, websites, stills, or generated media into video; capture web pages; and connect generation services or MCP servers when needed. Source footage is optional, and VEED Fabric can generate footage only with the user's approval. For captioning, the agent transcribes the video, designs the captions, renders the result, opens a video previewer, and accepts follow-up requests such as changing subtitle color, position, or emphasis. Transcription can use VEED, local offline WhisperX, or a user-supplied service; all providers must produce per-word timings, which are stored in a common transcript format. Caption styles are authored in HTML and CSS. The project includes VEED's closed-source but free-to-use HTML renderer, which does not require a headless browser; Chrome can be used as an alternative rendering backend. V1 primarily targets captions, while motion graphics, charts, and brand-book-matched styling are less exercised. OpenEdit supports Apple Silicon Macs and Windows x64 PCs. Intel Macs are unsupported, Linux support is planned, and Windows requires Git, Node, and ffmpeg. It can be installed from the repository or with `npx skills add veedstudio/open-edit`.

Outreachr is a free, open-source, local-first fundraising CRM for founders. It provides a private desktop workspace backed by a founder-owned SQLite database rather than an Outreachr-hosted account or subscription, covering investor research, target ranking, introductions, outreach, meetings, diligence, and committed capital. The application searches bundled investor research, ranks targets against factors such as stage, sector, geography, and published check evidence, and shows the supporting records and sources behind each score. Founders can move firms through research, introductions, meetings, diligence, soft-circle, commitment, pass, and not-now stages; draft, review, approve, and send one-to-one initial messages through Gmail or Outlook; and reconcile selected relationship metadata from existing mailboxes. Its outreach controls include duplicate-send prevention, sending caps, domain cooldowns, suppression lists, and founder approval before messages reach a mail provider. Codex or Claude may propose drafts, research, tasks, and pipeline changes from explicitly permitted local context, but agents cannot send email or bypass approval.

Python is a programming language and the implementation environment maintained in the CPython repository. The videos describe its use for back-end applications, bots, scientific analysis, API scripts, file-operation primitives, database and vector-search result processing, coding-session dashboards, document-converter bindings, and compiled agent skills. CPython can be built on Unix-like systems and Windows, with optional profile-guided and link-time optimization; installable distributions and documentation are provided through python.org.

QuerySplat is an open-source implementation for predicting and rendering 3D Gaussian splats from multiple images of a scene while decoupling geometry and appearance representations. Its inference pipeline preprocesses custom images, predicts 3D Gaussians, and uses VGGT-Omega components for camera and depth prediction; optional test-time optimization can refine the result. It can export Gaussian PLY files, predicted input cameras, per-view depth and confidence data, and depth-derived point clouds. The repository requires Linux, an NVIDIA CUDA-capable GPU, and CUDA-enabled PyTorch.

Rust is a systems programming language and toolchain developed by the Rust project. Its compiler, standard library, documentation, package manager and build tool Cargo, formatter rustfmt, linter Clippy, and editor support are maintained in the main repository. The language uses a rich type system and ownership model to enforce memory and thread safety at compile time, while targeting fast, memory-efficient software for critical services, embedded devices, and integration with other languages. Rust is primarily distributed under the MIT and Apache 2.0 licenses, with portions under BSD-like licenses.

Shitty is a terminal emulator for macOS and Linux/Wayland, built for low latency, fast startup, and predictable resource use. It keeps terminal state on the CPU and renders changed cells with native compute backends: Metal on macOS and Vulkan on Linux. It supports VT52 through VT5xx and ECMA-48 controls, ISO-2022 character sets, common xterm extensions, primary and alternate screens, scrollback reflow, hyperlinks, shell marks, mouse and keyboard input, resizing, HiDPI and fractional Wayland scaling, and Unicode 17 grapheme clusters including emoji sequences and wide CJK characters. Resize frames are rendered in the same transaction as bounds changes, and updates are damage-driven. The project provides native macOS and Linux/Wayland frontends as a small self-contained binary with embedded fonts. Its README describes more than 5,000 black-box tests driven through a real PTY, a fuzzed parser state machine, and a default security model that prevents applications from reading selections or controlling the host window unless explicitly permitted. The same terminal code is also distributed under the user-facing name Pretty.

Soup is an open-source Python CLI for fine-tuning and post-training large language models from a YAML configuration and a single training command. It supports supervised fine-tuning, LoRA, QLoRA, preference optimization, quantization, automatic batch-size and GPU detection, and local training without a cloud service. Its optional layer-streaming mode keeps the frozen base model out of VRAM and feeds it to the GPU one decoder layer at a time, reducing memory use for low-VRAM systems. The project describes training an 8B model on a 4 GB laptop GPU with this mode; layer streaming is marked beta.

SQLite is a self-contained, serverless, zero-configuration, transactional SQL database engine implemented as a C library. It was created and is primarily maintained by D. Richard Hipp and the SQLite development team and is widely used as an embedded database in applications, devices, and browsers.

Stickman Video Director is a Codex skill that turns source copy, notes, articles, or a topic into a planned one-minute stickman video workflow. It first produces a six-scene directing proposal covering English narration, reference translation, visual action, camera movement, transitions, background music, sound effects, color treatment, and continuity; only after explicit approval does it generate six standalone Gemini Omni Flash prompts for roughly ten-second segments. Each scene specifies three timing beats and repeats key character, line-weight, palette, audio, dialogue, transition, and negative constraints so the prompts can be used independently, then assembled into a continuous video. It adapts composition and camera paths for 9:16, 16:9, or 1:1 formats and supports white-background/black-stickman or black-background/white-stickman styles with up to three accent colors. Global changes return the workflow to the proposal stage, and the skill is designed to avoid adding unsupported facts or claims. It requires no API or MCP integration and is distributed under the MIT license.

Swiftlet is an open-source Swift and Metal runtime for running Qwen3-Next and Qwen3.5/3.6 mixture-of-experts models locally on Apple devices, including iPhones. It keeps the models' small dense core in memory and streams routed expert weights from storage on demand, allowing 35B and 80B models to run with comparatively low RAM. The project is available as a Swift package, command-line interface, OpenAI-compatible server, and iOS app; its repository describes the runtime as working end to end and notes that only about 3B parameters are active per token in the supported models.

Token Saver is a local Claude Desktop extension developed at Marktechpost AI Media Inc. It searches PDFs stored on the user's computer using local hybrid retrieval—BM25 and embedding search—then sends Claude only the passages relevant to a question, with exact page citations, instead of placing the entire document in the context window. The extension is distributed as a one-click .mcpb file and bundles its search engine, requiring no Python, terminal, or configuration files; documents remain on the local machine. It also supports document-folder selection, document listing, disambiguation when multiple files match, and reusing a loaded document for follow-up questions.

Virtual Mac on iPad is a community project that runs macOS in a hardware-virtualized virtual machine on compatible Apple-silicon iPads. Its tooling extracts and rebuilds the virtualization frameworks required from macOS's shared cache, allowing macOS applications such as Xcode, Terminal, Final Cut Pro, Logic Pro, and Pixelmator Pro to run on the device. The project requires a jailbroken iPad Pro with an M1 or M2 chip or an M1 iPad Air running iPadOS 14 through 16.3.1. It is distributed as a package through the project's Cydia repository and installed with Sileo; the project states that it is unaffiliated with Apple, Dopamine, UTM, and VirtualBuddy.

vPhone Workstation is a native macOS SwiftUI application for managing virtual iPhones. It provides a library with live running or stopped state, iOS and cloudOS builds, search, and VM details including variant, CPU, memory, disk, network, device, and UDID. Users can start, stop, foreground, clone, rename, export, configure, and delete virtual machines. Its creation wizard selects a VM variant, firmware pairing, and CPU, memory, and disk resources, while streaming progress from vphone-cli and producing an exportable log. vPhone Workstation is a graphical front end for vphone-cli, which boots iOS research virtual machines through Apple's Virtualization.framework using PCC research VMs. It also checks host readiness, including the required macOS version, Apple Silicon hardware, vphone-cli availability, research-guest permissions, and AMFI configuration. The repository documents macOS 15 or newer as a requirement and provides Homebrew installation and source-build instructions.

Waku Agent is an open-source, local-first personal assistant and AI-agent harness built by seanchen.io. It runs on a laptop through a terminal, local browser dashboard, voice input, or Telegram gateway, and is organized around four components: a harness for tool execution, an approximately 95-line plain-Python agent loop, memory, and evaluation/LLM operations. Its memory uses semantic, episodic, and procedural stores in a local SQLite database, with a gate that decides whether to retrieve or save memory and a pass that determines what to retain. The project includes deterministic tests alongside LLM-as-judge evaluation with a release gate. The dashboard displays message flow through the harness, including gate decisions, tool calls, loop iterations, memory updates, traces, costs, and latency; it also includes views for workflows, tools and MCP connectors, memory, and the SQLite state database. It can be installed with pip and exposes the `waku` command. The dashboard runs a local web server at `127.0.0.1:7777`, and the Telegram gateway starts when `TELEGRAM_BOT_TOKEN` is configured. Model providers are selected through a provider setting and API key, with support stated for Anthropic, OpenAI, Gemini, DeepSeek, MiniMax, Kimi, GLM, OpenRouter, OpenCode Zen, and OpenCode Go.

whisper.cpp is a C/C++ implementation of OpenAI's Whisper automatic speech-recognition model for speech-to-text inference. It is designed for offline, on-device use and supports CPU-only inference, GPU and NPU acceleration, integer quantization, mixed F16/F32 precision, voice activity detection, and a C-style API. The implementation targets macOS, iOS, Android, Linux, Windows, WebAssembly, Raspberry Pi, Docker, and other platforms, with optimizations including ARM NEON, Apple's Accelerate and Metal frameworks, x86 AVX, POWER VSX, Vulkan, CUDA-related NVIDIA support, ROCm, OpenVINO, and other backends. The repository includes the whisper-cli example, which transcribes 16-bit WAV audio using Whisper models converted to ggml format; its zero-dependency implementation is intended for integration into applications such as on-device captioning tools.