2,419 tools and products — trending open source, and what gets used in AI and other work.
omg.dev is an open-source, self-hosted parallel coding-agent harness and control plane developed by BennyKok. It runs coding-agent sessions on a local computer or hosted Computer, keeps them running when the web interface disconnects, and provides a single web UI for parallel sessions, transcript reading, follow-up instructions, chats, bots, schedules, and notifications. The project supports agents including Claude Code, Codex, Grok, Cursor, OpenCode, Jcode, GitHub Copilot, and Pi, using the user's existing agent subscriptions or API keys. Local installations run through a Bun CLI and expose the interface on localhost; remote phone access can be provided through Tailscale. The repository is licensed under the MIT License.
The Temporal TypeScript SDK is an open-source framework for authoring workflows and activities in TypeScript or JavaScript against Temporal's distributed orchestration engine. Temporal executes asynchronous, long-running business logic with durable state and recovery across failures, while the SDK provides client packages for starting and interacting with workflows and worker packages for running workflows, activities, and Nexus Operations. Worker-level execution uses Node-specific capabilities including native modules, worker threads, virtual-machine isolation for the workflow sandbox, AsyncLocalStorage, and async hooks. The repository distributes the SDK as version-aligned @temporalio/* packages and officially supports Node.js 20, 22, and 24; alternative server-side JavaScript runtimes may work for some client features but are not officially supported for workers.
loop.js is a TypeScript framework and CLI for loop engineering: it runs an agent against a declared goal and verification criterion until a separate, skeptical Verify agent judges the work complete. Each round starts with fresh context and reads state and handoff notes from disk; the worker agent cannot approve its own work, and a failed verification supplies a reason for the next round. The framework provides round, spending, and per-round timeout guards; typed exits; crash recovery; idempotent resumption from a stored cursor; and a compare-and-set lock with heartbeat to prevent overlapping live runs and take over dead ones. It can run from a terminal, be embedded in a product, or be scheduled through crontab, launchd, Task Scheduler, or Modal's cloud.
Codex Orchestration is a Codex plugin that coordinates multiple models and role-based stages within a single task. It can assign optional Planner, Advisor, and Designer roles, plus a required Executor; the selected Codex model passes work between them, checks each result, and returns the final answer. The Planner creates and revises a plan, the Advisor reviews it and can return PLAN_APPROVED, the Designer produces an optional visual or UX handoff, and Executors implement the approved plan, potentially in parallel. The workflow stops after approval or after a safety limit of eight reviews, showing unresolved issues if approval is not reached. It is installed through the Codex plugin marketplace and requires Python 3.11 or newer.
self-learning-skills is a meta-skill for AI coding agents, including Claude Code, Cursor, Codex, and agents that read AGENTS.md or similar standing-instruction files. It recognizes hard-won, reusable procedures such as non-obvious commands and recurring operational workflows, then captures them as skills, rules, or project instructions for later sessions. The capture includes known dead ends and excludes one-off details and secret values. The project stores learned procedures as new skills/<name>/SKILL.md files for Claude Code, Codex, and Agent Skills clients; as .cursor/rules/learned/<name>.mdc files for Cursor; or in AGENTS.md or project notes for other agents. It can be installed with the community skills CLI, as a Claude Code plugin, or manually by copying the relevant files.
Three.js Game Skills is a set of self-contained Codex and Claude Code agent skills for building playable Three.js browser games. Its threejs-game-director skill routes work across gameplay, graphics, UI, asset generation, audio, debugging, and release verification, so users can request an outcome without manually selecting specialist skills. The package includes SKILL.md files, references, checklists, prompt templates, helper scripts, and a Vite, TypeScript, and Three.js scaffold. Generated games include deterministic test hooks and seeded randomness, with Playwright templates for smoke tests, visual-regression baselines, and bot playtests; the documented workflow also checks browser behavior, mobile viewports, visual output, UI, performance, and release readiness. It is created by Majid Manzarpour and can be installed for Codex or Claude Code through the skills CLI or the repository's installer.
Pumpkin is an open-source Minecraft server written entirely in Rust. It is designed for fast, efficient, customizable hosting, using multithreading and adhering to vanilla game mechanics. The project targets Java Edition and Bedrock Edition clients, although Bedrock support is marked as a work in progress. It provides configuration through TOML and implements Minecraft protocol features including encryption, packet compression, server status and ping, world loading and saving, chunk generation, entity spawning, player movement, inventories, combat, commands, permissions, chat, and plugin support. It also supports RCON, queries, and proxy systems including BungeeCord, BungeeGuard, and Velocity. Pumpkin is under heavy development. The server is licensed under GPLv3; its plugin APIs are dual-licensed under MIT or Apache-2.0.
Catch2 is an open-source, C++-native unit testing framework for unit tests, test-driven development (TDD), and behavior-driven development (BDD). It uses readable test names, assertions written as ordinary C++ boolean expressions, and sections for localized sharing of setup and teardown code. It also provides simple BDD macros and basic microbenchmarking; benchmarks are opt-in rather than run by default. Catch2 v3 is distributed as a normal library with multiple headers and separately compiled implementation, while the v2.x branch provides the previous major version and C++11 support.
Gigatoken is an open-source tokenizer for language-model training data, implemented in Rust and distributed as a Python package. It tokenizes large text corpora at gigabytes per second, supports common tokenizer formats across x86 and ARM hardware, and provides compatibility modes for Hugging Face Tokenizers and Tiktoken. Its native API can read text files directly and encode them with parallel processing, while the compatibility modes are designed as drop-in replacements but incur additional overhead. The repository reports throughput of up to roughly 1,000 times that of Hugging Face tokenizers in its benchmarks.
Palmier Pro is a Swift-native macOS video editor built for AI, with generative video and image models integrated into the timeline. It can expose a local MCP server at http://127.0.0.1:19789/mcp for connections from Claude, Codex, and Cursor, and includes an in-app agent that can create and edit within the same project. It requires macOS 26 (Tahoe) or later on Apple Silicon. Releases through v0.7.6 and the corresponding published source are available under GPLv3; later binary releases are proprietary.
ReUI is a free, open-source React component library for the shadcn/ui ecosystem, developed by Keenthemes. It provides more than 1,000 copy-and-own registry examples across 71 categories, presented in realistic dashboard layouts rather than isolated demos, including data grids, filters, timelines, steppers, trees, calendars, kanban boards, and other interface blocks. The components are composed from shadcn/ui primitives and Tailwind CSS, with 19 custom in-house component types. Its Data Grid uses TanStack Table and virtualization and supports features such as drag-and-drop, pinning, resizing, and infinite scrolling. The in-house components have both Radix UI and Base UI versions and are distributed as TypeScript source through the registry rather than as an npm package, allowing the source to be copied into a React project. ReUI is released under the MIT License.
deepsec is an open-source, agent-powered vulnerability scanner and security harness from Vercel Labs for on-demand review of large codebases within the user's own infrastructure. It identifies candidate vulnerabilities, uses AI models to investigate them, supports triage and optional revalidation to reduce false positives, and can run work across parallel worker machines. The CLI initializes a repository with `npx deepsec init`, stores state and findings in a `.deepsec/` directory, and resumes interrupted runs while skipping files already analyzed. Subsequent commands separate fast pattern scanning from AI processing and revalidation, and findings can be exported as Markdown directories or JSON. Model access can use Vercel AI Gateway, direct OpenAI or Anthropic credentials, or a custom HTTPS provider; scan cost and duration can be capped with command-line options.
MuScriptor is an open multi-instrument music transcription model developed by Kyutai and Mirelo. It converts audio recordings into note events, MIDI, or engraved sheet music with instrument information, and provides a locally hosted web UI, command-line interface, and Python package. Its architecture is a transformer decoder-only model; small, medium, and large variants are distributed through Hugging Face and downloaded and cached automatically after authentication. Sheet-music output uses MuseScore 4 or newer to generate MusicXML and PDF scores, including instrument-specific and tablature PDFs. The project can be run with uvx or installed from PyPI, and the model weights require acceptance of the CC BY-NC 4.0 license. The README notes that transcription works best with a steady tempo and performs significantly worse on rubato recordings.
LLM Space v4 is a local-first desktop workbench for prototyping and developing AI agents. It lets users version prompts, system messages, tools, and model settings; trace each model call and tool run in an agent loop; replay historical runs to debug failures; and evaluate agent performance across runs. It can generate prompts and tools with AI and turn a thread into a runnable LangGraph agent. Threads, project files, and API keys remain on the user's computer, while the project describes the app as cloud-ready for managed agents. The desktop application uses Electrobun with a React, Tailwind CSS, and shadcn/ui interface, and is built as a Bun monorepo around Pi Agent Core.
OpenAI Evals is an open-source framework for evaluating large language models and systems built with them, including tool-using agents and prompt chains. It includes a registry of benchmark evals, supports custom model-graded and private evals based on a user's data, and uses a completion-function protocol for advanced workflows. The package can be installed with pip and run locally with an OpenAI API key; the repository also documents optional result logging to Snowflake and configuration through the OpenAI Dashboard.
Claude Opus 4.5 is an AI foundation model developed by Anthropic for complex reasoning, software engineering, and agentic workflows. It is available through Anthropic's API and Claude applications.
Together AI (branded Together) operates an "AI native cloud": a platform for deploying, running, and scaling AI models and applications. The service provides infrastructure and tooling for developing and serving machine learning and large language models.
Vercel Connect is a Vercel service for connecting autonomous agents and applications on the Vercel platform to external systems, APIs and event sources while mediating and restricting data and event access. Developed by Vercel as part of its agentic infrastructure, it provides connectors and permission controls that determine what data and events an agent can read, transmit, or act on. The service is exposed via Vercel's agent framework and platform APIs and runs on Vercel's hosting infrastructure.
The Vercel AI SDK is an open-source software development kit from Vercel that provides components and utilities to integrate large language models and other AI features into web applications. It supports multiple model providers and both server- and client-side usage, and is maintained by Vercel on GitHub.
Fluree is an enterprise AI data platform that turns raw, structured, and unstructured data into trusted, queryable knowledge graphs. Its platform uses governed data models, taxonomies, golden records, entity resolution, semantic layers, and GraphRAG-powered retrieval to prepare enterprise data for conversational analytics, enterprise search, decision intelligence, and AI agents. Users can ask questions, create dashboards, build applications, and deploy governed agents over the resulting knowledge graph.
Freesolo is a full-stack platform for post-training of small models. The company offers a managed post-training service (mentioned in videos as 'Flash') that uses an agent-driven workflow combining supervised fine-tuning and reinforcement learning to produce specialized, downloadable small-model weights and to deploy models via an OpenAI-compatible API.
Prosed is a web service that assembles newsletters, posts, and other content into a publish-ready book. It converts a creator's existing content and ideas into a formatted manuscript for publishing.
PromptQL is a collaborative AI platform that provides shared AI threads and a self-organizing team knowledge base. It captures conversational context and generates suggested edits to interconnected wiki-style pages representing skills, knowledge, and a semantic layer.
AskCodi is an AI coding assistant and developer platform that provides multi-model AI chat, code generation, refactoring, and custom agent capabilities. It offers an OpenAI-compatible API, IDE integrations, and developer-focused workflows.
Moxie Docs is a hosted service that automatically generates and maintains documentation from GitHub repositories. It re-checks docs on every merge, opens cleanup pull requests when documentation drifts, and provides context for AI agents plus hosted public help centers.
Wispro is a desktop voice dictation assistant for Windows that converts natural speech into polished, formatted prose and pastes it into the active application, enabling instant insertion of dictated text.
Notte is a browser framework and platform for building and running AI agents, providing tools to simplify web-based LLM tasks, handle CAPTCHAs, and offer developer tooling for agent development and deployment.
A macOS power-management app that automatically manages Mac power usage and charging. It measures power-saving actions locally, limits charging to 80% to reduce battery wear, and learns when to fully charge the battery before the user disconnects the Mac.
OpenCode Superapp is a native workspace developed by OpenCode that integrates local, self-hosted, or cloud AI models with realtime voice interaction. It offers private dictation, Computer Use (control of local computer tasks), extensions, widgets, and secure remote access. The site describes the product as Codex-inspired.
Cosyra is a mobile coding terminal for iPhone and Android that provides cloud-hosted Ubuntu terminals to run developer tools and CLIs (examples listed include Claude Code and Codex CLI) without requiring users to manage their own VPS.
Megaphone is a podcast hosting, monetization and ad‑tech platform that provides episode publishing, analytics, dynamic ad insertion and an ad marketplace for publishers and advertisers. The platform was acquired by Spotify in 2020.
AgentLoop is a developer tool that runs unattended coding loops on a user's machine, executing plans authored in ChatGPT by launching Codex-based workers and independent critic processes. It automates iterative code generation and evaluation until configured quality standards are met.
Blaxel is an infrastructure platform for autonomous AI agents that provides isolated microVMs (which boot in milliseconds and resume in ~25ms), persistent shared memory, and programmable networking to run, connect, and operate agents at scale. It offers a distributed shared-storage feature (presented in videos as "Agent Drive") that can be mounted as a POSIX-style drive so agents in separate sandboxes can share files, cache dependencies, exchange context/memory, and coordinate parallel work in real time.
Pulse is a team knowledge-management product that records a company's decisions, commitments, and lessons and answers questions in plain English. It is permission-aware (showing content only to those with access) and targets teams of roughly 5–500 people; the site lists the product as in early access.
Last 30 Days is an open-source AI-agent research skill developed by mvanhorn. It searches configured sources in parallel—including Reddit, X, YouTube transcripts, TikTok, Hacker News, Polymarket, GitHub, arXiv, Techmeme, and other web and social platforms—then ranks material using signals such as upvotes, likes, views, comments, engagement, and real-money prediction-market odds. An AI-agent judge synthesizes the results into a cited brief focused on recent information, with source-specific outputs such as quotable transcript passages, Reddit comment scores, GitHub activity, and Polymarket probabilities. Reddit, Hacker News, Polymarket, and GitHub work without additional configuration. A setup wizard can enable further sources using user-provided API keys, browser sessions, or locally available source connectors. The skill can be installed through the Claude Code marketplace or with the Skills CLI for Claude Code, Codex, Cursor, Copilot, Gemini CLI, and other agent-skill hosts.
GeoLibre is a free, open-source, cloud-native GIS platform for visualizing, exploring, and analyzing geospatial data. It supports GeoParquet, GeoPackage, and shapefile layers, with live styling, attribute tables, saved projects, plugins, and an optional Python package for use in Jupyter notebooks. The same workspace runs in web browsers, native desktop apps, mobile apps, and Jupyter notebooks while keeping data local. Its browser-based geoprocessing tools run through WebAssembly without a server or data leaving the machine. The project is built with Tauri v2, React, TypeScript, MapLibre GL JS, DuckDB-WASM Spatial, and deck.gl, and also provides R and Python packages plus installers and mobile distribution.
bitchat is a decentralized peer-to-peer messaging app with two transport layers: Bluetooth mesh for offline communication and the Nostr protocol for internet-based messaging. It does not require accounts, phone numbers, or central servers. The app discovers nearby peers over Bluetooth LE and can relay messages across a mesh network for up to seven hops, automatically falling back to Nostr relays when internet connectivity is available. It supports location-based channels using geohash coordinates, IRC-style commands, and end-to-end encrypted private messages using the Noise Protocol on the mesh and BitChat private envelopes over Nostr. The project also documents message compression, battery-saving modes, and an emergency data-wipe gesture. The repository provides native iOS and macOS support and links to App Store and Play Store distributions. It is released into the public domain. The README notes that nearby radios can observe a persistent per-device identifier derived from the identity key, despite the app requiring no account or phone number.
Box3D is an open-source 3D physics engine for games, written in portable C17. It provides rigid-body simulation with collision detection, continuous collision detection, contact events, convex hulls, capsules, spheres, triangle meshes, height fields, multiple shapes per body, collision filtering, ray and shape casts, overlap queries, sensors, and a character mover. Its physics system includes a robust Soft Step rigid-body solver, continuous physics, island-based sleeping, and revolute, prismatic, distance, motor, weld, and wheel joints with limits, motors, springs, friction, and force events. The data-oriented system uses multithreading and SSE2 or Neon SIMD, supports cross-platform determinism and recording and replay, and is optimized for large piles of bodies. The core library has no dependencies beyond the C runtime and libm on Unix, can be integrated with CMake as the box3d::box3d target, and builds on Windows, Linux, macOS, and WebAssembly; its samples use sokol and Dear ImGui.
Awesome LLM Apps is an open-source GitHub collection of Python templates for AI agents, agent skills, and retrieval-augmented generation (RAG) applications. Its examples include starter and advanced agents, multi-agent teams, voice and multimodal agents, MCP-based agents, memory-enabled agents, data analysis, web scraping, and other application patterns. The repository provides runnable source projects and step-by-step tutorials. Individual agents can be cloned and run with an API key, while agent skills can be installed with a command such as `npx skills add` and used with coding agents including Claude Code, Codex, and Cursor. The projects support models from providers and ecosystems including Claude, Gemini, GPT, DeepSeek, Llama, and Qwen. The repository describes the collection as free and open source under the Apache-2.0 license.
An open-source Python package of shared utilities for Unsloth, supporting large-language-model training and fine-tuning workflows such as full fine-tuning, LoRA, quantized training, and reinforcement learning. The associated Unsloth project provides beginner-oriented notebooks, supports models including gpt-oss, Gemma, Qwen, Llama, and Mistral, and can export fine-tuned models to GGUF, Ollama, vLLM, or Hugging Face.
Hubble.md is a free, open-source Markdown and HTML note-taking app for people and AI agents. It provides a Notion- or Apple Notes-style editor with Markdown shortcuts, slash commands, and frontmatter properties, while allowing an agent to work directly with the notes folder; edits made by an agent are live-reloaded in Hubble. Beyond Markdown notes, Hubble can display HTML-based apps built from a notes folder, including views such as tables, bookshelves, and maps. It is distributed as an Electron desktop app for macOS, Windows, and Linux, and the repository also contains a web app, Markdown editor packages, an HTML-app runtime, a filesystem sync engine, and a `hubble` command-line interface.
Native SDK is an open-source toolkit for building native desktop applications. It uses declarative markup in `.native` files for views, plain TypeScript compiled to native code at build time or Zig for application logic, and an engine that renders directly into real operating-system windows without a browser, WebView, or JavaScript runtime in the application binary. The CLI can scaffold, run, check, and build applications, with live view updates during development and an optimized release-binary build. Its component catalog includes controls such as buttons, tabs, text fields, dialogs, charts, and virtual lists, while styling is organized around replaceable design tokens. The toolkit also includes an embedded automation server for agents.
mise is an open-source command-line tool for managing development tools, environment variables, and project tasks from a single mise.toml configuration file. It installs and switches between tools such as Node.js, Python, CMake, Terraform, Go, and jq, and prepares the configured environment before commands run. Project-specific environment variables can include values from .env files, while tasks define commands and dependencies for activities such as building, testing, linting, validation, and deployment.
Guava is a set of core Java libraries developed by Google. It adds collection types such as multimaps and multisets, immutable collections, a graph library, and utilities for concurrency, I/O, hashing, primitives, strings, and related tasks. The distribution provides separate JRE and Android flavors. The JRE flavor requires JDK 1.8 or higher; both flavors are published as Maven artifacts under the com.google.guava group ID with the guava artifact ID, and can also be added through Gradle. APIs marked with @Beta may change or be removed, while non-beta APIs are intended to remain binary-compatible.
DwarfStar is a self-contained, model-specific native inference engine for running DeepSeek V4 Flash, GLM 5.2, and, on high-memory systems, DeepSeek V4 PRO locally. It combines model loading, prompt rendering, tool calls, persistent KV state, an HTTP server, and a coding agent rather than serving as a general GGUF runner. The engine supports Metal on Macs, CUDA including multi-GPU systems, and ROCm on Strix Halo hardware; it also provides SSD streaming for machines with insufficient RAM, distributed tensor and pipeline parallelism, and server-side micro-batching for multi-user inference. The repository includes tooling and data for GGUF, imatrix, quality, and speed testing, and acknowledges code and design contributions from llama.cpp and GGML.
A DataExpert-io GitHub repository that curates learning resources for data engineering. It organizes a 2024 career roadmap, free beginner and intermediate boot-camp materials, required software, hands-on projects, interview guidance, books, communities, newsletters, company and tool directories, engineering blogs, whitepapers, and data-engineering creator accounts. Its categorized references cover areas including orchestration, data lakes and clouds, warehouses, data quality, analytics, data integration, semantic layers, modern OLAP, LLM application libraries, real-time data, and data lineage.
npmx.dev is an open-source web frontend for browsing the npm registry, developed as a fast, modern alternative interface rather than a replacement for the registry. It provides package search and details, README and metadata views, source-code browsing with syntax highlighting and line permalinks, generated API documentation when available, version diffs, changelogs, publish timelines, package comparisons, and shareable URL-driven views. It also surfaces dependency and package-health information such as outdated dependencies, OSV vulnerability warnings, provenance indicators, license and install-script warnings, install size including transitive dependencies, download statistics, repository signals, and JSR availability. The interface supports dark and light modes, localization across dozens of locales including right-to-left languages, keyboard workflows, screen readers, and package, team, and organization administration through the local npm CLI.
Ladybird is an independent open-source web browser developed around its own web-standards-based engine rather than an existing browser engine. Its multi-process architecture separates the main user interface, WebContent renderer processes, image decoding, and network requests; each tab has its own sandboxed renderer, while image decoding and network connections run out of process. The project is pre-alpha and intended for developers, with builds for Linux, macOS, Windows through WSL2, and other Unix-like systems. It is licensed under the 2-clause BSD license.
Nimbus is an open-source documentation-site scaffolder by Cloudflare that builds Astro sites directly in a repository. It writes layouts, components, styles, routes, content collections, and theme tokens as editable source files, while its plumbing is provided through npm packages. The generated sites include agent-readable Markdown and MDX page twins, llms.txt and llms-full.txt files, JSON-LD, sitemaps, robots.txt, per-page Open Graph images, full-text search, theming, accessible navigation, and optional versioned documentation. Nimbus also provides a registry for adding editable components, utilities, and agent-handoff recipes to an existing project. Sites produce static output for deployment anywhere, with Cloudflare as a first-class target, and include prose, structure, MDX, and configuration validation. The project is built on Astro, Sätteri, Tailwind, and optionally React; its README describes it as work in progress and pre-1.0.
Drawesome is a drawing toolbar and canvas component for React applications. It provides seven pen types—pencil, pen, brush, fineliner, highlighter, fountain pen, and another configurable pen tool—that vary stroke width according to movement speed or direction, plus an eraser that removes areas of strokes rather than deleting whole strokes. The React component supports configurable tools, colors, sizes, opacity, toolbar placement, appearance, controls, themes, tooltips, keyboard shortcuts, and initial or externally managed strokes. Drawings can be exported as SVG or PNG, downloaded directly, or accessed as stroke data through its ref methods. It is installed with npm, requires React 18 or later as a peer dependency, and has no runtime dependency beyond React.
Let's Seal is an open standard, reference implementation, and free service for proving that a file is unaltered, existed by a certain date, and was sealed by a specific certificate. It preserves native signature formats such as PAdES/X.509 signatures for PDFs and uses OpenTimestamps to anchor time to Bitcoin. Each seal is also recorded in an append-only transparency log with a Bitcoin-anchored root, allowing independent auditing. The project provides a hosted web app, the `sealbot` command-line tool, a REST API and SDKs, and a self-hosted engine with its own certificate authority; verification is available online or offline without an account. The repository states that the engine, SDKs, and standard are open source under the Apache-2.0 license.
world-model-optimizer converts OpenTelemetry traces from agent workflows into a routing policy. It scores models on held-out tasks, selects a model for each request, and provides world models for testing changes to prompts, tools, and runtime code.
Snowflow is a real-time procedural snow-rendering demonstration built with WebGPU, Babylon.js, and hand-written WGSL. It generates terrain, snow deformation, procedural characters, inverse-kinematics foot placement, cloth, surf wakes, water spells, atmosphere, and post-processing at load time without textures, meshes, HDRIs, or animation assets. Its terrain uses a nested-ring geometry clipmap whose vertex shader performs world placement, level-of-detail morphing, and displacement. Layered gradient noise produces the underlying heightfield, while a persistent GPU terrain-state buffer handles additive deformation through scrolling, relaxation, and splatting in a ping-ponged full-screen pass. The demo also provides GPU snow shading, cascaded PCSS shadows, procedural character and cloth systems, and in-browser settings, performance, and debug views. It requires a WebGPU-capable desktop browser and a discrete or recent integrated GPU; it intentionally has no WebGL fallback.
Qwen Scribe is a private, local transcription and system-wide dictation application for Apple Silicon Macs. It runs Qwen3-ASR through mlx-qwen3-asr on the Mac's Metal GPU, keeping audio and transcripts on the computer without an account, API key, or cloud service. Users can drag audio or video into a local web interface, transcribe with automatic or forced language selection, vocabulary hints, word timestamps, and SRT export, and search or export local transcript history. A menu-bar helper provides push-to-talk dictation in text fields, displaying listening and transcription states in a non-focus-stealing HUD before pasting the result at the cursor. The project requires macOS 14 or newer, Apple Silicon, Python 3.12 or newer, and ffmpeg for non-WAV audio and video. It offers Qwen3-ASR 1.7B and 0.6B model options; model weights are downloaded separately and require roughly 3.4 GB or 1.2 GB of unified memory respectively, in addition to application overhead. It is distributed as a downloadable beta Mac app or can be built from source; the repository describes the project as version 0.2.1-beta.1.
CodeJury is a terminal-first, knowledge-grounded multi-agent software delivery pipeline for scoping requirements, implementing changes, running tests, and gating pull requests. It divides delivery into six terminal-driven stages and uses isolated development branches, human approval gates, deterministic QA, and ensemble code review. The system indexes a repository into a persistent code graph and uses semantic search to retrieve relevant code during later work. Its review stage sends the same change to independent models from different providers; both must approve it for the change to pass, while a rejection returns the reason to the developer. A configurable panel can use up to six specialist reviewers and a foreperson to reconcile their findings. The project is distributed as a Python package installable with pip and documents support for macOS, Linux, and Windows. Its repository describes a one-command run and operation without an API key.
netherite is a from-scratch C and CUDA reimplementation of Minecraft 1.11.2, checked against the original Java game. It provides a batched CUDA reinforcement-learning environment in which multiple worlds step in lockstep on one GPU, with semantic-camera observations; the repository reports a recorded batch reaching 7,200 live worlds. Linux x86_64 supports the full build, CUDA training, and Java-oracle workflow, while macOS provides CPU and Metal game backends without CUDA support. The project does not ship Mojang content and requires a legal Minecraft copy and JDK 8.
TurboVLA is a compact vision-language-action model and official implementation for robotic manipulation, developed by researchers at Huazhong University of Science and Technology and Huawei Technologies. Instead of routing visual input through a large language model, it independently encodes visual observations and language instructions, exchanges information through lightweight bidirectional vision-language interaction, and uses a compact decoder to predict continuous action chunks. The repository includes training and evaluation code for LIBERO and RoboTwin, with model checkpoints released separately. The project reports 32 Hz inference on an RTX 4090 with less than 1 GB of VRAM; on LIBERO, it reports a 97.7% average success rate, 31.2 ms inference latency, and 0.2 billion parameters.
SpinWin is a macOS menubar utility that visually rotates or continuously spins another app’s window without private APIs or disabling System Integrity Protection. It hides the selected window off-screen through the Accessibility API, captures its contents with ScreenCaptureKit, and displays the frames in a transparent, borderless overlay whose CALayer transform applies preset 90-degree rotations, arbitrary angles, or continuous spinning. The overlay is sized to the rotated content’s bounding box to avoid clipping and can be dragged or freely rotated. It requires Screen Recording and Accessibility permissions; the overlay is a live, non-interactive picture, and the project notes limitations involving Mission Control, off-screen rendering throttling, and windows that resize themselves.
ctrlb-decompose is a log-processing tool from ctrlb that converts large, repetitive log files into compact, LLM-ready patterns with typed variables, quantiles, anomaly information, severity scores, and example values. Its single-pass pipeline extracts and normalizes timestamps, applies CLP encoding to replace integers, floats, IP addresses, and strings with compact placeholders, clusters log types with the tree-based Drain3 algorithm, and merges wildcard positions with decoded values for semantic typing such as IPv4, UUID, duration, enum, and integer. It accumulates DDSketch quantiles, HyperLogLog cardinality estimates, top-k values, temporal buckets, and reservoir-sampled lines, and provides human-readable and LLM-optimization output modes. The project runs as a CLI, a Rust library, or client-side WebAssembly in the browser; its README states that logs remain on the local machine.
PerceptionBench is a benchmark for evaluating atomic visual perception in multimodal large language models (MLLMs). It isolates perception from reasoning and domain knowledge by defining ten atomic perceptual capabilities, including counting, depth, localization, OCR, and perception-related hallucination, based on an error taxonomy derived from failures across existing benchmarks. The released dataset contains 3,000 verified open-ended questions with short, uniquely determined answers; 1,800 are decomposed from failures in source benchmarks and 1,200 are newly authored using supplementary images. The repository provides the benchmark's code and links to its paper and dataset.