1,038 tools and products — trending open source, and what gets used in AI and other work.
Codex ChatGPT Web is an unofficial launcher that uses browser automation to make ChatGPT Web models, including available Pro models, selectable as native models inside the Codex coding agent. It preserves ChatGPT Web features such as context, images, streaming, and separate usage limits, and its Full harness mode connects ChatGPT tool calls to the current Codex task's files, terminal, tools, and approvals through MCP. A Zero Risk mode lets users paste prepared prompts and send them manually while using a separate MCP connector. The launcher bundles its browser and runtime, supports macOS, Windows, and Linux releases, and is distributed under the MIT license. It is not an OpenAI API or OpenAI-affiliated product; its browser automation depends on the ChatGPT interface and can be affected by UI changes.
Open Slide is a presentation framework built for AI coding agents. An agent describes a deck in natural language and writes React pages, while Open Slide provides a fixed 1920 × 1080 canvas, scaling, navigation, hot reload, presentation and presenter modes, and an in-browser inspector. Its agent skills support deck creation and slide authoring; the inspector records element comments as source markers that an agent can apply. The framework also manages images, video, fonts, and SVG logos, and exports decks as static HTML, PDF, or editable PPTX files. It is distributed as a pnpm/Turbo monorepo with an `@open-slide/cli` scaffolder and `@open-slide/core` runtime, and produces static builds deployable to standard static hosts. The repository is MIT-licensed.
px0 is a browser-based IDE for reviewing AI-generated code, developed by px0-ai. It provides native Git and GitHub integration for pull-request and unpushed-commit review, including scoped diffs, inline draft comments, staging, commits, and pushes. Its navigation tools include fuzzy file search, symbol outlines, workspace-wide regular-expression search, and virtual rendering for large files, while its browser interface can hand edits to local coding harnesses such as Claude Code, Gemini CLI, Cursor Agent, OpenCode, Codex, Aider, and Goose. px0 is distributed as a static Go binary with no runtime dependencies and can run locally or on a remote server, VM, or container; the project is MIT-licensed.
CodexBar is a menu-bar and desktop application by Peter Steinberger for monitoring usage limits, reset times, credits, spending, and status for AI coding and model providers. Its macOS app supports macOS 14 and later, while a Qt-based Linux desktop app and bundled CLI are also available. It reads provider data through existing OAuth or device-flow sessions, API keys, browser cookies, local configuration files, provider CLIs, and other supported local sources, without storing passwords. It presents provider-specific session, weekly, and monthly windows with reset countdowns, usage and spending charts, cost scans, notifications, status indicators, and optional widgets; supported providers include Codex, Claude, OpenAI, Cursor, Copilot, Gemini, OpenRouter, and many others. The project is distributed under the MIT license.
Coucou is an open-source desktop companion for Claude Code and other coding-agent sessions, appearing in the macOS notch or as a top-of-screen island on Windows. It displays session events such as files read, edits, and commands run; surfaces Claude Code permission requests for Allow/Deny approval; opens the relevant terminal on macOS; provides built-in Claude chat; and accepts dropped files as context. It can also attach a macOS window as Claude context and display optional integration pills for services including GitHub, Vercel, Notion, and others. On macOS, a hook script forwards Claude Code events over a Unix socket and waits for approval responses, while a borderless NSPanel and a SwiftUI Canvas render the animated Mochi character. The native Swift 6, SwiftUI, and AppKit application has no third-party dependencies. On Windows, a Tauri 2 application uses Rust, TypeScript, a transparent always-on-top window, Canvas 2D rendering, and a named pipe for hook events. The project states that it collects no telemetry, requires no account, and stores credentials in the macOS Keychain or Windows Credential Manager. Coucou is distributed from the project's releases and can also be built from source. The code is licensed under MIT; the Mochi name, character, icon, sounds, and media are separately reserved by the author. The Windows installer is temporarily unavailable according to the README because its unsigned build was flagged by Microsoft Defender.
DSCODE is a macOS terminal coding agent and CLI built on DeepSeek Harness. Its persistent shell retains the working directory, environment, background jobs, edits, searches, patches, and test runs, while the TUI, CLI, and scripts share one session runtime. Sessions can hand tasks to other sessions, run side questions in read-only child sessions, delegate work to child agents in isolated Git worktrees, and use an independent read-only model to review changes and approve permissions. The `dscode exec` command supports non-interactive scripts and CI, with output streamed to stdout and tool activity and session identifiers sent to stderr. It also supports MCP servers, triggers, cross-session memory, multiple model providers, and a workspace-write sandbox with human-approval and automatic-review modes. Distribution is available through npm, Homebrew, GitHub release tarballs, and source; the project requires macOS 14 or later and is licensed under MIT. It is an independent community project, not an official DeepSeek product.
onetake is a Claude Agent Skill for creating product launch films, teasers, and feature demos as continuous one-take animations rather than sequences of cuts. It rebuilds product interfaces from screenshots in HTML, places them on the camera plane, and uses a motion library of measured spring, entrance, carry, collision, simulation, camera, and fluid-light movements. Its workflow requires a beat sheet in which each boundary identifies what survives into the next beat, then renders deterministic frames with sampled linear-light motion blur and event-based sound. Before a film is shown to a human, probe.py records frame contents, detects beat boundaries, and uses a continuity oracle to check whether visual elements carry across them. The verification also rejects uniform shot timing, missing stillness, clipped audio, fast movement without blur, and subjects leaving the frame. The repository includes rendering, verification, probing, reference-analysis, still-image, footage, narration, and sound scripts, with draft output at 1080p30 and final output at 4K60. It requires Python with Playwright and Chromium, NumPy, SciPy, Pillow, and Matplotlib, plus FFmpeg and Node; narration additionally uses faster-whisper and Kokoro TTS. It is distributed under the PolyForm Noncommercial 1.0.0 license, which permits personal, educational, research, and other noncommercial use but not commercial use.
ReelMimic is a local 2D animation tool that uses Claude Code or Codex to create a new video from a reference video and a brief. It analyzes the reference's shot count, pacing, BPM, transitions, framing, colors, and camera movements, then produces a storyboard, character and asset plan, and style frames for approval. After approval, multiple AI agents build different parts in parallel, while separate agents review completed shots and request fixes with before-and-after screenshots. The tool supports seven drawing engines: vector/motion graphics, hand-painted watercolor, crayon picture book, pixel art, paper cut-out stop-motion, whiteboard doodle, and anime cel; new styles are defined in Markdown. It runs on the user's computer, uses the user's Claude Code or Codex account, and does not reuse the reference video's footage, characters, logos, or assets. The repository is MIT-licensed and has been tested primarily on Windows, with macOS and Linux support described as less tested.
Agent Console is an open-source, local-first observability tool by LockedIn Labs for AI coding agents. It reads Claude Code and Codex transcripts to show token usage by cache reads, cache writes, output and uncached input, model and session activity, list-price cost estimates, burn rate and dollars per hour. A self-hosted hub can combine privacy-limited usage metadata from multiple machines over TLS, while keeping prompts, replies, file paths and file contents local. It also provides optional Claude Code OpenTelemetry and Kong or LiteLLM metric ingestion, a Prometheus endpoint, Grafana dashboard, presenting mode, alerts, and policy hooks. It runs locally from source or release downloads on macOS, Linux and Windows, requires Node.js 22 or newer for the source route, and is licensed under the MIT License.
Rungic is an AgentOS product that runs alongside Android on compatible phones, providing a Linux desktop with KDE Plasma and desktop applications such as Firefox, Blender, Krita and VS Code. The Linux environment runs in an LXC container on Android's kernel rather than in a virtual machine or emulator, while an Android app connects its display, touch input, sound and camera to the phone, a floating window or a TV. Users can install software through apt, Flatpak or Discover and use the phone as a touchpad and keyboard when casting. Rungic includes a bundled assistant whose reference integration uses a realtime voice model for conversation and Codex for background tasks. The assistant plans tasks, opens applications, clicks, types and returns created files in the conversation; its work can run in a separate desktop workspace so the user can continue using the phone. The system supports multiple agent workspaces, shared project folders, desktop control, phone controls and proactive system-care suggestions that record evidence, investigation results, repair plans and user decisions. Rungic exposes these capabilities through two MCP servers, command-line tools, D-Bus services and standard Linux interfaces, allowing other agents to be adapted to it. Rungic is under active development and in private preview. Its device adaptation currently requires a compatible, unlockable Android phone with an appropriate GKI kernel, ARM64 support and a Snapdragon processor with an Adreno GPU; the repository documents tested Motorola devices and separate preparation, Linux image-building and installation stages. Device preparation or bootloader unlocking can erase user data, and the repository advises backing up the phone and checking manufacturer and application restrictions.
Jeeves is a 9B reasoning Jev-style decision model for producing calibrated probabilities for yes/no, multiple-choice, and rating questions. It can support application decisions such as ticket routing, urgency flagging, and customer-frustration scoring through a Jev-compatible API. The model is based on Qwen3.5-9B with LoRA and a pointer head. It uses supervised fine-tuning and CISPO reinforcement learning to reason before deciding, then scores answer options with a pointer head and converts the scores into probabilities. A block-4 diffusion drafter accelerates reasoning-chain generation. The repository includes training and evaluation code, data-preparation scripts, released model weights, an inference server, CUDA and Metal execution paths, FP8 support, and a Python SDK designed as a drop-in replacement for Jev's SDK. The server accepts structured state and question definitions and returns answers, probabilities, confidence values, and optionally reasoning text.
hypoarena is a fully offline scientific hypothesis-discovery workbench and command-line tool. It generates deterministic synthetic literature containing planted causal chains, competing hypotheses, paraphrase clusters, distractors, and contradictions, then runs a generate–debate–evolve loop through pluggable agent adapters that propose, critique, and revise hypotheses. Its pipeline includes span-level citation and claim-grounding checks, paraphrase deduplication using hashing, n-gram Jaccard, TF-IDF, and MinHash LSH, Bradley–Terry/Elo tournaments for ranking hypotheses or agents, hypothesis evolution operators, and Bayesian evidence accumulation. It produces Markdown and self-contained HTML reports with rankings, grounding flags, deduplication clusters, belief updates, run metadata, and limitations. The core runtime depends on NumPy; an optional CPU-only PyTorch extra demonstrates a trainable ranker. Examples and tests run without network access, real model calls, or downloaded weights, and the project is licensed under the MIT License.
iCode is a lightweight, extensible, fully offline terminal development platform and toolkit for orchestrating AI agents and workflows, developed as an open-source project by openJiuwen-ai. Its terminal interface shows the active agent, model, approval mode and session; renders Markdown and diagrams; and provides workflow execution views with per-node models, tool calls and token spending. Agents are configured rather than coded and can include instructions, tools, sub-agents, skills, MCP servers and memory. Model profiles from different providers can be switched during a conversation, while file diffs, turn-based rollback, token and cache usage, model/tool timelines, and an embedded shell expose and control execution. iCode does not include models, telemetry, analytics or crash reporting; it uses only the providers, MCP servers and hooks configured by the user. The repository describes the project as a work in progress, requires uv to run, and distributes it under the Apache License 2.0 alongside separately licensed third-party components.
A Claude Code skill for creating motion-design videos in code using HTML, Playwright, ffmpeg, Python, and Chromium. It computes each frame from a deterministic seek(t) function, renders frames with Playwright, blends subframes for motion blur, and encodes the result with ffmpeg. Its motion system includes closed-form springs, log-space camera keyframes, shared-element handoffs, masked word-by-word text, color floods, and ports of selected 21st.dev components. The workflow moves from inputs and a beat map through still-frame approval, full rendering, pop detection, audio generation, and final muxing. Audio processing finds musical drops by band energy, aligns the song to a key visual frame, places sound effects at measured peaks, and applies two-pass loudness normalization. The repository also includes a remake mode for frame-locked recreations with shot analysis, side-by-side verification, and royalty-free replacement audio. The code is MIT-licensed; downloaded third-party media remains under its respective license.
imajev is an open, locally runnable multimodal typed-decision model and repository maintained by Mohit Garg. It accepts up to two images, application state, and typed questions with user-defined answer options, then returns probabilities for each option together with an unknown probability, confidence, and abstention status. Example tasks include checking a product listing against its photo, comparing a returned item with a reference image, routing support tickets, and evaluating records against policies. Rather than generating an answer, imajev binds each option to a decision code and reads the model's logits at a decision position through a float32 output head. Its vision tower is frozen while LoRA adapters are trained on the language-model projections; the released 2B, 4B, and 9B variants are based on Qwen3.5 models. Unknown is trained as a first-class outcome for missing, contradictory, or out-of-scope evidence, allowing applications to route low-confidence cases to a person. The repository provides a local server and API with MLX and PyTorch paths, calibration files, playground applications, training scripts, and an ImajevBench evaluation suite. The code, adapters, and stated base models are distributed under Apache-2.0. The README describes English-only operation and limits of up to two images, 32 KB of state, eight questions per request, and 254 options; it also cautions that calibration and testing on application-specific data are needed before setting automation thresholds.
Blueprint Animation is a Claude Design skill for presenting UX redesigns and design rationale as continuous blueprint animations. In Redesign mode, it turns the Before screen into a blueprint, animates only the changed parts, and reveals the After screen; in Explain mode, the unchanged screen is converted into a blueprint module by module with guides, dimensions, and labels. Each presentation uses 3–6 numbered steps containing a focus stage, blueprint conversion, construction or annotation, UI reveal, and an explanation of the fix or design decision. The skill reproduces Figma designs using their values, fonts, and exported icons, checks built screens against the Figma export, and includes rules for motion, text placement, UI swaps, performance, and handoff QA. It is distributed as a ZIP upload for Claude Design and is licensed by Oğuz under CC BY-NC 4.0; commercial use requires written permission.
Agent Office is a self-hosted 3D workspace for coordinating coding agents across GitHub repositories. It creates a floor for each project, places Claude Code, Codex, OpenCode and other supported agent workers at desks, streams their terminals, and lets users assign issues, queue tasks, provide Git worktrees, inspect changes and open pull requests. Workers can also list, hire, message and dismiss other workers through the agent-office MCP server or command-line interface. The office includes shared chat, voice, screen sharing and a whiteboard, plus a 2D /lite interface for phones and slower computers. It runs locally or can be deployed to AWS, Azure, Railway, Fly.io, Dokploy or an Ubuntu/Debian server, with multi-user accounts and per-user Claude and GitHub sign-ins. The repository describes it as work in progress, subject to breaking changes, and distributes it under the MIT license.
Naive-N0.5-Flash is an open-weight mixture-of-experts language model from NaiveAI for coding and AI research. It has 309 billion total parameters, 15.5 billion active parameters, and a native one-million-token context window implemented with a hybrid of 128-token Sliding-Window Attention and DeepSeek Sparse Attention; the latter's indexer scans the history while its backbone attends to the top 2,048 selected tokens, with no full-attention layers. The model is released under the MIT License with inference code, and supports FP8 mixed-precision inference through Transformers on FP8-capable NVIDIA GPUs; the repository reports approximately 315 GB of model weights. NaiveAI also describes NaiveRT, an inference system using mega-kernel fusion, Programmatic Dependent Launch, and speculative decoding.
seiso is a Markdown convention and linter for project documentation written by AI and read by people and coding agents. Its convention assigns each document a kind, such as how-to, reference, or ADR; keeps each fact in one canonical location; requires pointers to named files or symbols; separates different concerns; and avoids rapidly changing values in long-lived pages. The reference implementation checks documents against the convention and reports where each violation occurs and how to fix it, including the decision needed when a judgment call is involved. Stable rules run by default, while experimental preview rules require an explicit option; the CLI also provides commands for initialization, rule explanations, parsing, and checks. It can be installed through Cargo, PyPI, npm, or pipx, and integrates with Claude Code hooks, pre-commit, and CI. The project is licensed under MIT.
OpenAI MCP Extensions is an open-source extension framework for adding ChatGPT-specific capabilities to MCP, allowing developers to build plugins that behave as native ChatGPT features. Supported extension points include sidebar entry points, custom file-extension handlers, composer mentions for searching plugin resources and adding references to messages, and extended forms with thumbnail-based choices. The repository provides TypeScript and Python SDKs for MCP servers and Apps, along with plugin documentation and a specification. It is licensed under Apache License 2.0.
Universal Modder is an open-source toolkit by rehan-remade for AI coding agents that create offline, single-player mods for PC games. Its skills and tools identify a game's engine and existing modding route, inspect code and files, back up saves, build a working mod slice, generate art, 3D assets and audio through the fal MCP server, test the result in the running game, package it, and record findings for later agents. The repository supports Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode and other agents that read AGENTS.md, and includes the Python `um` CLI for game scanning, asset generation and processing, Windows automation, save management, video capture, publishing checks and a shared knowledge base of game-specific field notes. Its rules restrict use to games owned by the user and offline or single-player work; they prohibit online cheats, anti-cheat or DRM bypasses, and distributing game files or decompiled code. The repository is MIT licensed and requires Python 3.10+ and ffmpeg; Blender is used for 3D-to-sprite rendering.
Antigravity Fixer is an open-source command-line tool for diagnosing and fixing Google Antigravity, Gemini Code Assist, and related 9router eligibility and login errors. It checks account-related conditions, removes stale credentials and caches, terminates stuck processes, and can revoke OAuth connections before guiding the user through Google sign-in and age verification. It supports the Antigravity CLI and desktop/IDE installations on their platform-specific credential stores and application paths, and includes a clean-only mode for local cleanup. The tool is written in Python, uses Camoufox for headless browser automation and Rich for its terminal interface, prompts for passwords at runtime without writing them to disk or logs, and is distributed under the MIT license.
DoxxNet is an Agentic Defined Network: a private, parallel network for people and software agents to browse, connect devices, call, message, and transfer files. Its P2P communications connect devices directly with end-to-end encryption, while its infrastructure includes private IP space, DNS, and hardware; agents can configure network controls through an API. The service provides encrypted voice and video calls, group and direct messaging, voice memos, on-device translation, and NetDrop transfers for full, unmodified files. It also offers DNS encryption, unwanted-domain blocking, private networking and routing controls, and agent identities with private network permissions. The site states that it does not retain provider-held communication logs and describes the network as zero-log infrastructure.
Fastino Labs is an AI company that builds open-weight language models for agentic AI. Its models are intended for fast type decisions such as intent routing, fact-checking, hallucination checks, and browser-agent tasks, and the video describes them as available for local use and through an inference API. The company’s site identifies GLiNER, GLiGuard, and the Fastino fine-tuning agent among its research and models.
Google DeepMind is an artificial-intelligence research and development organization that builds AI models and systems for general use, science, robotics, speech, image and video generation, music, and coding. Its featured systems include the Gemini family, Gemma open models, Veo video generation, Lyria music generation, Gemini Robotics, and AI systems for scientific discovery. The site also publishes research and provides access to models through products and developer platforms such as Google AI Studio, the Gemini API, and related enterprise tools.
An application programming interface from xAI that provides programmatic access to Grok language models, including Grok 2, for generating and analyzing text through software applications.
MLX is an array framework for machine learning on Apple silicon, developed by Apple machine learning research. It provides Python, C++, C, and Swift APIs modeled on NumPy, plus higher-level packages such as mlx.nn and mlx.optimizers with PyTorch-like interfaces. Its composable transformations support automatic differentiation, automatic vectorization, and computation-graph optimization; computations are lazy, graphs are built dynamically, and operations can run on supported CPUs and GPUs using shared unified memory rather than explicit data transfers. MLX is distributed through PyPI and supports model training and deployment workflows including language-model training and generation, LoRA fine-tuning, image generation, and speech recognition.
Plaud is an AI-powered note-taking platform that consolidates in-person conversations, phone calls, and online meetings into a single workflow, offering automated transcription and summaries. The service is presented at plaud.ai and the site states it has over 2.5 million users.
Trace is an AI-powered browser developed by Breadcromb that creates a private knowledge layer by unifying tabs, documents, and conversations and uses that context to assist the user. The product is described as local-first by default and is offered as a free download.
Open-source Python toolkit that rewrites AI-generated text to read human-written. Ships four documented methodologies — translation chaining, multi-turn LLM rewriting, detection-guided feedback loops, and mixed-engine translation — plus a Standard Pipeline that chains two LLM rewrite passes (DeepSeek/OpenRouter, second pass carries the first as history) with two cross-engine translation hops (EN → ZH → JA → FI → EN across different engines), so no single engine's structural fingerprint survives. Claims 100% key-information retention on 50 verified text pairs; the repo warns detector scores are probabilistic and against misrepresenting authorship.
An open-source set of reinforcement-learning training environments for Pollen Robotics' Microduck bipedal robot, built on mjlab with MuJoCo Warp and PPO. It trains walking, standing, recovery, sitting, ground-picking, kicking, rolling, and roller-skating policies, with flat, rough, slope, and backlash variants where applicable. The environments encode a sim-to-real pipeline using BAM actuator physics for Dynamixel XL330 servos, domain randomization for battery voltage, voltage sag, command delay, and friction, and simulated gear backlash with output-side encoder feedback. Policies use a shared 61-dimensional observation contract for runtime hot-swapping, are trained at 50 Hz, exported to ONNX with the observation normalizer embedded, and deployed by the Microduck runtime. Training runs locally through a CUDA GPU or can be submitted to Hugging Face Jobs; the project is licensed under Apache 2.0, while its 3D model files use CC BY-SA-NC.
ODS is a self-hosted local AI server stack developed by Osmantic for Linux, Windows with WSL2 and Docker Desktop, and Apple Silicon macOS. Its installers detect the host hardware, select an installable model, generate credentials, and launch a preconfigured stack containing local LLM inference, Open WebUI, a management dashboard, LiteLLM, embeddings, speech-to-text, text-to-speech, agents, n8n workflows, Qdrant-based RAG, search, ComfyUI image generation, and operational and privacy tools. Local mode is the default; cloud and hybrid modes can route through OpenAI, Anthropic, or Together APIs. The stack uses a bootstrap model so chat can begin while the hardware-selected full model downloads in the background. A manifest-based extension system discovers service folders and merges their Docker Compose definitions into the stack, while the `ods` CLI manages services, models, modes, configuration, presets, and lifecycle operations. The repository documents GPU and unified-memory paths for NVIDIA, AMD, Intel Arc, and Apple Silicon hardware, along with model selection, recovery, validation, and operator workflows. ODS is distributed as an Apache-2.0-licensed open-source repository and can be installed with shell or PowerShell installers. The repository identifies version 2.6.0 as its stable release and states that no cloud account or subscription is required for local operation.
video-use is an open-source video-editing tool for coding agents such as Claude Code, Codex, Hermes, and Openclaw. It takes raw footage in a folder and produces an edited MP4, with support for removing filler words and dead space, applying FFmpeg color-grading chains, adding short audio fades at cuts, burning customizable subtitles, and generating animation overlays through HyperFrames, Remotion, Manim, or PIL. Its pipeline transcribes each source with ElevenLabs Scribe, packing word-level timestamps, speaker diarization, and audio events into a Markdown view. The agent reasons over that transcript and requests filmstrip, waveform, and word-label composites from timeline_view only at decision points, rather than processing every video frame as visual context. It then creates an edit decision list, renders the result, evaluates the rendered video at every cut boundary for visual jumps, audio pops, and subtitle problems, and can revise and re-render up to three times. The tool uses an approval-oriented workflow in which the agent inventories sources, proposes an editing strategy, waits for confirmation, executes the edit, evaluates it, and persists session memory in project.md. It requires FFmpeg and an ElevenLabs API key; yt-dlp is optional for downloading online sources. Outputs are placed in an edit directory next to the source footage.
prompts.chat is a curated collection and community-maintained library of prompts for AI chat models, originally known as Awesome ChatGPT Prompts. Users can browse prompts, submit additions through prompts.chat, and access the collection as a website, CSV file, Markdown document, or Hugging Face dataset; submitted prompts sync to the repository automatically. It supports use with ChatGPT, Claude, Gemini, Llama, Mistral, and other modern AI assistants. The project also provides an interactive prompt-engineering book, a game-based prompting experience for children, a CLI, a Claude Code plugin, and an MCP server. Its open-source application can be self-hosted with custom branding, themes, authentication, and PostgreSQL; the repository documents setup through `npx prompts.chat new` or a manual clone and installation. Source code and site-authored content use the MIT License, while prompt content and data are dedicated to the public domain under CC0 1.0 Universal.
Everything Claude Code is a Claude Code plugin and toolkit containing configurable agents, skills, slash commands, rules, hooks, contexts, MCP server configurations, scripts, and tests for AI-assisted software development. Its agents delegate tasks such as planning, code review, security review, testing, and build-error resolution. Skills define workflows including test-driven development, security review, continuous learning, and verification; hooks run on Claude Code tool or session events to load and save context, suggest compaction, and evaluate sessions; and rules provide persistent guidance for security, coding style, testing, Git workflow, delegation, and performance. Node.js scripts support Windows, macOS, and Linux and detect a project's preferred package manager from environment settings, project configuration, package.json, lock files, or global configuration. The repository can be installed as a Claude Code plugin or copied into Claude configuration directories manually. It includes a Node.js test suite and is distributed under the MIT license.
Marketing Skills is a collection of Markdown-based skills for AI coding agents, built by Corey Haines. It provides specialized workflows for conversion optimization, copywriting, SEO, analytics, advertising, growth engineering, sales, and related marketing tasks, and supports Claude Code, OpenAI Codex, Cursor, Windsurf, and agents implementing the Agent Skills specification. The skills share context and cross-reference one another. The product-marketing skill serves as the foundation, with other skills consulting it for product, audience, and positioning information before handling tasks such as CRO, copywriting, SEO audits, analytics setup, email flows, pricing, or sales enablement. Skills can be installed individually or as a collection through the Skills CLI, Claude Code's plugin system, a repository clone or submodule, or SkillKit. The repository is built and maintained through community contributions, with issues and pull requests welcomed. It is free and MIT-licensed.
Camofox Browser is an open-source anti-detection headless browser server for AI agents, built by the team behind jo and powered by Camoufox, a Firefox fork that applies fingerprint spoofing at the C++ level. It exposes browser automation through a REST API with accessibility snapshots, stable element references for clicking and typing, navigation and search macros, screenshots, downloads, file uploads, structured extraction, YouTube transcript extraction, and OpenAPI documentation. Browser contexts provide isolated cookies and storage, with optional persistence, cookie import, proxy and GeoIP configuration, VNC-based interactive login, and Playwright session tracing. It can run from npm, Docker, Fly.io, or Railway, and supports lazy browser startup and idle shutdown. The repository is licensed under MIT.
DeerFlow is an open-source long-horizon super-agent harness developed in the ByteDance repository. Built on LangGraph and LangChain, it orchestrates a lead agent with sub-agents, progressively loaded skills, memory, tools, and sandboxed execution environments for research, coding, file work, report generation, slide creation, web-page creation, and other multi-step tasks. The runtime provides a filesystem for uploads, workspaces, and outputs; configurable web search, web fetching, browser control, Bash and file tools; MCP servers; reusable Markdown-based skills; long-term memory; scheduled tasks; and messaging-app channels. Sub-agents receive scoped context, tools, and termination conditions, and can run bounded parallel work before returning structured results to the lead agent. Sandbox modes include local execution, isolated Docker containers, Kubernetes-backed execution, and E2B-based environments, depending on configuration. DeerFlow can run locally or with Docker Compose and exposes a Web UI and Gateway at the documented local address. Its setup wizard generates configuration for model providers, search, sandbox, Bash, and file-writing options; the project also documents integrations for LangSmith, Langfuse, and Monocle tracing, personal access tokens, custom agents, MCP background tasks, and extensible Python packages. Version 2.0 is a ground-up rewrite of the earlier 1.x Deep Research framework; active development is on the 2.0 line.
Lightpanda Browser is a headless browser built from scratch in Zig for AI agents and web automation. It is not based on Chromium, Blink, or WebKit; its implementation includes an HTML parser, V8 JavaScript support, a DOM tree and APIs, network loading, cookies, form input, proxy support, network interception, and optional robots.txt compliance. It can dump pages as HTML, Markdown, PNG, or PDF, or expose a Chrome DevTools Protocol and WebDriver BiDi server for clients such as Puppeteer and Playwright. Its native agent mode uses an LLM to navigate pages, click through flows, fill forms, and extract structured data. Sessions can be exported as deterministic, token-free PandaScript JavaScript and replayed without a model at runtime. The browser also provides an MCP server over stdio or HTTP, with isolated or shared browsing sessions for multiple agents. The project distributes nightly binaries for Linux and macOS, Docker images for Linux amd64 and arm64, and packages for Homebrew and the Arch User Repository. It can also be built from source with Zig and its V8, Libcurl, and html5ever dependencies.
Andrej Karpathy Skills is a set of guidelines for Claude Code and Cursor that configures coding agents to avoid common LLM coding pitfalls. Its four principles are Think Before Coding, which requires explicit assumptions, clarification of ambiguity, and stated trade-offs; Simplicity First, which limits speculative features and unnecessary abstractions; Surgical Changes, which restricts edits to the requested scope; and Goal-Driven Execution, which turns tasks into verifiable success criteria and test-driven loops. The guidelines can be installed as a Claude Code plugin, copied into a project's CLAUDE.md file, or applied through the repository's Cursor rule. The repository is licensed under the MIT License.
Browser Use is an AI browser-automation framework and hosted service for assigning web tasks to agents. Its agents operate websites through a browser by opening pages, clicking controls, entering text, filling forms, and extracting structured data. The project provides a Python library for embedding browser agents in applications, a CLI for agent-driven one-off tasks, and a Browser Use Cloud API with hosted agents and managed browser infrastructure. The library supports different LLM providers, custom tools, structured output, reusable browser profiles, and parallel or scheduled automation. The open-source library is free to use and licensed under the MIT License; the hosted service adds scalable browser infrastructure, proxy rotation, stealth browser fingerprinting, CAPTCHA handling, persistent storage, and integrations.
TradingAgents is an open-source, multi-agent LLM financial trading framework developed by TauricResearch for research. It models a trading firm through specialized agents: fundamental, sentiment, news, and technical analysts produce market assessments; bullish and bearish researchers debate those assessments; a Trader Agent proposes the timing and size of a trade; and risk-management and portfolio-management agents evaluate and approve or reject it before execution on a simulated exchange. The framework uses LangGraph and can be run through an interactive CLI, Docker, or as a Python package via the TradingAgentsGraph class and its propagate function. It accepts market and ticker inputs, analysis dates, configurable research depth and debate rounds, and LLM backends including hosted providers, Ollama, and other OpenAI-compatible servers. It gathers financial, market, news, macroeconomic, and social-sentiment data, and supports persistent decision logs and optional LangGraph checkpoint recovery. TradingAgents is intended as a research scaffold rather than financial, investment, or trading advice. Results can vary with the language model, sampling settings, data sources, analysis period, and other nondeterministic factors.
LLM Wiki is a cross-platform desktop application maintained by nash_su that turns imported documents into a persistent, organized, interlinked knowledge base. It uses a three-layer structure of immutable raw sources, LLM-generated wiki pages, and configuration rules, with ingest, query, and lint operations; generated pages use YAML frontmatter, source traceability, an index, an operation log, and [[wikilink]] cross-references. Its ingest pipeline first analyzes each source for entities, concepts, connections, contradictions, and recommended structure, then generates or updates wiki pages, summaries, indexes, research queries, and human-review items. SHA256 caching skips unchanged files, while a persistent queue supports serialized processing, retries, cancellation, and crash recovery. It accepts PDFs, Office documents, EPUB/MOBI, Org mode, images, media, web clips, and URL batches, with optional vision-based image captions and MinerU processing for complex PDFs. Retrieval combines tokenized keyword search, optional LanceDB vector search through OpenAI-compatible embedding endpoints, and graph expansion based on direct links, shared sources, Adamic-Adar similarity, and page-type affinity. The application also provides Louvain community detection, graph insights, deep research through Tavily, SerpApi, or SearXNG, a Rust tool-using chat agent, review workflows, Mermaid and KaTeX rendering, a Chrome web clipper, and project export/import. It includes a token-protected local HTTP API and MCP server for searching, reading files, traversing the graph, rescanning sources, and calling the backend agent. The application is built with Tauri v2, React, TypeScript, and Rust, has macOS, Windows, and Linux builds, and is licensed under GPL-3.0.
Skills is a command-line tool for the open agent skills ecosystem. It installs reusable instruction sets defined in SKILL.md files into supported coding agents such as OpenCode, Claude Code, Codex, Cursor, and others. The `npx skills` CLI resolves skills from GitHub, GitLab, other Git URLs, local paths, or direct file and archive URLs; lists available skills, selects individual skills, and installs them project-wide or globally by symlink or copy. It can also generate a prompt for a skill without installing it, launch a supported agent with that prompt, search for skills, update or remove installations, and create new SKILL.md templates. Skills use YAML frontmatter with a name and description, and the CLI discovers them in standard agent and repository skill directories. The project is licensed under the MIT License.
Higress is a cloud-native API gateway and AI gateway developed as a vendor-neutral CNCF project. Based on Istio and Envoy, it supports Kubernetes Ingress and Gateway APIs, microservice discovery, security controls, unified LLM API management, and hosting MCP (Model Context Protocol) servers through plugins. The gateway is extended with WebAssembly plugins written in Go, Rust, or JavaScript. Its plugin mechanism supports streaming request and response processing, authentication and authorization, rate limiting, audit logging, observability, WAF features, AI model routing, token rate limiting, caching, and conversion of OpenAPI specifications into remote MCP servers. Higress provides an out-of-the-box console and can run with Docker or be deployed on Kubernetes using Helm.
CowAgent is an open-source AI assistant and agent harness developed by LinkAI. It plans complex tasks step by step, runs tools and skills, controls computers and external services, maintains long-term memory and a personal knowledge base, and supports multi-agent teams with separate roles, models, skills, knowledge, and workspaces. Its architecture routes messages from web and messaging channels through an agent core that reasons over available models, tools, skills, memory, and knowledge before returning a response. Memory uses a three-tier structure of conversation context, daily memory, and long-term MEMORY.md, with a nightly Deep Dream process that distills memories; the knowledge base is an automatically curated Markdown wiki with cross-references and a browsable knowledge graph. Built-in tools include file and terminal operations, web fetching and search, scheduling, memory retrieval, vision, and browser automation, while MCP servers can be added through configuration. Skills can be installed from the Skill Hub, GitHub, ClawHub, or URLs, or authored through conversation. CowAgent supports multiple LLM providers and routes chat, vision, image generation, speech, and embedding workloads separately. It can run locally, in Docker, on a server, or through its macOS and Windows desktop client, with a web console for configuration and a command-line interface for service and skill management. The project is licensed under the MIT License and was formerly named chatgpt-on-wechat.
JeecgBoot is an open-source enterprise AI low-code development platform. It combines online form, workflow, report, dashboard, portal, and application builders with an AI application platform for model management, chat, knowledge bases, workflow orchestration, MCP plugins, and other AI applications. Its AI Skills workflow accepts natural-language requirements and can generate database tables, SQL, front-end and back-end CRUD code, menu permissions, and rendered pages for a runnable business system. The platform also provides a conventional code generator for single-table, tree, one-to-one, and one-to-many data models, allowing generated code to be configured online and then manually merged and extended. Its knowledge-base features use large language models and RAG for document parsing, vector-store integration, retrieval, and question answering. The platform uses a separated Vue 3, TypeScript, Vite, and Ant Design Vue front end with a Java back end based on Spring Boot, Spring Cloud Alibaba, and MyBatis-Plus. It supports monolithic and microservice deployments, granular button, data, and field permissions, enterprise modules such as users, roles, organizations, messaging, and scheduled tasks, and integrations including Flowable, JimuReport, and JimuBI. JeecgBoot is distributed under the Apache License 2.0.
Fay is an open-source digital-human agent framework for connecting 2D, 3D, mobile, desktop, and web avatars—or OpenAI-compatible and DeepSeek large language models—to business systems. It provides interfaces for text and voice interaction, avatar control, administration, automatic broadcasting, and intent handling, and can connect interchangeable avatar, ASR, TTS, and language models. The framework supports offline operation, streaming interaction, multi-user concurrency, configurable voice commands, custom knowledge bases and personas, wake-word activation and interruption, scheduled proactive dialogue, agent tool calling, MCP tool management, robot-expression output, and server or standalone deployment. It runs from source with Python on Windows, macOS, and Ubuntu and includes a local management page.
LangBot is an open-source platform for building and operating AI-powered instant-messaging bots. It connects large language models to platforms including Discord, Telegram, Slack, LINE, QQ, WeChat, WeCom, Lark, DingTalk, KOOK, and Matrix, supporting multi-turn conversations, tool calling, multimodal input, streaming output, and agent workflows. The platform includes a built-in retrieval-augmented generation knowledge base, integrations with systems such as Dify, Coze, n8n, Langflow, and Deerflow, and an event-driven plugin system with MCP support. A browser-based management panel configures, manages, and monitors bots, while its multi-pipeline architecture supports separate bot workflows and operational monitoring. LangBot can be run through LangBot Cloud, a one-line uvx launch, Docker Compose, or other deployment methods. It also exposes an MCP endpoint that mirrors its HTTP API for programmatic management of bots, pipelines, plugins, and models.
claude-red is a curated library of offensive-security skills for Claude, maintained by SnailSploit. Each skill is a structured SKILL.md file covering a particular attack surface, including web vulnerabilities, identity systems, wireless, cloud, mobile, IoT, exploit development, fuzzing, reconnaissance, API security, containers, cryptography, privilege escalation, post-exploitation, and AI security. The files are intended to be installed into a Claude Skills environment, where matching skills load on demand from conversational triggers, or they can be supplied manually as system files or project prompts. The repository includes an installation script with category and target options, a skill index, a roadmap, contribution guidance, and a review-oriented skill template. It is distributed under the MIT license and lists authorized red-team engagements, bug-bounty triage, security research, CTF preparation, and operator training as use cases.
PentAGI is a self-hosted, AI-powered penetration-testing platform for security professionals, researchers, and ethical hackers. It uses a multi-agent system with specialized researcher, developer, executor, adviser, planner, and reflector roles to investigate targets, plan and execute testing tasks, operate professional security tools, and generate vulnerability reports. The platform runs operations in isolated Docker containers and includes more than 20 security tools, an isolated web scraper, external search integrations, PostgreSQL with pgvector for persistent results and semantic memory, and an optional Graphiti/Neo4j knowledge graph. Its agents use long-term, working, and episodic memory; chain summarization converts conversation chains into a structured ChainAST, summarizes older sections and oversized pairs, and rebuilds the chain when the result is smaller. Optional execution monitoring detects repeated or excessive tool calls and invokes a mentor for alternative strategies, while task planning decomposes a request into actionable subtasks before specialist agents execute it. PentAGI provides a React and TypeScript web interface, Go-based REST and GraphQL APIs with Bearer-token authentication, flow-scoped file management, Markdown and PDF report downloads, and monitoring integrations built around OpenTelemetry, Grafana, VictoriaMetrics, Jaeger, and Loki. It supports multiple hosted and local LLM providers, including OpenAI, Anthropic, Google Gemini, AWS Bedrock, Ollama, DeepSeek, GLM, Kimi, Qwen, and MiniMax, and can be deployed with Docker Compose or Podman. The project states that it is an autonomous and assistant-guided penetration-testing platform rather than a CALDERA-style breach-and-attack-simulation system with predefined campaigns; testing should be limited to systems for which the operator has authorization.
Agent Skills is a curated skill registry and npm-based CLI for extending AI coding agents such as Claude Code, Cursor, Copilot, and Antigravity with packaged instructions, templates, and reference files. The CLI fetches a catalog, lets users search and select skills, caches downloads locally, and installs them to selected agents by copying or symlinking them at global or project scope; it also supports updating, removing, auditing, and managing cached skills. The repository describes the catalog as an open-source, human-curated library with CI static analysis, content hashing, lockfiles, path isolation, symlink guards, sanitization, audit logging, and Snyk Agent Scan checks. It also includes an MCP server that exposes progressive-disclosure tools for listing, searching, reading, and fetching skill files. The CLI source is MIT-licensed; maintainer-authored skills are generally CC BY 4.0, while third-party skills retain their own licenses.
Transformers is Hugging Face's open-source model-definition framework and Python library for inference and training with pretrained text, computer vision, audio, video, and multimodal machine-learning models. It centralizes model definitions behind a unified API and provides the Pipeline interface, which preprocesses inputs and returns task-specific outputs for applications such as text generation, automatic speech recognition, image classification, and visual question answering. Models can be downloaded and cached from the Hugging Face Hub, and the library supports moving models between PyTorch, JAX, and TensorFlow workflows while exposing model internals for customization. It requires Python 3.10 or later and PyTorch 2.5 or later for the documented installation.
VoxCPM is an open-source tokenizer-free text-to-speech system from OpenBMB for multilingual speech generation, voice design, and voice cloning. Its latest major release, VoxCPM2, directly generates continuous speech representations in the latent space of AudioVAE V2 through a four-stage diffusion-autoregressive pipeline: LocEnc, TSLM, RALM, and LocDiT. It accepts text, optional natural-language voice descriptions, or reference audio with transcripts, and produces streaming or batch speech audio; VoxCPM2 supports 30 languages, voice-style controls, controllable and continuation-based cloning, and native 48 kHz output. The project provides a Python package and command-line interface, a local web demo, fine-tuning through full supervised training or LoRA, and serving options including an OpenAI-compatible vLLM-Omni endpoint. Its code and model weights are released under the Apache-2.0 license. The project warns that generated voices can be misused for impersonation, fraud, or disinformation, and notes that voice-design and controllable-cloning results can vary between runs.
Atlas is a desktop application for source control and shared context across coding-agent sessions, developed by Pacifio. It runs Claude Code, Codex, its native agent, and agents from the ACP registry against a codebase, recording prompts, tool calls, file changes, and patches in local session storage. When a commit is created, Atlas links it to the session that produced it, preserving those checkpoint relationships through rebases and amends and allowing the session to be queried later. Before sending a prompt, Atlas resolves local @ mentions, combines project notes and agent memory, performs on-device semantic retrieval, and can add a handoff from a previous session. Markdown knowledge files, CLAUDE.md, AGENTS.md, plans, decisions, failures, file changes, skills, papers, and past sessions can therefore be shared between agents. The application also provides a code editor, Git operations and diffs, a terminal, a project knowledge base, research tools, a browser, spatial boards, and agent activity history. Atlas is local by default: code, notes, and sessions remain on the machine, secrets are scrubbed before persistence, and local mode works without an account or network. Optional organisations can sync across devices and teammates. The repository states that macOS is the supported platform; Linux and Windows builds are untested. It is distributed under the MIT license.
Voicebox is a local-first, open-source AI voice studio developed in the jamiepine/voicebox repository. It combines voice cloning, text-to-speech, speech-to-text dictation, voice profiles, audio effects, and agent voice output in a desktop application. It generates speech through seven switchable TTS engines, including Qwen3-TTS, Qwen CustomVoice, LuxTTS, Chatterbox, HumeAI TADA, and Kokoro, and supports voice cloning from reference audio, preset voices, multilingual generation, automatic text chunking, crossfading, and multi-track story editing. Whisper-based transcription powers global dictation, in-app recording, and the transcription API; an optional local Qwen3 LLM can refine dictation or rewrite text according to a voice profile's persona. The application exposes REST endpoints for speech generation, agent speech, transcription, and profile management, plus a built-in MCP server with tools for speaking, transcribing, and browsing captures and profiles. It runs models and stores voice data locally, using Tauri with a React/TypeScript frontend and FastAPI backend; platform backends include MLX on Apple Silicon and PyTorch on CUDA, ROCm, XPU, DirectML, or CPU. The repository is licensed under the MIT License and provides macOS, Windows, and Docker distribution, with Linux build-from-source instructions.
AstrBot is an open-source all-in-one AI agent chatbot platform and development framework that connects instant-messaging platforms with LLM services, plugins, multimodal features, MCP, skills, knowledge bases, persona settings, and automatic context compression. It supports platforms including QQ, Telegram, WeCom, Feishu, DingTalk, Slack, Discord, LINE, and others, and can connect to services such as OpenAI-compatible APIs, Anthropic, Gemini, Ollama, LM Studio, Dify, Coze, and Alibaba Cloud Bailian. Its plugin system offers one-click installation of more than 1,000 plugins, while its agent sandbox isolates code and shell execution and supports session-level resource reuse. AstrBot provides a WebUI and web ChatUI with agent sandbox and web-search support, and can be deployed with uv, Docker or Docker Compose, desktop applications, or other documented deployment methods.
WeChat Bot is an open-source IM agent and command-line tool based on Wechaty. It connects WeChat QR-code logins, Lark events, Telegram's Bot API, and WhatsApp Cloud API to AI services including ChatGPT, DeepSeek, Ollama, Claude, Pi, and other configured providers for message replies and analysis. The project routes received messages through a pipeline that captures WeChat messages locally in JSONL, sends them to a selected model or agent, and posts the response back to the IM channel. Its `wb wx` commands use OpenCLI wx-cli to access local WeChat chats, contacts, group members, favorites, and cached Moments; `wb stats` and `wb analyze` provide local statistics or AI-assisted group and friend analysis. Replies can be restricted with contact and group allowlists, prefixes, and group-mention rules, and non-text messages are not automatically sent to the reply pipeline. The repository is a Node.js project requiring Node.js 18 or later and can be installed with npm or run in Docker. It is licensed under the MIT License. The README warns that WeChat Web protocols can trigger account warnings or bans and recommends narrow allowlists and usage scopes.
Supermemory is an AI memory and context engine with a hosted API, application, plugins, MCP server, and local runtime. It extracts facts from conversations, tracks temporal changes and contradictions, automatically forgets expired information, and maintains user profiles containing stable facts and recent activity. Its API combines memory retrieval with hybrid RAG search, allowing applications and AI agents to query personalized context and knowledge-base documents together. It also processes uploaded PDFs, images, videos, and code, and can synchronize data from services such as Google Drive, Gmail, Notion, OneDrive, GitHub, and web pages. SDKs are available for JavaScript and Python, with integrations for AI frameworks including the Vercel AI SDK, LangChain, LangGraph, OpenAI Agents SDK, Mastra, Agno, and n8n. Supermemory can provide persistent memory to compatible assistants through plugins or its MCP server. The local distribution runs as a single binary or through npx, uses local embeddings by default, supports OpenAI-compatible and Ollama-backed models, and exposes the same Memory API on localhost for offline or self-hosted use.
An independent MCP server and CLI that connects Claude Code or other MCP clients to a locally running TradingView Desktop application for chart analysis and workflow automation. It communicates through Chrome DevTools Protocol on a user-enabled localhost debug port, rather than connecting to TradingView servers or intercepting network traffic. The bridge exposes tools for reading quotes, OHLCV data, indicator values, Pine Script drawings and tables; changing symbols, timeframes, indicators, panes and tabs; injecting, compiling and debugging Pine Script; drawing chart objects; managing alerts and watchlists; using replay mode; capturing screenshots; and streaming locally read chart data as JSONL. Its CLI provides equivalent pipe-friendly commands, while MCP uses a stdio transport. Data processing remains local, and the project states that it does not execute real trades or bypass TradingView access controls. It requires the installed TradingView Desktop app, a valid TradingView subscription, Node.js 18 or later, and an explicitly enabled Chrome DevTools debug port. Because it relies on undocumented internal TradingView application interfaces, updates to TradingView Desktop may break compatibility. The source code is MIT-licensed, but the license does not grant rights to TradingView software, data or trademarks.