A compact multilingual text-to-speech model with zero-shot voice cloning, available in 0.6B and 0.1B parameter versions for high-efficiency speech synthesis.

A public collection of over 100 ready‑to‑import AI “persona skills” (prompt templates) for agents, covering writers, celebrities, historic texts, social‑media influencers, and multi‑agent teams. Compatible with OpenClaw/StepClaw and the nuwa‑skill format, it lets developers instantly give LLMs a distinct voice or run collaborative agent crews.

Collie is a mobile‑first web UI (PWA) that lets you monitor and interact with AI agents running inside terminal multiplexers (Herdr, tmux, zellij). It works over a private Tailscale network, offering a dashboard, push notifications, quick‑reply commands, special‑key keypad, file attachments, and multi‑machine “crew” support. Installation is a single script; the service runs locally on loopback and requires careful security handling because it provides remote shell access.

An open-source voice-input tool for macOS, Windows, and Linux that uses AI to transcribe and polish spoken text, inserting it directly at the cursor.

A lightweight, CPU-optimized text-to-speech application that enables fast, local audio generation and voice cloning without requiring a GPU.

k‑skill is a collection of 125 command‑line / HTTP “skills” that let AI agents interact with Korean public services, e‑commerce sites, finance portals, and more. Install via a single `npx` command, configure API keys, then call any skill through a lightweight proxy (or directly via Claude Code). It covers travel, real‑estate, law, finance, shopping, health, weather, culture, and developer utilities, providing a Korean‑focused toolbox for building autonomous agents.

Figwright is a free, locally‑run MCP server + Figma plugin that lets AI agents read from and write to Figma files. It provides 113 bidirectional tools (inspection, export, creation, editing, grounding) and two ready‑made agent skills (codegen and build). The whole stack runs on your machine, works with the free Figma tier, and is extensible via custom skills.

An open‑source, Apache‑2.0 licensed fork of the TradingAgents framework that adapts the multi‑LLM analyst‑debate pipeline to China’s A‑share market. It adds three new analyst roles (policy, hot‑money, lock‑up), uses free Chinese data providers, enforces A‑share trading rules (T+1, price limits, lot size), and offers both a CLI and a Streamlit UI. Users configure any OpenAI‑compatible LLM (MiniMax, DeepSeek, Qwen, etc.) with separate quick‑think and deep‑think models, run analyses with a single command, and can later evaluate direction‑accuracy of the generated recommendations.

Natively is a free, source‑available desktop app (macOS / Windows) that captures meeting audio, transcribes it locally with Whisper‑style models, and feeds the transcript to any LLM you provide (including offline Ollama models). It offers real‑time answer suggestions, stealth mode, dual‑channel audio, local RAG via SQLite vector search, persona‑based prompts, and a dashboard for history export—all running on‑device to keep your data private.

garak is an open‑source CLI framework that automatically probes large language models for security‑relevant failures (hallucination, prompt‑injection, data leakage, toxicity, jailbreaks, etc.). It supports Hugging Face, OpenAI, Bedrock, NIM, Replicate, local GGUF, and generic REST models, and provides a large library of ready‑made probes and detectors. Results are shown as PASS/FAIL tables and saved as JSONL logs for deeper analysis.

Tianshu Harness is a TypeScript‑based AI coding‑assistant runtime. It provides a shared core engine for both a terminal TUI and a Tauri desktop GUI, adds a “Cognitive Virtual Machine” with 72 safety hooks, supports multi‑agent orchestration, prefix‑cache for low token cost, Zen/Plan modes, and a star‑domain system for different cognitive styles. Install via a one‑click script, npm, or pre‑built binaries for macOS/Windows/Linux, and use the `rivet` CLI to interact with LLMs for multi‑turn coding tasks.

A state-of-the-art multilingual text-to-speech system that uses a Dual-AR architecture to provide realistic voice cloning and fine-grained emotional control via natural language tags.

PentestGPT is an MIT‑licensed Python CLI that chains LLMs (Claude Code, Codex, or many other providers) into an autonomous penetration‑testing pipeline. It runs staged phases (recon → exploit → walkthrough for CTF, or asset discovery → vulnerability identification → report for real pentests), can persist sessions, and offers a legacy interactive mode where a user guides three cooperating LLMs. Install with `make install` (Python 3.12+, uv) or via Docker, then run `pentestgpt --target <IP>` (or `pentestgpt-legacy` for the interactive version). The tool reports optional anonymous telemetry and achieved ~86 % success on a published benchmark.

Toolcraft is an open‑source starter kit and UI library for building custom, browser‑based design tools (image filters, gradient generators, animation editors, etc.). It ships a React + TypeScript + Tailwind stack, a pan/zoom canvas, a rich component library, layer/timeline support, export utilities, and a set of prompts that let AI coding assistants (Codex, Claude, Cursor) scaffold and extend the app via a simple CLI (`npx @pixel-point/toolcraft create`). The generated project runs locally (default http://localhost:3002) and can be deployed anywhere. MIT‑licensed, cross‑platform, and intended for developers who want AI‑assisted rapid prototyping of niche design workflows.

A family of tiny neural text-to-speech voices (294k to 2.3M parameters) that run in real-time on $3 microcontrollers or in the browser without cloud dependencies.

Agent Client Protocol (ACP) is a language‑agnostic JSON‑RPC standard that lets code editors talk to generative‑AI coding agents. The repository provides the Rust schema crate, versioned JSON‑Schema files, and links to official SDKs (Kotlin, Java, Python, Rust, TypeScript). It defines a stable protocol version (1) negotiated at runtime, enabling editors and agents to interoperate without custom adapters.

Loop Engineering is a CLI framework that lets you design, run, and score autonomous LLM‑agent loops (e.g., daily triage, PR babysitter) around a code repository. It provides ready‑made patterns, a STATE.md persistence file, safety checklists, and a progressive L1‑L3 scoring system to move from report‑only to fully unattended automation.

A fast and local neural text-to-speech engine that provides offline speech synthesis using espeak-ng for phonemization.

Maxun is an open‑source, no‑code web‑data platform that lets you scrape, crawl, and extract information (including from PDFs and images) and output it as clean, AI‑ready formats. It provides visual robots, an LLM‑powered extraction mode, a CLI/SDK, and can be self‑hosted, making it a practical tool for building datasets and APIs for LLM applications.

Autoprompt‑Skill is a Node‑based CLI that wraps multiple LLM coding agents (Claude, Codex, OpenCode, etc.) to automate the write‑test‑review loop. Install via npm, pick a provider, and run `autoprompt activate PROVIDER -- "<goal>"` (optionally with `path=` and concurrency flags). Benchmarks on Terminal‑Bench 2.1 show a 45 % drop in failures (73/89 solves vs. 60/89) at the cost of higher runtime and token usage.

easy‑stock is a desktop Electron app (Go backend + React UI) that pulls A‑share market data, news and forum posts, structures them into a domain model, and lets a locally‑run LLM (Hermes) interpret the information. It offers market overviews, trend‑theme radar, short‑term “连板” analysis, AI‑generated single‑stock reports, portfolio health checks, and automated “big‑V” (influencer) recap. All data and AI conversation history stay on the user’s machine; the app works with any OpenAI‑compatible model and is released under a non‑commercial license.

OpenRSI is an open‑source stack that makes “AI improving AI” a concrete, executable research problem. It includes the Frontis‑MA1 35 B model trained to draft, improve, debug, and recombine code, the OpenMLE full‑stack (Gym for task creation, RL pipelines for learning operators, and Evo for long‑horizon search), and public datasets. Results show sizable gains on MLE‑Bench and NatureBench when the model and search system are combined. The repo provides detailed guides for reproducing training or running a single search, and all code and models are released under a CC‑BY‑NC 4.0 license.

DSH‑Desktop‑EAC is a native desktop client that bundles DeepSeek Harness (the open‑source AI‑agent framework) with a full Node.js runtime, a plugin marketplace, auto‑updates, and UI customisation. It lets users run multi‑agent LLM workflows with a double‑click, sharing the same `dsh` core and data as the CLI while adding safety features like plugin snapshots and rollback.

OpenProgram is a Python framework that turns ordinary functions into LLM‑backed agents. The docstring becomes the system prompt, `llm()` calls the model, and the surrounding logic stays in deterministic code. It provides four core mechanisms – Agentic Functions, a DAG‑based context graph, workflow gates that enforce validation, and a process‑wide event bus – enabling multi‑agent, self‑modifying, and proactive AI assistants. Installable via a one‑line script, it ships with a terminal and web UI, supports Anthropic/OpenAI/Gemini providers, and includes plug‑in harnesses for GUI automation, research workflows, and memory. Licensed AGPL‑3.0 and accompanied by a KDD 2026 workshop paper.

HelixDB is an open‑source Rust database that natively blends graph and vector storage, exposing a language‑agnostic query API for AI agents. It ships with a CLI that can spin up a local instance, scaffold projects, and even hand off to LLM‑based code generators. SDKs for Rust, TypeScript, Python and Go let you build queries as code, which are sent as JSON ASTs to the server. A managed cloud offering adds HA, auto‑scaling readers and integrated full‑text/vector search, making HelixDB suitable for RAG, agent memory, and knowledge‑graph applications.

FyAgent is a cross‑platform desktop app that consolidates the configuration of AI‑assistant tools (model/provider selection, skill/plugin installation, prompt and memory editing, and service credentials) into a single local UI, helping users manage multiple AI agents consistently.

Token Tracker is a local‑only dashboard and desktop app that automatically records how many tokens you spend with 38 AI coding tools (Claude, Gemini, Copilot, Cursor, etc.). It shows usage trends, model‑wise cost estimates, heat‑maps, widgets, a pet companion, achievements, and optional cloud sync/leaderboard – all without requiring an account or sending prompts to any server.

baoyu-design packages Claude Design as a local agent skill, enabling full‑featured UI/UX design (mockups, prototypes, decks, design‑system import, Figma .fig decoding, PPTX export, etc.) directly inside agents like Cursor, Claude Code, or Codex, with all artefacts stored as self‑contained HTML in your repo.

FastGPT is an open‑source platform for building AI agents and LLM‑driven applications. It provides a visual Flow editor, knowledge‑base (RAG) support, plugin hot‑reloading, debugging tools, and can be run locally via Docker or deployed on Sealos/Kubernetes. A hosted SaaS version and a commercial edition are also offered.

A local-first AI subtitling tool that converts audio and video into timestamped subtitles offline, with direct integrations for DaVinci Resolve, Adobe Premiere Pro, and After Effects.

Dograh AI is an open‑source, self‑hostable platform for building and testing voice conversational agents. It provides a visual workflow editor, telephony integrations (Twilio, Vonage, etc.), and a modular LLM/STT/TTS stack that you can replace with any provider. One‑command Docker setup gets the whole system running locally, and Python/Node SDKs let you create agents programmatically. The project is BSD‑2 licensed and backed by a Slack community.

N.E.K.O. is an open‑source AI companion that combines voice, text, and visual understanding with a five‑layer memory system and multiple avatar forms. It runs locally (Python 3.11) and can be deployed via one‑click binaries, Docker, or source build. Features include proactive chat, real‑time multimodal interaction, agent‑style tool execution, a plugin marketplace, and support for 14+ LLM providers. The project is Apache‑2.0 licensed and has active Discord/QQ/Steam communities.

A native macOS application that uses local AI models to provide instant, privacy-focused voice-to-text transcription and a conversational AI assistant.

A documented workflow and set of tools (plugins, prompts, lint rules, sub‑agent orchestration, memory handling) that let developers use Claude Code / Codex as a disciplined coding teammate. Includes a “Superpowers” discipline, parallel sub‑agents, quality‑gate ESLint config, and a full onboarding Playbook.

lat.md is an npm‑distributed CLI that creates a markdown‑based knowledge graph for a codebase. It lets you write linked documentation in `lat.md/`, annotate source files with `@lat` comments, and validates the graph. Agents and developers can search it semantically (offline MiniLM or remote OpenAI/Vercel embeddings) and use commands like `lat check`, `lat locate`, and `lat expand` to keep design decisions, constraints, and test specs in sync with the code.

ShadowBroker is an open‑source, self‑hosted dashboard that aggregates dozens of public OSINT feeds (air‑traffic, maritime AIS, satellite imagery, seismic sensors, radio scanners, Shodan, Telegram, etc.) into a single, real‑time map. Built with Next.js, MapLibre GL, and a FastAPI/Python backend, it offers 40+ toggleable layers, a server‑side recon toolkit, optional Shodan integration, and an HMAC‑signed AI‑agent command channel (compatible with Claude, GPT, LangChain, etc.). An experimental decentralized mesh (InfoNet) provides obfuscated messaging and a governance “Sovereign Shell” economy. Deployment is Docker‑first (or Helm for Kubernetes); no user accounts or telemetry are collected. Ideal for analysts, security researchers, and anyone wanting a unified view of global open‑source intelligence, with optional AI‑augmented analysis.

A set of Claude Code/Gemini‑CLI plugins that turn design knowledge into machine‑readable skills and ready‑to‑run commands, letting AI agents execute full design workflows (research, strategy, UI, interaction, systems, ops, critique) directly from the assistant.

A smart macOS dictation tool that uses Gemini 3.5 Transcribe to turn spoken words into polished text inserted directly at the cursor.

NotFair Plugin is an open‑source library of 45 “skills” that let LLM agents (Claude, Codex, Gemini, etc.) perform concrete SEO, GEO, and paid‑media tasks—audits, keyword research, ad‑budget optimisation, GA4 analysis, and more—by connecting to live marketing platforms through a universal NotFair MCP. It provides per‑task, testable workflows, safe read‑only review before any changes, and can be installed via the plugin marketplaces of several AI‑agent platforms.

Hyper‑Extract is a Python‑based CLI (and library) that uses LLMs with function‑calling to turn PDFs, Word files, markdown, etc., into structured knowledge bases—lists, graphs, hyper‑graphs, temporal graphs, and more. It ships with 80+ zero‑code YAML templates for domains such as finance, law, medicine, and general biographies, and supports incremental updates with full provenance. Works with cloud providers (OpenAI, Anthropic, Gemini, DeepSeek, Alibaba Bailian) or locally‑hosted vLLM models, and can export results to Obsidian, GraphML, CSV, JSON‑LD, or serve them via the Model‑Context‑Protocol. Install with `uv tool install hyperextract` (or pipx), configure a provider, then run `he parse …` to extract and `he search …` to query.

elizaOS is an open‑source TypeScript framework and product stack for autonomous AI agents. It provides a model‑agnostic runtime, a plug‑in architecture, a user‑facing app (web/desktop/mobile), a CLI for scaffolding, optional cloud services, and on‑device inference of Gemma‑based models. The monorepo contains the core runtime, first‑party plugins, UI, and tooling; a separate repo holds bootable Linux/Android distributions. MIT‑licensed.

Skyvern is an open‑source platform that adds LLM‑driven, vision‑based automation to Playwright. It lets developers and non‑technical users control browsers with natural‑language prompts, build reusable workflows, and run either locally (pip or Docker) or via a managed cloud service.

Decepticon is an open‑source, Docker‑based platform that lets LLM‑driven agents run full‑scale red‑team engagements (recon, exploitation, lateral movement, etc.) inside an isolated Kali sandbox. It provides a management plane (LiteLLM, PostgreSQL, Neo4j, LangGraph), on‑demand specialist tools, a web dashboard, and a Python SDK for custom integration. Benchmarked at ~98 % success on the XBOW validation suite, it is licensed under Apache‑2.0 and supported via Discord, live hosted app, and sponsorship.

Pipelock is an open‑source Go proxy that sits between AI agents and the internet, scans HTTP/WebSocket/MCP traffic for secret leaks, prompt‑injection, SSRF and other attacks, and records every decision as a signed receipt that can be verified offline. It offers strict, balanced and audit modes, a free evidence viewer, and an optional enterprise dashboard.

NanoClaw is a minimal, open‑source AI‑assistant framework that runs each agent inside its own Docker container for strong isolation. It installs automatically, registers Anthropic (Claude) credentials via OneCLI’s secure vault, and lets you add chat‑platform adapters (Slack, Telegram, Discord, WhatsApp, etc.) on demand. Agents can be wired per‑channel, shared, or hybrid, and you can schedule recurring tasks. The default model is Claude, but you can swap in OpenAI, OpenRouter, Ollama, etc. Customization is done by asking the built‑in Claude Code AI to edit the tiny codebase, avoiding sprawling config files. Security‑focused, no user accounts, MIT‑licensed.

An open source AI toolkit for building real-time, on-device voice agents and applications with low-latency speech-to-text and text-to-speech capabilities.

A self-hosted personal internet radio station featuring an AI DJ that curates music from your library and provides spoken commentary and transitions.

OpenAlice is a locally‑run AI orchestrator for individual traders. It lets you plug in LLM agents (Claude, Codex, Pi, etc.) and connect them to market data, research workspaces (Git‑backed folders), and broker accounts. Agents can fetch data, run quantitative analyses, generate reports, and even draft trade orders, which you review and commit via a Git‑style interface. The product ships as a desktop app (macOS/Windows) with CLI/Docker options, and includes scheduling, issue tracking, and a central Inbox for follow‑up. Trading execution is marked as beta and intended for paper or test accounts first.