Hope Agent is an open‑source, Rust‑based desktop AI assistant that can run as a native app, a headless server with a web UI, or an IDE‑integrated backend. It offers goal‑driven task automation, persistent memory, design‑space generation, and full computer control (file, shell, browser) with strong security (local‑first data, tool approval, Docker sandbox). Installable on macOS, Windows and Linux via package managers or Docker, it ships with dozens of LLM provider templates, multi‑platform IM integrations, and a plug‑in skill system.

A Python library for embedding and detecting imperceptible watermarks in audio files, utilizing neural network-based approaches to ensure robustness against manipulation.

Arkloop is an open‑source, local‑first AI agent platform. It bundles a Go backend, SQLite storage, and an Electron‑based desktop app (with a React web UI) that supports multi‑model routing, tool calling, persistent or semantic memory, and integration with Telegram, Discord, QQ, Feishu, and WeChat. Installable via GitHub releases, Homebrew, or AUR, it runs entirely on the user’s machine, avoiding any cloud infrastructure.

ai4j is a Java‑first SDK (JDK 8+) that unifies access to many LLM providers and adds agentic features (tool calling, multi‑agent coordination, A2A, RAG). It ships core libraries, an agent runtime, a coding‑assistant CLI/TUI/IDE bridge, Spring Boot starter, and a plugin system. Install via Maven/Gradle, configure a provider (e.g., OpenAI), and you can call chat/completion APIs in a few lines of code or run the built‑in coding agent locally.

Ollama4j is a Java library that wraps the Ollama local LLM server, providing text generation, chat, tool calling, embeddings, image inputs, model management, and Prometheus metrics. It is distributed via Maven Central/Gradle and integrates with any Java project.

Bert-VITS2 is a text-to-speech system that integrates a VITS2 backbone with multilingual BERT to produce high-quality, natural speech synthesis.

Flounder is an open‑source framework that lets LLM coding agents run autonomous, sandboxed security audits. Given a repo, contract address, transaction hash, or similar clue, it prepares a workspace, maps the attack surface, lets the model write and execute PoC tests in an OCI sandbox, verifies findings locally, optionally confirms them against a real target, and generates markdown reports. It supports many scenarios (blind audits, incident investigations, bug‑bounty work) and works with various model providers via a daemon‑based control plane.

glm‑coding‑helper is a Tampermonkey userscript plus a local CPU/GPU OCR server that automates the frantic “rush” purchase of Zhipu GLM Coding Plan subscriptions. It enables early button clicks, cycles through plan options, solves Chinese point‑selection captchas locally (using YOLO + PP‑OCRv6), and offers configurable auto‑click, multi‑window, and safety settings. The project provides portable Windows builds and online installers for macOS/Linux, with a one‑click start script that sets up a Python environment and launches a FastAPI OCR service.

PyOpenCL is a Python library that provides full, Python‑friendly access to OpenCL GPUs and other parallel devices. It offers RAII‑style resource cleanup, automatic error‑to‑exception conversion, and fast C++ bindings, with cross‑platform binary wheels and an MIT license, making it a practical tool for custom GPU‑accelerated code in AI and scientific computing.

Claude Night Market is a plugin marketplace that extends Anthropic’s Claude Code IDE with 23 safety‑guarded plugins for git workflows, test‑driven development, spec‑driven development, autonomous agents, and many domain‑specific utilities. Install the marketplace, pick the plugins you need, and run high‑level commands like `/attune:mission` to manage the full feature lifecycle inside Claude Code.

aicommit2 is a CLI that automatically creates Git/YADM/Jujutsu commit messages using a wide range of LLM providers, with features like multi‑provider streaming, diff compression, code review, hook integration, and LazyGit support.

A collection of samples for Azure Cognitive Service Text-to-Speech, enabling developers to integrate natural-sounding AI voices into their apps via the Speech SDK or REST API.

DSH Taskboard is a DeepSeek Harness plugin that adds a native SQLite‑backed task‑board UI, a JSON CLI, and a set of agent tools for managing projects, tasks, and automations. Humans approve and accept work; agents claim, work on, and submit tasks for review. Installation requires building the plugin, adding it to a Harness profile, and restarting the Harness server.

A macOS‑only Swift tool that records mic and system audio locally, then transcribes both tracks on‑device using a Core ML speech model, producing speaker‑tagged transcripts without sending any data to the cloud.

A singing voice conversion tool that enables real-time voice changing and high-quality voice cloning with an improved GUI and faster training.

Bitterbot is a local‑first AI agent that keeps a biologically‑inspired, persistent memory, runs autonomous “dream” cycles to consolidate knowledge, can execute tools (web browsing, code, WhatsApp), and optionally participates in a P2P skill marketplace for USDC. Install via Node ≥ 22, run the onboarding wizard, and interact through a browser UI at http://127.0.0.1:19001.

PMB is an open‑source Python library that gives LLM‑based coding assistants a local, SQLite‑backed memory. It automatically indexes code, PDFs, git history, and user‑recorded facts, then injects the most relevant pieces back into the agent via the Model‑Context‑Protocol. The system works offline, offers a hybrid BM25‑+‑vector search (≈35 ms warm latency), provides a visual dashboard, and includes hooks that both recall memory before the model thinks and auto‑write observed actions. Install with `pip install pmb-ai`, run `pmb setup` to wire an agent, and the agent will remember decisions, lessons, and project context across restarts without any API keys or cloud calls.

Atomic Agent is an open‑source, local‑first AI assistant that runs on your machine using quantised models via `llama.cpp`. It executes tool‑call JSON loops to browse, edit files, run shell commands, manage memory, and more, all while keeping state on disk and offering a TUI/CLI and HTTP side‑car for integration.

EvoTrace is a local‑first tool that imports Claude Code or Codex session histories, mines coherent coding episodes, and compiles them into verified, Docker‑ready tasks, preference data, and reward candidates for AI training and evaluation. It runs on the DeepSeek Harness UI, uses a four‑agent review pipeline, and keeps all data under the user’s control.

CyberCode is an open‑source, locally‑run AI coding assistant built on the Claude‑Code interaction style. It offers permanent memory, self‑evolution, a TUI, a cross‑platform desktop GUI, remote control via Telegram/Feishu, model‑agnostic provider support, code‑graph indexing, token‑optimisation, scheduled tasks, and extensible plugins/skills.

A professional AI subtitling software that uses advanced ASR models and custom alignment algorithms to provide high-accuracy local transcription and translation for video editors.

ORG‑2 is a Rust‑based desktop tool that records the entire workflow of AI coding agents (prompts, tool calls, file edits, etc.), lets teams replay sessions like a video, and provides “AI‑blame” linking every line of code back to the originating agent decision. It supports 20+ agent CLIs, offers collaboration features, a built‑in terminal/Git workspace, and optional browser automation. Distributed as native installers for macOS, Windows and Linux, it is AGPL‑3.0 licensed.

GPA is a unified auto-regressive transformer model that integrates speech recognition (ASR) and text-to-speech (TTS) into a single system with near-SOTA performance.

Session Kit is a local command‑line dashboard that lists all your Claude Code and Codex AI coding sessions, shows their state (e.g., waiting for input), and lets you jump to any session by typing its stable number. Sessions survive terminal or SSH disconnects, can use different provider accounts, and are protected by a safety model that re‑verifies the exact provider process before acting. It installs via a verified release, runs on Linux (systemd) or macOS 14+, and stores all data locally with no telemetry.

Agentlas OS is an open‑source, cross‑platform framework (Hephaestus engine) that lets you build, borrow, and own portable LLM‑driven agents or teams. Agents are packaged method documents with strict JSON contracts, stored in a private “Agent Cloud” or fetched from a public Hub, and can run on any supported host (macOS, Windows, Linux) with any LLM you already use. Installation is a single script; the command surface (`/agentlas build`, `/agentlas hub`, `/agentlas upload`, etc.) and a visual Desktop UI let you compose, verify, and execute agents while keeping credentials and permissions local.

A minimal, MIT‑licensed *skill* that lets AI agents obtain disposable phone numbers and read SMS verification codes via a commercial API, enabling automated sign‑ups and 2FA without human interaction.

Confucius4-R2T2 is a low-latency, high-accuracy real-time speech recognition model that provides stable, append-only transcriptions to prevent text revisions in live applications.

Bend is an open‑source language that compiles to native‑speed CPU/GPU code while requiring formal proofs for user‑defined invariants (`LAWS.bend`). The compiler checks proofs in seconds, enabling AI‑generated code to be automatically verified against safety or correctness rules. It offers implicit parallelism, a Python‑like syntax with dependent types, and targets C, Metal, CUDA, and JavaScript. The project includes a compiler, standard library, benchmarks, academic papers, and a minimal formatter LSP, but it is still young—code is verbose, tooling is limited, and many libraries are missing.

MARK LIV is a cross‑platform, voice‑first AI assistant (Windows/macOS/Linux) that uses Gemini 3.1 Flash Live for real‑time conversation. It displays a software‑rendered 3‑D human head that lip‑syncs, blinks, and tracks gaze, serving as a visual status indicator. Features include push‑to‑talk, local wake‑word, self‑echo guard, unlimited session memory, undo, plugin system, system control, web search, hardware monitoring, and many voice‑driven utilities. No GPU or extra dependencies are required; only a 25 KB facial asset and the existing PyQt6/numpy stack are used.

dbdemos is a pip‑installable Python toolkit that automates the setup of Databricks Lakehouse demo bundles—installing notebooks, clusters, pipelines, dashboards and optional ML models with a single command.

KaoBuddy is a PWA that lets you upload PDFs, docs, handwritten images, and B‑site video subtitles, then uses an LLM (DeepSeek, Kimi, OpenAI, etc.) to extract exam‑relevant knowledge points, generate explanations, flash‑cards, practice questions and full mock exams. It builds a daily study plan, stores everything locally in IndexedDB, and runs via a FastAPI backend + React‑TypeScript frontend. Windows users can use a portable zip; macOS and other platforms run from source or Docker. MIT‑licensed.

A free, open-source voice-to-text tool that performs all transcription locally on your computer to provide private, offline dictation into any application.

Jupynium.nvim is a Neovim plugin that syncs a Jupytext‑style file (`*.ju.py`) to a live Jupyter Notebook opened in Firefox via Selenium. It provides one‑way (Neovim → browser) live preview, execution shortcuts, optional completions, folding, and a small Lua API, enabling a full notebook workflow without leaving the editor.

Sec‑Gemini SDKs are thin client libraries (Python, TypeScript) and a web‑component widget that let developers call Google’s experimental cybersecurity‑focused AI model. The repo is a simple integration layer, not an officially supported Google product.

ClaudeBox is a Docker‑based tool that launches Anthropic’s Claude Code AI coding assistant inside an isolated container. It provides per‑project Docker images, persistent authentication/history, and a set of pre‑configured development profiles (C/C++, Python, Rust, Go, etc.). Features include multi‑instance support, firewall allowlists, macOS clipboard bridging, tmux integration, and a simple task engine. Installation is via a self‑extracting `.run` installer that also sets up Docker if needed. Users create “slots” (persistent Claude sessions) per project, add profiles, and run Claude with custom flags, all while keeping project data isolated and reproducible.

A Node.js skill that lets AI agents browse, compare, and purchase real products from the Filtmall marketplace, handling discovery, checkout, order tracking, and after‑sales via a JSON‑based CLI.

Voicetypr is an open‑source, offline‑first dictation app for macOS and Windows. It captures speech via a global hot‑key, transcribes locally with Whisper (and Parakeet on Apple Silicon) or optionally via cloud STT services, can clean up the text with LLMs, and inserts the result at the current cursor. A companion CLI lets scripts or AI agents use the same transcription pipeline, returning plain text or JSON. The app is built with Tauri (Rust backend) and React, offers LAN‑based private transcription servers, and is released under AGPL‑3.0.

EmperorClaw is a self‑hosted Node.js/Next.js app that adds an organisational layer for AI agents: task board, client directory, knowledge base, secure file storage, chat, pipelines, and full audit logs. Agents register via a generic MCP API (OpenClaw, Hermes, or any compatible runtime). The system runs in Docker, stores state in PostgreSQL, and is licensed under a Fair‑Source model that becomes Apache 2.0 after two years.

Ref MCP is a Node‑JS MCP server that lets LLM coding agents search documentation and fetch only the relevant excerpts, dramatically cutting token usage and cost.

A fork of ida‑pro‑mcp that provides a hardened, multi‑instance MCP server for headless IDA Pro, enabling AI agents (Claude Code, etc.) to drive reverse‑engineering tasks via a secure HTTP API.

BIDARA is a GPT‑4‑based chatbot that guides users through the Biomimicry Institute’s Design Process, helping them define problems, translate them into biological questions, discover natural models, abstract strategies, and emulate nature‑inspired solutions. It runs as a Discord bot, a web app, or via a reusable system prompt for OpenAI’s Playground.

A family of open-source music foundation models for high-fidelity music generation, audio tokenization, and lyrics transcription conditioned on lyrics and tags.

CCPlugins provides 24 ready‑made Claude Code CLI commands (e.g., /cleanproject, /commit, /security-scan, /scaffold) that automate repetitive dev tasks, add git safety checkpoints, and deliver fast, deterministic results. Install with a one‑liner script; commands work on any language or OS and can be chained in CI/CD pipelines. The project is actively maintained and MIT‑licensed.

AVA is a TypeScript library that lets you load CSV/JSON data, ask natural‑language questions, and receive AI‑generated analysis, code/SQL, and visualizations. It automatically picks in‑memory JavaScript for small datasets and SQLite/IndexedDB for larger ones, works in browsers and Node.js, and provides query suggestions and chart generation via LLMs.

LocalGPT is a Rust‑based, locally‑run tool that turns natural‑language prompts into 3‑D Bevy worlds (`localgpt‑gen`) and provides a privacy‑first AI assistant (`localgpt`) with persistent, searchable memory, autonomous background tasks, and a tiny HTTP API. It works with any LLM (local or cloud), runs as a single binary, and includes sandboxed security features.

Axe is a Go‑based CLI that runs LLM‑powered agents defined in TOML files. Agents are single‑purpose, can use built‑in sandboxed tools (file ops, shell commands, web fetch/search), delegate to sub‑agents, and keep persistent markdown memory. It supports Anthropic, OpenAI, Ollama, OpenCode, and Bedrock, offers token‑budget limits, retry policies, dry‑run mode, JSON output, and can be composed with standard Unix pipelines, cron, or Docker. Installation is via pre‑built binaries, `go install`, or building from source.

Sentient is an open‑source, self‑hostable AI personal assistant that can chat (text/voice), remember user preferences, automate multi‑step and recurring tasks, and proactively manage email/calendar, with integrations for 20+ apps.

An LLM‑orchestrated CLI tool that turns a high‑level request into a spec, spawns Claude‑based design agents and Codex‑based code‑generation agents, runs isolated builder worktrees, validates with frozen checks, and finally emits a single PR or local finish record—no human approval gates required.