worktrunk: a git worktree manager that simplifies parallel development for AI agents
A CLI for git worktree management designed to facilitate running multiple AI agents in parallel by making worktrees as easy to use as branches.
agentmemory: Persistent, token‑efficient memory for any coding LLM agent
agentmemory is a self‑hosted, SQLite‑backed memory engine for AI coding assistants. It auto‑captures agent actions via hooks, stores them with hybrid BM25/vector/graph search, and serves the knowledge to any MCP‑ or REST‑compatible agent (Claude Code, Copilot CLI, Gemini CLI, etc.). Benchmarks show ~95 % recall @5 and >10× token savings, while a real‑time viewer lets you inspect memories. Install with `npm i -g @agentmemory/agentmemory`, start the server, and connect your agent with `agentmemory connect <agent>`.
agent-governance-toolkit: a deterministic policy enforcement and governance layer for autonomous AI agents
A governance framework for autonomous AI agents that provides deterministic policy enforcement, identity management, and audit logging to prevent unauthorized actions.
agent-framework: Microsoft Agent Framework – a cross‑platform framework for building and operating production‑grade AI agents and workflows.
Microsoft Agent Framework is an open, multi‑language (Python/.NET) toolkit for creating production‑ready AI agents and multi‑agent workflows, offering provider‑agnostic orchestration, middleware, observability, declarative YAML definitions, skill‑based knowledge, and hosting options such as Microsoft Foundry.
hindsight: a biomimetic long-term memory system for agents that learns from experiences and world facts
An agent memory system that enables AI agents to learn from experiences and world facts using biomimetic data structures, moving beyond simple conversation history.
graphiti: a temporal context graph engine for AI agents that tracks how facts change over time
A framework for building temporal context graphs that allow AI agents to track evolving facts and maintain historical provenance from raw data streams.
iPolloWork: a local-first visual AI workbench for creating editable code, documents, and multimedia content via agents
A local-first visual AI workbench that enables agents to create and maintain editable code, documents, presentations, and videos in a single workspace.
voice-pro: an all-in-one AI dubbing studio for YouTube processing, voice cloning, and multilingual translation
An AI-powered web application for speech recognition, translation, and multilingual dubbing that integrates ASR, TTS, and voice cloning into a single workflow.
opencodex: a universal provider proxy that lets you run any LLM with Codex, Claude Code, and Grok Build
A local proxy that enables the use of any LLM provider with OpenAI Codex, Claude Code, Claude Desktop, and Grok Build by translating their API responses.
mesh-llm: a distributed inference system that pools GPU and memory resources across machines to run large models
A distributed LLM inference system that pools GPUs and memory across multiple machines to run large models via an OpenAI-compatible API.
omnigent: an open-source meta-harness for orchestrating and governing multiple AI agents across devices
An open-source meta-harness that provides a common orchestration layer for AI agents, allowing users to swap runtimes, enforce policies, and collaborate in real-time across devices.
EverOS: a local-first memory runtime for agents that uses Markdown as the canonical source of truth
A local-first memory runtime for AI agents that uses Markdown as the source of truth to provide portable, persistent long-term memory across apps and devices.
semantic-router: a programmable routing layer for building Mixture-of-Models systems across heterogeneous LLM infrastructure
A programmable routing layer for Mixture-of-Models systems that optimizes LLM request routing across heterogeneous compute and infrastructure to improve quality, cost, and latency.
openmed: a local-first clinical text pipeline for PII de-identification and medical entity extraction
A local-first healthcare AI runtime for clinical entity extraction and PII de-identification that runs on-device across Python, Swift, Android, and browsers.
wigolo: wigolo: Local-First Web Intelligence for AI Agents
wigolo is a local-first web intelligence toolkit for AI agents that provides search, fetch, crawl, extract, cache, find-similar, research, and agent tools without requiring API keys for core functionality. It runs as an MCP server, REST endpoint, or SDK, keeping all data private in `~/.wigolo/` and offering optional LLM-based synthesis for advanced tasks. ## Key Features - Ten integrated web tools (search, fetch, crawl, extract, cache, find_similar, research, agent, diff, watch) - Works as MCP server (for coding agents), REST endpoint, or embeddable SDK (TypeScript/Python) - Local-first design: data stays under `~/.wigolo/`; no third-party calls unless LLM synthesis is explicitly opted in - Core tools require zero API keys; uses public search engines and on-device ML models - Returns verbatim excerpts with byte-offset source spans and explainable scoring per result - Zero cost per query after initial setup (caches results locally for instant re-query) - Integrates with major AI agent frameworks: LangChain, CrewAI, LlamaIndex, Vercel AI SDK - Supports direct wiring to popular agents: Claude Code, Cursor, Codex, Gemini CLI, OpenCode, VS Code, Windsurf, Zed, Antigravity ## How It Works wigolo operates as a single Node.js process that exposes three interfaces: MCP (JSON-RPC over stdio for agent communication), REST API, and SDKs. The search tool uses 18 direct adapters with rank fusion and on-device ML reranking. The fetch tool employs a tiered router that escalates from plain HTTP to headless browser based on observable signals (SPA markers, anti-bot challenges). Results are stored in a local cache combining keyword and vector indices. Optional LLM synthesis (for research/agent tools) can be configured via environment variables (e.g., `WIGOLO_LLM_PROVIDER=gemini`). All heavy components (browser engine, embedding models) are lazy-loaded and stored locally under `~/.wigolo/`. ## Limitations - Some challenge-protected sites rely on IP reputation; datacenter IPs may fail where home connections succeed (see self-hosting guide for proxy opt-in) - Without an LLM API key, the `research` and `agent` tools return raw evidence briefs instead of synthesized answers - Initial setup requires ~1.5 GB of free disk space and downloads the browser engine and on-device models - Requires Node.js >= 20 ## Ecosystem - CLI: `npx wigolo <tool>` for terminal use - Interactive shell: `wigolo shell` for NDJSON piping - REST API: `wigolo serve` exposes JSON endpoints - Docker: Official images on GitHub Container Registry and Docker Hub - Framework integrations: Official packages for LangChain, CrewAI, LlamaIndex, and Vercel AI SDK - Agent skills: 11-pack skill catalog managed via `wigolo skills` ## License AGPL-3.0 ## Status Public beta ## Links - Documentation: [docs/README.md](docs/README.md) - Examples: [examples/README.md](examples/README.md) - Installation: [docs/installation.md](docs/installation.md) - Configuration: [docs/configuration.md](docs/configuration.md) - Tools reference: [docs/tools.md](docs/tools.md) - SDKs: [docs/sdks.md](docs/sdks.md) - Self-hosting: [docs/self-hosting.md](docs/self-hosting.md) - Skills: [docs/skills.md](docs/skills.md) - Contributing: [CONTRIBUTING.md](CONTRIBUTING.md)
markstream-vue: Markstream — Streaming Markdown renderers for AI chat
A Vue 3 / Nuxt / VitePress package that renders Markdown incrementally as it is streamed (e.g., from LLM token streams) to minimize flicker and provide a smooth experience in AI chat interfaces.
note-gen: a local-first Markdown app that uses AI to organize scattered information fragments into structured notes
A local-first Markdown app that uses AI to organize scattered fragments of information—like voice recordings and screenshots—into structured notes.
ms-swift: ms‑swift: scalable lightweight infrastructure for fine‑tuning LLMs and multimodal models
ms‑swift is a lightweight, full‑stack framework for fine‑tuning, serving, and evaluating hundreds of LLMs and multimodal models, offering LoRA‑style efficient training, advanced parallelism, quantization, RLHF, and a Gradio‑based UI.
banana-slides: an AI-native presentation generator that creates editable slides and narrated videos from natural language prompts and documents
An AI-native presentation app that generates editable, professional slides from ideas, documents, or images using natural language editing and the nano banana pro model.
airllm: a memory-efficient inference engine that runs massive LLMs on consumer GPUs by loading one layer at a time
AirLLM is a library that enables running massive LLMs (up to 671B parameters) on low-end GPUs by loading only one model layer at a time into VRAM.
DeepSeek-Reasonix: DeepSeek‑Reasonix: a config‑driven, plugin‑based AI coding agent for the terminal.
DeepSeek-Reasonix is a terminal AI coding agent that uses DeepSeek models via a config‑ and plugin‑driven harness, delivered as a single static Go binary optimized for DeepSeek’s prefix cache to keep token costs low.
WrenAI: an open-source generative BI engine that lets AI agents deploy governed dashboards from any database
An open-source generative BI engine that provides a context layer for AI agents to generate governed SQL, create charts, and deploy shareable dashboards from any database.
screenpipe: a local-first AI memory system that captures screen and audio to power searchable history and AI agents
A local-first AI memory tool that captures screen and audio activity to create a searchable history and provide context for AI agents.
openai-agents-python: a provider-agnostic framework for building multi-agent workflows with support for voice and sandboxed execution
A lightweight, provider-agnostic framework for building multi-agent workflows, supporting text, voice, and sandboxed execution across various LLMs.
LightRAG: a lightweight graph-based RAG framework that bridges vector and graph retrieval for efficient large-scale indexing
A lightweight, graph-based RAG framework that combines knowledge graphs and vector embeddings to provide deep contextual understanding and efficient incremental updates.
vox-director: an automated pipeline for creating editorial paper-collage explainer videos from a single topic
An agent skill that automates the production of Vox-style paper-collage explainer videos, handling everything from scripting and keyframes to motion and audio.
caveman: Caveman – token‑saving plug‑in for AI coding agents
Caveman is a plug‑in that makes AI coding agents speak in terse, caveman‑style language, cutting output tokens by up to 65% on prose while preserving all technical content.
browser-use: an AI agent framework that enables LLMs to control a web browser for automated web tasks
A library and framework that enables AI agents to control a web browser to perform tasks like filling forms, extracting data, and QA testing through natural language commands.
sglang: a high-performance serving framework for LLMs and multimodal models featuring RadixAttention and broad hardware support
SGLang is a high-performance serving framework for LLMs and multimodal models that optimizes inference throughput and latency across diverse hardware via advanced caching and scheduling.
CodeWhale: a model-agnostic terminal coding agent that supports BYOK and granular permission controls
An open-source, model-agnostic coding agent for the terminal that can read code, edit files, and run commands using any LLM provider.
mempalace: a local-first AI memory system with verbatim storage and pluggable vector backends
A local-first AI memory system that stores conversation history as verbatim text and uses semantic search for high-precision retrieval without summaries.
mnemosyne: Mnemosyne: Local-First AI Memory Layer
Mnemosyne is a local-first, SQLite-backed memory layer for AI agents that provides working memory, episodic memory, and a temporal knowledge graph (TripleStore) without requiring external services or cloud dependencies.
ECC: ECC: Agent Harness Operating System for AI Coding Assistants
ECC provides an agent harness operating system that equips AI coding agents with a coordinated engineering system—agents, skills, commands, hooks, memory, and security scanning—to plan, test, implement, review, verify, remember, and improve code.
inkos: a professional story-creation agent system with multi-dimensional continuity auditing and open-world interactive fiction capabilities
An AI Agent system for long-form story creation and interactive fiction that ensures narrative consistency through a multi-agent auditing pipeline and structured long-term memory.
apfel: apfel: Apple's on-device LLM as a CLI and OpenAI-compatible server for macOS
apfel brings Apple's on-device LLM to the terminal and as a local OpenAI-compatible server for private, free AI on Apple Silicon Macs.
OpenOutreach: OpenOutreach: Self-hosted AI sales agent for B2B lead generation
A self-hosted, open-source AI sales agent that automates B2B lead generation by finding company profiles from licensed data, using AI to qualify the best leads, and sending personalized email from the user's own mailbox.
herdr: a terminal-based agent multiplexer for managing and coordinating multiple AI agents
A terminal-based agent multiplexer that allows users to monitor, manage, and coordinate multiple AI agents through persistent sessions and a socket API.
open-connector: OpenConnector – a unified, self‑hostable connector catalog that gives LLM agents safe, reusable access to thousands of SaaS tools
OpenConnector is an open‑source gateway that lets AI agents call 1 000+ SaaS services via a catalog of pre‑built Actions. It securely stores API keys/OAuth tokens, provides inspectable request/response schemas, and can be run locally, on Cloudflare Workers, Fly.io, or via OOMOL’s hosted SaaS. Integration is possible through a TypeScript SDK, a CLI, MCP, or direct OpenAPI calls, and a web console offers admin and debugging UI.
apify-mcp-server: an MCP server that connects AI agents to thousands of ready-made web scrapers and automation tools
An MCP server that enables AI agents to use thousands of Apify Actors for web scraping, data extraction, and automation, featuring dynamic tool discovery and agentic payments.
open-code-review: an AI-powered code review CLI tool that combines deterministic engineering with an agentic workflow for high-precision defect detection
An AI-powered code review CLI tool that combines deterministic engineering with an agentic workflow to provide high-precision, line-level code defect detection.
SmartSub: SmartSub – One‑click video subtitle, translation, and AI voice‑over on your desktop
SmartSub (妙幕) is an open‑source cross‑platform desktop app that lets you download videos, transcribe audio with local or cloud ASR models, translate subtitles via dozens of services, proofread them, generate voice‑overs (including zero‑shot voice cloning), and finally burn or mux the subtitles back into the video—all offline‑first and with optional GPU acceleration.
openlake: a distributed storage engine for GPU workloads that optimizes KV cache offloading and model training
A high-performance distributed storage engine built in Rust for GPU workloads, optimizing LLM inference and training by providing fast KV cache offloading and S3-compatible object storage.
Vibe-Skills: VibeSkills: a Skill‑orchestration harness for AI agents that routes local Skills, plans work in L/XL levels, records progress, and validates results via 17 checks.
VibeSkills orchestrates locally installed Skills to plan, execute, and verify complex tasks, providing level‑based task breakdown, Skill selection, execution tracking, and a 17‑point verification check.
llama.cpp: a high-performance C/C++ inference engine for running LLMs and VLMs locally across diverse hardware
A plain C/C++ implementation for LLM and VLM inference that enables high-performance local execution across a wide range of hardware with minimal setup.
DiffSynth-Studio: an open-source diffusion model engine for cutting-edge exploration and training of generative image, video, and audio models
An open-source diffusion model engine that provides a unified framework for integrating, training, and deploying cutting-edge generative models for image, video, and audio.
OpenBiliClaw: a cross-platform personalized content discovery agent that builds local psychological profiles to break information cocoons
A local, privacy-focused AI Agent that builds a deep psychological profile of the user to proactively discover personalized content across multiple platforms like Bilibili, YouTube, and X.
PyRIT: an open-source framework for proactively identifying risks in generative AI systems
PyRIT is an open-source framework that enables security professionals and engineers to proactively identify risks in generative AI systems.
cc-haha: a graphical desktop workspace for AI coding agents with visual diff review and remote IM integration
A desktop workspace for Claude Code that provides a graphical interface for managing AI coding sessions, reviewing diffs, and controlling agentic permissions.
natively-cluely-ai-assistant: a private, low-latency interview copilot with undetectable stealth mode and local RAG memory
A real-time AI interview copilot and meeting assistant that provides low-latency, private, and undetectable live assistance during interviews and meetings.
remove-ai-watermarks: a comprehensive utility to strip visible and invisible AI provenance marks from images and video
A tool for removing visible labels, invisible pixel watermarks, and AI metadata from AI-generated images and videos.