A curated list of on-device AI agent runners, harnesses, and tooling — agents whose control loop runs locally on your desktop, laptop, or workstation rather than on a hosted cloud platform.
The underlying model may be a hosted API (Anthropic, OpenAI, xAI, Mistral…) or a fully local engine (Ollama, llama.cpp, vLLM). What unifies this list is that the harness lives on your machine.
Last updated: 2026-04-06. This list is updated periodically as new projects emerge — expect refreshes on a rolling basis.
See SCOPE.md for the full inclusion/exclusion rules and grey-area guidance. In short:
In scope: CLI/TUI coding agents, sysadmin-style agents, tooling/orchestrators around agent CLIs, MCP servers, multi-agent frameworks that run locally, computer-use agents, desktop GUI agents, personal AI assistants, terminal assistants, harnesses, and desktop-agent protocols.
Out of scope: cloud-hosted agent platforms, IDE-first tools where the CLI is only a companion, pure inference engines (Ollama, llama.cpp, vLLM) without an agent loop, and frameworks that don't support tool use / MCP directly.
Each entry is tagged with shields.io badges. Stars and last-commit update dynamically from GitHub.
| Badge | Meaning |
|---|---|
 |
Built by the model vendor (OpenAI, Anthropic, Mistral, GitHub, Moonshot, xAI…) |
 |
Independent / community project |
   |
Primary interface |
   |
Local-model support |
 |
Supports MCP (Model Context Protocol) |
 |
Supports tool use / function calling |
  |
Primary use case |
Capability badges (Local/MCP/Tools) are best-effort — verify against upstream docs before relying on them.
- 1. AI Coding Agents (CLI)
- 2. Tooling Around Agent CLIs
- 3. Multi-Agent Frameworks
- 4. Harnesses & Specifications
- 5. Computer-Use & Agent Operating Systems
- 6. Desktop GUIs & Cowork Clones
- 7. Personal AI Assistants
- 8. Terminal Assistants
- 9. Protocols
- 10. Wallets & Payments
- 11. Skill Collections
- 12. Voice & Speech Agents
- 13. Remote Access & Mobile UIs
- 14. Memory
- 15. Gateways & Proxies
- 16. Browser Extensions & Web Agents
- 17. Research Assistants
- Related Resource Lists
- Contributing
- Disclaimer
CLI agents designed to work inside a code repository. In scope: tool use, MCP support, local or cloud model support. Out of scope: GUI apps, pure inference engines, IDE-only tools where the CLI is just a companion.
OpenAI's lightweight coding agent that runs in your terminal.
GitHub Copilot CLI — brings the Copilot coding agent directly to your terminal.
Moonshot's Kimi Code CLI — vendor coding agent for the Kimi model family.
Minimal CLI coding agent by Mistral.
AI pair programming in your terminal. Works with most major models including local ones via Ollama / LiteLLM.
An AI programming agent for the command line.
Glamourous agentic coding for all — Charm's Bubble Tea-powered TUI coding agent.
Open-source, extensible AI agent that goes beyond code suggestions — install, execute, edit, and test with any LLM. Strong MCP support, straddles code and sysadmin use cases.
Open-source autonomous agent powered by Grok.
"The agent that grows with you" — from Nous Research. Scope / in-repo vs sysadmin usage TBD.
OpenHands (formerly OpenDevin) — AI-driven development platform. Runs agents locally in a sandboxed environment, supports many model backends including local ones.
The open source coding agent — provider-agnostic TUI coding agent.
An open-source asynchronous coding agent from the LangChain team.
AI agent toolkit: coding agent CLI, unified LLM API, TUI & web UI libraries, Slack bot, vLLM pods.
LLM-based agent for general-purpose software engineering tasks, by ByteDance.
Tools that wrap, manage, orchestrate, or extend the CLI agents above.
Local-first desktop and web app for browsing, searching, and analyzing AI agent coding sessions across Claude Code, Codex, Gemini, OpenCode, Copilot, and more.
Developer productivity tools for Claude Code and other AI assistants.
Coding Agent Session Manager for Claude Code / Gemini CLI / Codex CLI / Cursor Agent / Copilot CLI / Cline CLI / OpenCode / Kimi CLI.
Desktop application for managing multiple OpenAI Codex CLI accounts.
A desktop app for isolated, parallel agentic development, from Coder.
Kanban-style task manager for orchestrating Claude Code, Codex, and other coding agents — get more out of any agent CLI.
Free, local chat where AI coding agents can tag each other, talk, and coordinate alongside the human.
Programmatically orchestrate thousands of agents and harnesses as a graph.
Agentic orchestrator for parallel coding agents — plans tasks, spawns agents, and handles CI fixes, merge conflicts, and code reviews.
BLACKBOX CLI for running multi-agents locally with a Judge to select the best task implementation.
Private Agent Fleet with Spec Coding — each agent gets its own GPU-accelerated desktop. Runs Claude, Codex, Gemini, and open models on a full private AI stack.
A filesystem designed for agents, from the Turso team.
Chrome DevTools for coding agents — exposes browser debugging to agents via MCP.
MCP server giving Claude terminal control, filesystem search, and diff-file editing.
Security scanner for Agent Skills, from Cisco AI Defense.
Give your AI agent eyes on the internet — read and search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu from one CLI.
Browser automation CLI for AI agents.
On-device multi-agent orchestration.
A complete AI agency at your fingertips — specialized expert agents with personalities, processes, and deliverables.
A reliable multi-agent orchestration framework.
The multi-agent evolution of OpenClaw.
Multi-Agent Harness for Production AI.
TypeScript multi-agent framework — one runTeam() call from goal to result. Auto task decomposition, parallel execution, three dependencies, deploys anywhere Node.js runs.
Recursive-Open-Meta-Agent — a meta-agent framework for building high-performance multi-agent systems.
A production-oriented multi-agent orchestration framework.
Intelligent automation and multi-agent orchestration for Claude Code.
Open formats, specs, and framework-level harnesses that cut across individual agent CLIs.
AGENTS.md — a simple, open format for guiding coding agents.
Specification and documentation for Agent Skills.
GitHub Agentic Workflows — a gh extension for defining and running agentic workflows.
omo — an agent harness (formerly oh-my-opencode).
Plan-first AI agent framework with approval-based execution, multi-language support, and built-in testing/review — built for OpenCode.
Open-source AI assistant framework with skills and agent architecture.
Open Agent Harness — research project from HKUDS.
Ship your code, on autopilot. An open source agent that lives on your machines 24/7 and keeps your apps running.
Create state-machine-powered LLM agents using XState.
Agents that drive the whole OS / desktop — mouse, keyboard, windows, filesystem — rather than being confined to a repo.
Native desktop automation CLI for AI agents. Controls any app through OS accessibility trees with structured JSON output.
Open agentic framework that uses computers like a human.
On-device GUI agent for operating Android apps, with reinforcement fine-tuning for efficient task execution.
AI desktop agent that sees your screen, controls your cursor, and completes tasks autonomously.
Open-source infrastructure for computer-use agents — sandboxes, SDKs, and benchmarks for agents that control macOS/Linux/Windows desktops.
An intelligent GUI agent for mobile phones.
Mobile-Agent: a powerful GUI agent family for operating mobile devices.
State-of-the-art (82% OSWorld Verified) computer-using agent, fully open-source.
Open-source Agent Operating System.
Open-source desktop AI agent that handles your documents, files, and daily workflows — locally, with any model.
Browser-based desktop where an AI agent operates every app through natural language.
Computer-use AI agent powered by OpenAI that interacts with E2B's virtual desktop environment through natural language. Harness runs locally; the sandboxed desktop is E2B-hosted.
Local-first alternative to Cowork-style computer assistants: real PTY terminal ops, explicit approvals, JSONL audit logs. Windows + Linux + macOS, model-agnostic.
Open-source multimodal AI agent stack connecting cutting-edge AI models and agent infra, by ByteDance.
Desktop GUI applications for running agents, including the rapidly growing crop of open-source Claude Cowork alternatives.
Amon — your AI coworker running on your desktop.
Desktop GUI for OpenClaw — turns CLI-based AI orchestration into a desktop experience.
Multi-model AI agent desktop client — connect any provider, extend with MCP & skills, control from your phone. Built with Electron + Next.js.
Self-hosted OS for personal AI agents. Security-first, multi-channel (WhatsApp, Telegram, Discord, Slack, iMessage), multi-provider (Claude, GPT, Gemini, Ollama).
Custom UI for LangChain's Deep Agents.
Claude Cowork alternative — browser automation & AI assistant powered by DeepSeek.
Open-source Cowork desktop. Local and free alternative to Claude Cowork.
7×24 desktop AI agent with visual interface, remote access, file management, and built-in AI browser.
Open-source alternative to Claude Cowork desktop app, by Kuse.
Full-stack open-source self-evolving general AI agent — a fully local alternative to platforms like Manus and Genspark.
Fully autonomous AI-based software development assistant.
Open-source Claude Cowork with 500+ SaaS app integrations.
Open-source Claude Cowork desktop AI assistant for programming, file management, and general tasks.
Open-source Claude Cowork for Windows, macOS, and Linux.
Desktop and web interface for the opencode AI agent.
Open-source alternative to Claude Cowork built for teams, powered by opencode.
Local desktop AI agent for Windows/macOS/Linux with Agent Skills (SKILL.md), autonomous coding, multi-agent teams, desktop automation, 15+ AI providers. No Docker, no terminal.
Desktop AI assistant / universal MCP client.
Desktop agent for any task.
General-purpose, always-on personal assistants that run on your machine.
Ultra-lightweight, efficient personal AI assistant.
Personal AI assistant easy to install and deploy on your own machine. Supports multiple chat apps with extensible capabilities.
Smart assistant that connects powerful AI to your personal world.
Context-aware AI desktop assistant that integrates multiple LLMs and MCP tools.
Open-source AI assistant framework, a Go-flavored openclaw.
Open-source AI personal assistant for email — reach inbox zero fast. Domain-specific: email.
Leon — your open-source personal assistant.
Run OpenClaw on a $5 chip — hardware-oriented agent OS with no Linux/Node/Raspberry Pi dependency.
Fastest, smallest, fully autonomous AI assistant infrastructure, written in Zig.
Your own personal AI assistant — any OS, any platform. 🦞
Embodied AI assistant from MINT-SJTU. Outlier: embodied/robotics — parked here pending a dedicated section.
A personal AI assistant for everyone.
🦞 Personal AI assistant with memory, skills, and scheduled tasks.
Fast, small, fully autonomous AI personal-assistant infrastructure — any OS, any platform.
Shell-integrated AI helpers — less "agent in a repo", more "AI that lives in your terminal".
All-in-one LLM CLI with Shell Assistant, Chat-REPL, RAG, and AI tools/agents. Supports OpenAI, Claude, Gemini, Ollama, Groq, and more.
A delightfully minimal, remarkably powerful AI shell assistant.
AI-powered, non-intrusive terminal assistant that lives inside tmux.
Open protocols and specifications for desktop agent interop.
AG-UI — the Agent-User Interaction Protocol. Bring agents into frontend applications.
Capabilities for on-device agents that need to hold funds or transact.
"Every AI agent deserves a wallet." Coinbase's toolkit for giving agents on-chain wallet capabilities.
Toolkit giving AI agents access to crypto wallets and on-chain functionality, powered by the 0xGasless SDK.
The leading agentic finance toolkit for AI agents.
Collections of portable agent "skills" — prompt/instruction packages designed to plug into multiple agent harnesses.
Production-grade engineering skills for AI coding agents.
Collection of Agent Skills for context engineering, multi-agent architectures, and production agent systems.
Claude Code Skills and 1000+ agent skills from official dev teams and the community — compatible with Codex, Antigravity, Gemini CLI, Cursor, and others.
An agentic skills framework and software development methodology that works.
Vercel's official collection of agent skills.
Collection of Apify Agent Skills for web scraping and automation.
AWS skills for agents.
The Ultimate Guide to Agent Skills: quickstart, resources, features, and toolkit.
Agent skills for Obsidian — teach your agent to use Markdown, Bases, JSON Canvas, and the Obsidian CLI.
Secure, validated skill registry for professional AI coding agents — extends Claude Code, Cursor, Copilot, Antigravity, and more.
Agents whose primary interaction modality is voice — speech-to-speech, voice companions, realtime voice agents.
Framework for building realtime voice AI agents.
Create your own AI desktop companion with customizable characters and voice conversations (~1s response). Long-term memory, visual recognition, voice cloning, Live2D.
Build local voice agents with open-source models.
Bridges that let you drive a desktop agent from your phone, tablet, or another device — keeping the agent loop on your machine.
Access your terminal and your AI agent from any device — phone, tablet, desktop.
Mobile-first web interface for opencode AI agents. Manage and code with multiple opencode agents from any device (phone, tablet, desktop). Git integration, file management, real-time chat, Docker deploy.
Manage agents remotely from your phone, desktop, and CLI.
Memory layers that give on-device agents persistent recall across sessions.
A memory upgrade for your coding agent.
Agent memory that learns over time.
The memory layer for AI agents — persistent, personalized memory across sessions.
Open-source context database for AI agents (e.g. OpenClaw) — unifies memory, resources, and skills via a filesystem paradigm with hierarchical, self-evolving context delivery.
Proxies and aggregators that sit between desktop agents and the services/tools they call.
Unified MCP server that aggregates multiple MCP servers into one.
Next-generation agentic proxy for AI agents and MCP servers.
Browser-resident agents and toolkits for driving the browser from an on-device agent.
Suite of tools for connecting AI to the web — query language and Playwright integrations for precise element interaction and data extraction.
AI browser-automation assistant, privacy-first. Alternative to Manus Browser Operator, Claude Chrome, and Agent Browser.
NativeMind — fully private, open-source, on-device AI assistant that lives as a browser extension.
Open-source, vision-first browser agent.
Autonomous toolkit that lets AI agents browse the web.
Agents focused on long-form research, deep diligence, and report generation.
Multi-agent company-research / due-diligence tool built on LangGraph and Tavily, using Gemini and GPT-class models on the backend.
An autonomous agent for deep financial research.
Other curated lists worth browsing alongside this one.
Curated list of awesome AI tools — broader in scope than this list.
Suggestions and PRs are very welcome — especially to fill in missing badges (local-model support, MCP, vendor vs third-party). Open a PR or reach out: public@danielrosehill.com.
When adding a project, please include:
- Name + GitHub link
- Stars + last-commit badges (dynamic shields.io)
- Type badge: 1st-party or 3rd-party
- Interface badge: CLI / TUI / GUI
- Local-model support (Ollama / Yes / No)
- MCP + tool-use badges where applicable
- Use-case badge: Code / Sysadmin
- 1–2 sentence description, language, author link
This list is non-exhaustive and best-effort. Inclusion is not an endorsement, exclusion is not a judgment. The on-device agent space is moving fast — expect gaps, staleness, and occasional mis-classification. Verify capability claims (especially local-model and MCP support) against upstream docs.
Maintained by Daniel Rosehill