Computer use
agents that operate a desktop or GUI
NanmiCoder/cc-haha NanmiCoder/cc-haha is the de-facto leader by adoption and breadth, offering a desktop-first, local-capable workstation that bundles code/desktop control and integrations. The single most useful decision for teams is whether they need a fully local/deterministic agent or are comfortable relying on cloud models and specialized device integrations — that choice narrows the useful projects quickly.
Which one matches your setup?
Answer any of the questions — the shortlist updates as you go. Recommendations come from the capability passports below, nothing else.
Comparison matrix
Axes are extracted from each project's docs by our review pipeline; the maturity score is computed from stars, growth and commit activity — not an opinion. Click a column to sort.
| Runs in | Models | Context | Cost to run | |||||
|---|---|---|---|---|---|---|---|---|
⭐ 14.3k +38/7d | CLIWeb app | AnthropicOpenAIBYOKOllama / local | Diff + Worktree | Approval gating | ●●●●● | Self-hostable | ●●●●● | Requires configuring your model provider API key (third‑party billing applies) |
⭐ 14.1k +48/7d | CLIWeb app | BYOKGemini | Whole-repo analysis | Human-in-the-loop | ●●●●● | Cloud API key | ●●●●● | Your LLM API key; provider charges apply |
⭐ 3.8k +225/7d | CLI | Fixed provider | Recording only | Manual review/edit | ●●●●● | Copilot cloud required | ●●●●● | Requires GitHub Copilot access |
⭐ 1.9k +69/7d | Web appCLI | BYOKOpenAIAnthropicGemini | Diff + related files | None mentioned | ●●●●● | Self-hostable | ●●●●● | Use your API key or third‑party relays for external models; desktop app is local-first |
⭐ 1.9k +94/7d | CLICoding-agent plugin | Fixed provider | Local UI only | None mentioned | ●●●●● | Fully local | ●●●●● | Free, local |
⭐ 1.7k +84/7d | CLIWeb appCoding-agent plugin | BYOKAnthropicOpenAIGeminiFixed provider | Device-only context | Review & retry | ●●●●● | BYO API keys (cloud) | ●●●●● | Requires your model API keys for model usage; local backend and Android client run from the repo. |
⭐ 1.3k +35/7d | CLICICoding-agent plugin | Fixed provider | Diff only | Basic filters | ●●●●● | Fully local | ●●●●● | Free, local |
⭐ 750 +51/7d | CLIWeb app | BYOK | Workspace-aware | Plan-mode gating | ●●●●● | Self-hostable | ●●●●● | Requires pi and your provider API keys (configured via pi) |
⭐ 575 +141/7d | CLI | BYOK | Project + related files | Revision & transaction checks | ●●●●● | Cloud APIs (your key) | ●●●●● | Your model provider API key (provider charges may apply) |
⭐ 12.2k +22/7d | CLIWeb app | BYOKOpenAIAnthropicGeminiFixed provider | Single-screen context | Basic filters | ●●●●● | Cloud APIs (your key) | ●●●●● | Your API keys (provider-dependent) |
⭐ 5.0k | Web app | BYOK | Document-level | Snapshots & diffs | ●●●●● | Self-hostable (BYOK) | ●●●●● | Free app; uses Genspark proxy by default or your API key for model costs. |
⭐ 4.2k +955/7d | Web appCLICI | BYOKOpenAIAnthropicGemini | Workspace access | Policy and audit | ●●●●● | Cloud APIs w/ key | ●●●●● | Requires CopilotKit license and your model API key. |
Capability profiles
Six axes, 0–5 each. The shape tells you the strategy: a wide hexagon is a generalist, a spike is a specialist. Showing the 8 most established — the rest are in the full catalog.
A desktop-first, local-first Claude Code workstation that combines worktree-aware code diffs, Computer Use desktop control, H5 remote access and IM integrations in one app.
Self-evolving agent that crystallizes each solved task into a persistent, reusable personal skill tree from a minimal (~3K-line) seed.
Converts a single on-screen recording (clicks, window switches, URLs, and optional narration) into a generalized, reusable Skill or scheduled Automation using the GitHub Copilot CLI.
A local‑first desktop AI agent that actually performs file-system operations, runs bash/managed processes, and extends via an MCP/Skills ecosystem with an optional remote Gateway.
Open-source, self-hosted MCP implementation of Computer Use that provides a local alternative to Codex Computer Use and integrates with multiple agent clients.
Built as a backend-managed mobile operator system with explicit orchestration, standby device dispatch, and separable planning/VLM roles for long-running phone workflows.
Provides a token-efficient, agent-native observe→act CLI that lets agents read compact accessibility outlines and perform alias-cached taps on both iOS Simulators and Android devices.
Provides a desktop workspace that runs multiple local pi RPC agent processes to manage multi-project sessions, history, Git integration and import of Codex/Claude local sessions in one unified GUI.
All repositories (25)
Local-first cross-platform desktop workspace for Claude Code / agents: multi-agent, Git worktrees, code diffs,…
Self-evolving agent: grows skill tree from 3.3K-line seed, achieving full system control with 6x less token co…
Agent S: an open agentic framework that uses computers like a human
AppAgent: Multimodal Agents as Smartphone Users, an LLM-based multimodal agent framework designed to operate s…
Free, open-source alternative to Microsoft Office with built-in AI agents — Word (.docx), Excel (.xlsx), Power…
Open-source AI coworkers that each get a computer of their own: a browser, files and tools, with every action…
Desktop app that records your on-screen work session and uses the GitHub Copilot CLI to reconstruct it as an i…
Open-source AI agent desktop app for Windows & macOS. One-click install Claude Code, MCP tools, and Skills — w…
A fully functional AI Agent desktop client that supports Webui access and can be creatively customized and exp…
👾 Open Computer Use – Open-Source Alternative to Codex Computer Use
Compiles a demonstrated GUI task into a program that reports VERIFIED only if an independent check agrees. pip…
OpenGUI is an Android GUI agent framework for phone-use AI that can see, plan, and operate real mobile apps th…
The first open-source Artificial Narrow Intelligence generalist agentic framework Computer-Using-Agent that fu…
Give your AI agent eyes and hands on iOS Simulator and Android emulator/devices.
PiDeck 是一个开源的桌面工作台，用于在本地项目目录中统一管理 pi Agent 会话，并支持导入 Codex、Claude 本地会话以便统一浏览和恢复。支持多项目工作区、会话历史、Git 集成、内置终端、模型配置和…
One chat. Every agent. A universal interface for AI agents running on your computer.
Autonomous self-evolving agents. Vision-grounded layered memory and self-written skills for LLM agents that op…
Open source previs software in the browser: block a scene, pose characters, author camera moves and cuts, then…
嘉立创EDA专业版(EasyEDA Pro)自动化：给 AI harness 装上画板的「手」—— 一套 typed 原理图/PCB 动作，CLI / Agent Skill / stdio MCP 三形态融合接入。承接…
DeepSeek Harness (DSH) plugin: a live iOS Simulator — and a USB-connected iPhone — inside the conversation. 22…
A macOS Dynamic Island for AI coding agents: Claude Code, Codex, Gemini CLI, Cursor, hooks, skills, and local…
One macOS app for Claude Code, Codex, and every agent runtime you use — scheduled runs, global hotkey launcher…
基于多模态视觉感知与 LLM Agent 的 macOS 微信自动化框架 | Visual RPA for WeChat
A menu-bar macOS agent loop: every N minutes it asks Claude/Codex what's overloading your Mac and suggests one…
The bioinformatics agent desktop for clinicians and wet-lab scientists — built on DeepSeek Harness. One-click…
Hosted alternatives
If running your own reviewer is more ops than you want, these managed services cover the same job.
Anthropic's hosted computer-use capability: Claude drives a virtual desktop through the API, so you get screen-grounded control without building the perceive-act loop or running a VM.
Try Claude Computer Use →A managed agent that operates a browser and desktop tasks on OpenAI's infrastructure, the hosted counterpart to running a local computer-use harness yourself.
Try OpenAI Operator →Rents ready-made cloud desktops and browsers for agents to control, so you point an existing agent at a managed machine instead of provisioning and securing one.
Try Scrapybara →