Terminal
Execute commands, manage environments, automate workflows
Mini Harness
Your AI Agent for Real Work
One workspace for models, tools and long-running tasks. Continue the same conversation from computer or phone, with the original executor calling models and running tools.
● Cross-platform● Local-first● Unified execution
Browser
Web forms, persistent logins, macOS apps and Android interaction
Search
Real-time information with multiple search providers
Files
Read, write, analyze, preview and deliver your files
Workflows
Background jobs, task follow-up and recoverable checkpoints
A flexible tool bus that connects frontier models with real-world capabilities. Extend with skills, custom tools, and provider adapters.
Explore tools →›_ Terminal◎ Browser⌕ Search▱ Files
Mini HarnessTool Bus♙ Skills⌘ Custom Tools⌘ APIs▱ Your Systems
Built-in safeguards, human-in-the-loop control, and transparent execution. You stay in control every step of the way.
- Permission-based tool access
- Real-time execution preview
- Stop, steer and resume supported tasks
- Execution logs and saved checkpoints
✓
Works with leading models and toolsUse your own cloud or local models. Carry history between compatible profiles.
◉ OpenAI✳ ClaudeOpenAI-compatible◈ DeepSeek〽 Skills▣ Qwen△ Local Models›_ SSHand more...
CONTINUOUS WORK
Cross-device execution
Same account, same conversation. Send, approve, stop or resume on the original executor.
Background & recovery
Results wake the owning chat. Recover stopped work from available checkpoints.
Model handoff
Switch compatible model profiles with recoverable history, confirmed by the executor.
Project memory
Reuse project facts and rules. Manage older tool output within the context budget.
Skills & subagents
Reuse workflows and delegate independent subtasks to isolated agent sessions.
Preview & deliver
Preview supported files, download deliverables and retain the delivered version.
Cross-device control needs the same account, supported clients and an online executor. App control and speech capabilities vary by platform.
Real-World Use Cases
Build ApplicationsFrom idea to full-stack app. Plan, code, test, and deploy with AI agents.
→
Analyze DataExplore data, generate insights, and create reports automatically.
→
Automate OperationsHandle repetitive tasks, system maintenance, and complex workflows.
→
Research & LearningBrowse, search, read and synthesize information from across the web.
→
DOWNLOAD
下载 Mini Harness
macOS、Windows、Linux、Android 与 iOS，本地运行，无需另装 Python。
累计下载6,236
macOSv1.0.7
↓ 下载 MiniHarness-macOS.zip提示「文件已损坏 / 无法打开」？解决方法
手动操作（三步）：
- 先把 app 拖进「应用程序」：解压后在 Finder 里把
Mini Harness.app拖到左侧「应用程序」；或直接留在「下载」文件夹也行，路径按实际写。 - 打开「终端」：
Command + 空格→ 输入 终端 / Terminal → 回车（或 应用程序 → 实用工具 → 终端）。 - 粘贴下面这一行并回车：
xattr -dr com.apple.quarantine "/Applications/Mini Harness.app"若 app 留在「下载」文件夹，请改用（注意用$HOME，双引号内的~不会展开）：xattr -dr com.apple.quarantine "$HOME/Downloads/Mini Harness.app"
Linuxv1.0.7
↓ 下载 MiniHarness-linux.tar.gz解压后怎么运行？
- 解压：
tar xzf MiniHarness-linux.tar.gz - 命令行启动或双击打开：
./desktop_main.dist/MiniHarness - 若提示缺少 WebKitGTK（Debian / Ubuntu）：
sudo apt install libwebkit2gtk-4.1-0
Android APKv1.0.7
↓ 下载 MiniHarness-android.apk装好之后怎么用？
- 打开 App，在 设置 → 模型中配置 API 地址、密钥和模型。文字对话无需语音组件。
- APK 内置本地 Python 引擎，可直接运行任务；需要访问电脑或服务器时，可连接 SSH 工作区。
- 安装包不内置 ASR 模型（APK 约 39 MB）。要用端上语音，在 设置 → 界面下载语音识别组件（约 842 MB，安装约 1 GB，建议 Wi-Fi）； 仅需下载一次，APK 更新会保留，装好后离线语音可用。
iOS IPAv1.0.7
↓ 下载 MiniHarness-ios.ipa如何侧载安装？
- 在电脑安装 Sideloadly，连接并信任 iPhone / iPad。
- 拖入 IPA，使用自己的 Apple ID 重新签名并安装；按系统提示启用开发者模式、信任开发者。
- 免费 Apple ID 签名有效 7 天，到期前重新签名或启用自动刷新。保持 Apple ID 和应用标识一致，覆盖安装可保留数据。
- 内置本地 Python 引擎；iOS 版暂不提供本地 Shell 和语音识别。作为执行端时需保持前台。
📦 版本 v1.0.7（2026-11-08 构建）· 校验和见 SHA256SUMS.txt ·
macOS 首次打开若被 Gatekeeper 拦截，请在「系统设置 → 隐私与安全性」中允许，或右键选择打开；
Windows 若提示 SmartScreen 未识别，点「更多信息 → 仍要运行」即可。
Ready to build with Mini Harness?
Download for your device and connect your favorite model.