OPENWARP

AI terminal for
local and SSH work.

Run an Agent locally to work on your projects and SSH servers. Connect Codex or Claude Code through MCP to read logs, upload builds, and deploy.

Download for Mac Documentation

macOS · Apple Silicon

产品演示模拟数据

一个 Agent，三种用法

选一个场景，点「发送」看完整执行过程：Agent 怎么规划、调用了哪些命令、在哪台机器上执行、最后给出什么结论。

Agent 运行在: 本机 · Native
命令执行在: 本机 Shell
模型: OpenAI 兼容接口

⌘K

你

检查这台电脑的 CPU、内存与磁盘，采样观察当前负载。

计划：识别环境 → 检查资源 → 持续采样负载。

✓识别本机环境本地 Shell1.4s

$uname -sm; sysctl -n hw.ncpu hw.memsize

Darwin arm64
8
17179869184

✓检查磁盘与内存压力本地 Shell1.4s

$df -h /; memory_pressure

/dev/disk3s1  460Gi  274Gi  186Gi  60%  /
System-wide memory free percentage: 42%

✓持续采样 CPU · 后台任务本地 Shell3.6s

$top -l 3 -s 1 -n 0 | grep "CPU usage"

CPU usage: 12.4% user, 5.8% sys, 81.8% idle
CPU usage: 11.7% user, 5.5% sys, 82.8% idle
CPU usage: 12.0% user, 5.6% sys, 82.4% idle

检查完成3 步 · 6.4s

CPU: 8 核 · 空闲 82%
内存: 16 GB · 压力正常
磁盘: 可用 186 / 460 GB
结论: 无持续高负载

检查完成：8 核 CPU，16 GB 内存；当前 CPU 空闲约 82%，内存压力正常，磁盘可用 186 GB。采样期间未见持续高负载。

检查这台电脑的 CPU、内存与磁盘，采样观察当前负载。

Native本地 Shell

MODEL PROVIDERS

Works with any OpenAI-compatible endpoint. API keys stay on your machine.

OpenAI
DeepSeek
Ollama
OpenRouter
LM Studio
vLLM
…

01 / WHY OPENWARP

No Agent to install on every server.

Your Mac carries the Agent and the keys. Servers only need SSH.

BeforeA full stack on every server

demo-serverAgentHarnessAPI Key

staging-dbAgentHarnessAPI Key

prod-apiAgentHarnessAPI Key

With OpenWarpOne stack, on your Mac

Your MacAgentHarnessAPI Key

demo-serverSSH

staging-dbSSH

prod-apiSSH

02 / ARCHITECTURE

一个客户端，
多种 Agent 框架。

原生桌面客户端提供交互，本地 Adapter 连接不同 Harness。模型请求与工具执行各走自己的通道。

01 · Client

OpenWarp原生桌面客户端终端 · Agent 对话
SSH 连接管理

02 · Local

本地协调层Adapter /
Agent Engine会话与配置
Runtime 对接 · 工具协议

03 · Harness

Native内置 Harness

Pi AgentPi Harness

DSHDeepSeek Harness

04 · Channels

模型请求 / 响应Model APIOpenAI-compatible API
远程服务或本地模型

工具调用 / 结果Local / SSH Tools本地项目或远端服务器
命令不通过模型服务执行

SSH MCP

Codex / Claude Code ──▸ 本地 MCP bridge ──▸ OpenWarp 授权与 SSH 工具 ──▸ 服务器

外部 Agent 使用自己的 Harness，不必经过 OpenWarp 的模型工作循环。

架构与实现详解 ↗

03 / SSH AGENT

Commands run on the server. Always.

Start the Agent in an SSH tab and every command runs on that host.

Explore the tools

Localdemo-server

developer@demo-server

$ hostname; whoami; pwd

demo-serverdeveloper/srv/demo-api

Local tab → your MacSSH tab → the server

04 / SSH MCP

Let Codex use your SSH connections. Server by server.

Codex and Claude Code use your saved connections. They only get what you switch on.

Set up MCP

Settings → MCP Access

Enable MCP

ServerCommandsUploadDownload

demo-server

staging-db

prod-api

Call log

05 / GET STARTED

Three steps.

Download for Mac

01
Install OpenWarp
GitHub Releases
02
Connect a model
Settings → Agent Engine
03
Open a workspace
Local tab / SSH tab
Localdemo-server
❯Why is demo-api returning 500?
- ◌Reading logs journalctl -u demo-api
- ›ERROR POST /orders 500
- ✓Root cause: orders.go:42, customer is nil

Frequently asked questions

Which platforms are available?

Prebuilt releases currently target macOS on Apple Silicon. A complete Windows desktop installer is not available yet.

Does local-first mean offline?

Configuration and runtime state live locally. Prompts and tool context go to your configured model endpoint. Offline use requires a local model service.

Is this an official Warp product?

No. OpenWarp is an independent community project built around the open-source Warp client. It is not affiliated with Warp.

What is not supported yet?

Full Warp cloud parity, subagents, computer use, and passive suggestions are not implemented. Check the supported-tools guide for the current scope.

AI terminal forlocal and SSH work.