🔶 Warp Agent CLI 正式发布
英文标题：🔶 The Warp Agent CLI
Warp 推出独立的 Warp Agent CLI，可在 Ghostty、iTerm 2、VS Code 以及 Windows 和 Mac 原生终端等环境中使用其多模型编码 Agent。该 CLI 基于 Warp 的终端基础设施构建，原生支持 Agent 会话复用、切换目录、操作全屏交互式终端程序，以及无需在远程机器安装二进制文件即可通过 SSH 运行 Agent。产品还提供按任务复杂度自动路由模型、多 Agent 编排、云端交接和自定义模型路由等能力，并支持订阅、按量积分、API Key、OpenAI 兼容端点及 SuperGrok 登录等推理配置方式。
41.0 KB
143.4 KB
196.4 KB
16.3 MB
4.6 MB
159.4 KB
14.9 MB
111.2 KB
49.6 KB
33.7 KB
10.9 KB
5.8 KB
182.7 KB
中文
亮点：
- Warp Agent 现已作为独立 CLI 提供，可在任意环境中使用——包括 Ghostty、iTerm 2、VS Code、Windows 和 Mac 内置终端——为高频使用终端工作流的开发者提供更出色的体验。
- Agent 本身具备内置模型路由能力，内置前沿模型和美国托管的开放权重模型，还可通过自定义模型路由器进行定制。
- 该 CLI 原生支持高级 Agent 工作流，例如多 Agent 编排和云端 Agent。
- 该 CLI 基于 Warp 的终端基础设施构建，因此能够原生复用 Agent 会话——这解锁了其他 Agent 所不具备的能力。
今天，我们很高兴发布 Warp Agent CLI：一款新的独立 CLI，让你可以在任何地方使用 Warp Agent。它与内置于 Warp Terminal 的多模型 Agent 相同，现在可用于 Ghostty、iTerm 2、VSCode、Windows 内置终端，或任何你偏好的终端。
Warp Agent CLI 具备现代 CLI 编码 Agent 应有的一切：这是一个为专业开发者打造的多模型、成本优化型运行框架。开箱即可访问前沿模型和开放权重模型，并可根据任务复杂度自动路由。
该 CLI 的真正目标，是解决我们在其他 Agentic CLI 中看到的、与 Shell 本身集成效果相关的缺口——这对于在终端中进行实际工作的开发者尤为重要。
从技术上说，我们通过将 CLI Agent 构建在 Warp 独特的终端基础设施之上实现了这一点：可以将我们的 CLI Agent 视为跨 Agent 会话的内置复用器。它支持更自然的交互，例如在 Agent 会话中切换目录、让 Agent 驱动全屏终端命令（如 sqlite 和 mysql），甚至可以跨 SSH 会话运行，而无需在远程端安装二进制文件。
下面来看看让 Warp Agent CLI 对长期使用终端的用户极具实用价值的功能。
可复用会话的终端 Agent
Warp Agent CLI 采用独特的会话复用架构。在一个 Agent 会话中，我们实际上运行并管理 pty 连接，并在 Agent 与底层 Shell 之间设置了一层间接层，这与 tmux 的架构类似。复用功能由 Warp 的终端基础设施管理，因此我们的 Agent 原生感知终端输入与输出（在 Warp 中我们将其称为“块”）。这带来了比任何其他 CLI Agent 更丰富的体验。以下是它解锁的一些功能。
持久会话与远程 Agent
Warp Agent CLI 允许你在 Agent 会话中切换目录，甚至无需安装远程二进制文件即可在远程机器上运行 Agent。这一切都得益于：即使会话的基础“状态”发生变化，Agent 会话仍会持续存在。
在 Agent 会话中切换目录。
当你需要处理跨多个仓库的项目，或希望在安装软件权限受限的云端机器上获得 Agent 协助时，持久会话非常有用。
由 Agent 驱动的全屏与交互式应用
你可以通过 Agent 会话运行 sqlite 和 python 等全屏、交互式终端应用，Warp Agent 也能控制这些应用。因此，在 REPL 中，你可以要求 Agent 编写查询或以交互方式调试代码；你也可以让 Agent 退出 vim……
Agent 使用 vim 编辑文档
例如，你可以要求 Warp Agent：
- 在 REPL 中生成 SQL 查询，并交互式调试数据问题。
- 使用 gdb 等调试器设置断点并调试正在运行的应用。
- 使用 htop 监控系统性能。
无缝终端输入：自动识别与 Tab 补全
与其他 CLI Agent 一样，你可以使用 ! 执行 Shell 命令。但 Warp CLI 还支持自然语言识别：我们内置了一个分类器，可自动区分 Shell 命令和提示词，并相应调用 Shell 或 Agent。
Warp 中的命令补全菜单
Warp Agent CLI 还拥有 Warp Terminal 的标志性功能之一：Tab 补全。开始输入命令后按下 Tab，即可查看参数和标志位建议。
面向高级 Agent 工作流的优秀 UI
Warp Agent CLI 是为自主开发时代打造的，而不只是用于手动向 Agent 输入提示词。因此，它全面支持运行云端 Agent 和编排多 Agent 工作流。
Warp 的独特之处在于，当你将 Warp Agent 与其云平台结合使用时，我们的运行框架不仅能委派任务给使用不同模型的子 Agent，还能委派给 Claude Code 和 Codex 等完全不同的运行框架。
为云端 Agent 选择所用的 Agent 运行框架
自动 Agent 编排
Warp 的 Agent 是一个编排型 Agent。它会将困难任务委派给子 Agent 进行拆解，并通过原生界面让你轻松了解每个被编排的 Agent 正在执行什么。你甚至可以使用方向键在子 Agent 与编排 Agent 会话之间切换。
Warp 通过编排管理子 Agent
云端 Agent 交接
你可以在 CLI 中开始工作，然后在想要合上笔记本电脑时将其交接到云端。所有云端 Agent 都会被集中跟踪，并可通过 Web 进行监控和引导，因此你可以在移动中继续工作。
通过斜杠命令交接云端 Agent
成本优化且可自定义的运行框架
Warp Agent 是一个帕累托高效的运行框架，可根据任务复杂度自动路由。开箱即可访问前沿模型和开放权重模型。
我们还投入更多工作，让开放权重模型能够在我们的运行框架中表现出色，并为自定义模型路由器提供一流支持，使你可以定义由哪些模型处理哪些任务。
自定义路由器配置文件
如何开始使用
安装 CLI
Mac / Linux（终端）：
curl -fsSL https://app.warp.dev/download/agent-cli | bash
Windows（PowerShell）：
Invoke-RestMethod "https://app.warp.dev/download/agent-cli.ps1" | Invoke-Expression
配置推理服务
安装 CLI 后，你有三种开始方式：
- 购买 Warp 订阅：起价为每月 18 美元，包含每月 20 美元的推理额度。如果你已有 Warp 订阅，可以登录后立即开始使用。
- 购买按需积分：现已无需月度订阅，起价为 10 美元。
- 添加 API Key、OpenAI 兼容端点，或使用 SuperGrok 订阅登录。
今天就试试看，并告诉我们你的想法！
English / 原文
Highlights:
- The Warp Agent is now available as a standalone CLI that you can use anywhere – Ghostty, iTerm 2, VS Code, built-in Windows and Mac Terminals – and provides a superior experience for developers who do heavy terminal workflows
- The agent itself has built-in model-routing capabilities , comes with frontier and US-hosted open-weight models built in, and can even be customized with custom model routers.
- The CLI hasnative support for advanced agent workflows , like multi-agent orchestration and cloud agents.
- The CLI is built on Warp’s terminal infrastructure, which allows it to mux agent sessions natively – this unlocks capabilities beyond what is possible in other agents.
Today we are excited to launch the Warp Agent CLI, a new standalone CLI that lets you use the Warp Agent anywhere. It’s the same multi-model agent that’s built into Warp Terminal, now available in Ghostty, iTerm 2, VSCode, the built-in Windows terminal, or whatever terminal you prefer.
The Warp Agent CLI has everything you’d expect from a modern CLI coding agent: it’s a multi-model, cost-optimizing harness built for pro developers. Out of the box, you get access to frontier and open-weight models and auto-routing based on task complexity.
The CLI was truly built to fix gaps we were seeing with other agentic CLIs related to how well they integrate with the shell itself— especially important for developers doing real work in the terminal.
Technically, we accomplished this by building the CLI agent on top of Warp’s unique terminal infrastructure: think of our CLI agent as a built-in mux’er across agent sessions. This allows more natural interactions, like being able to switch directories in agent sessions, have the agent drive full-screen terminal commands (e.g. sqlite and mysql), and even run across ssh sessions with no remote binary install.
Let’s dive into the features that make the Warp Agent CLI extremely useful for folks who live in the terminal.
A multiplexing terminal agent
The Warp Agent CLI has a unique mux’ing architecture. Within an agent session, we are actually running and managing pty connections with a layer of indirection between the agent and the underlying shell, a similar architecture to how tmux works. The mux’ing is managed by Warp’s terminal infrastructure, so our agent is natively aware of terminal inputs and output (what we call “blocks” in Warp). This allows for a richer experience than any other CLI agent. Here are some of the features it unlocks.
Persistent sessions and remote agents
The Warp Agent CLI lets you switch directories while in an agent session, and even run agents on remote machines without installing remote binaries. This is all because the agent session continues even when the base “state” of the session changes.
Changing directories in an agent session.
Persistent sessions are useful when you are working across multi-repo projects, or when you want agent assistance on cloud machines where you might have limited permissions to install software.
Agent-driven full-screen and interactive apps
You can run full-screen and interactive terminal apps like sqlite and python through the agent session. Warp Agent can also control those apps. So within a REPL you can ask our agent to write queries or interactively debug code. Or you can ask the agent to quit vim…
Agent editing document using vim
For example, you can ask the Warp Agent to:
- Generate SQL queries within a REPL and interactively debug data issues
- Use a debugger like gdb to set breakpoints and debug running apps
- Monitor system performance with htop
Seamless terminal input: autodetection and tab-completion
As with other CLI agents, you can use ! to run a shell command. But Warp CLI also supports natural language detection: we ship with a classifier that automatically distinguishes shell commands from prompts, and invokes either the shell or agent accordingly.
Command completion menu in Warp
The Warp Agent CLI has one of Warp Terminal’s flagship features: tab completions. Start typing a command, hit tab, and see suggestions for arguments and flags.
Great UI for advanced agent workflows
The Warp Agent CLI is built for the age of autonomous development, not just manually prompting agents. As such, it has full support for running cloud agents and orchestrating multi-agent workflows.
Unique to Warp, when you couple the Warp Agent with our cloud platform, our harness can delegate not just across subagents with different models, but with entirely different harnesses like Claude Code and Codex.
Choosing with agent harness to use for a cloud agent
Automatic agent orchestration
Warp’s agent is an orchestration agent. It delegates to subagents to break apart hard tasks, with a native interface that makes it easy to see what each orchestrated agent is doing. You can even use arrow keys to switch between subagent and orchestrator agent sessions.
Warp managing subagents with orchestration
Cloud agent handoff
Start work in the CLI, then hand it off to the cloud when you want to close your laptop. All cloud agents are tracked centrally and can be monitored and steered via the web so you can continue work on the go.
Cloud agent handoff via slash command
Cost-optimizing and customizable harness
Warp Agent is a pareto-efficient harness with auto-routing based on task complexity. Out of the box, it comes with access to frontier models and open-weight models.
We also have invested more to make open weight models great within our harness, with first class support for custom model routers where you define which models handle which tasks.
Custom router configuration file
How to get started
Install the CLI
Mac / Linux (Terminal):
curl -fsSL https://app.warp.dev/download/agent-cli | bash
Windows (PowerShell):
Invoke-RestMethod "https://app.warp.dev/download/agent-cli.ps1" | Invoke-Expression
Configure inference
Once you’ve installed the CLI, you have three ways to get started:
- Purchase a subscription to Warp, starting at $18 a month for $20 of inference / month. If you have an existing Warp subscription, you’ll be able to login and get started immediately.
- Purchase ad hoc credits, now available without a monthly subscription, starting at $10.
- Add an API key, an OpenAI compatible endpoint, or log in with your SuperGrok subscription.