离开座位,任务不停。
扫码连接手机,随时接着做。查看进展、补充想法、批准下一步,回到电脑时,一切都已衔接。
电脑需保持在线。仅配对你信任的设备,它们可访问你的全部会话。可随时在设置中撤销设备访问。

让代码跟上你的想法。 Loop,在本地工作的开源 coding harness。
npx @loop-harness/loop web
模型、节奏、工具与设备都由你掌握:在本地运行,按需接入,需要时从手机接管。
扫码连接手机,随时接着做。查看进展、补充想法、批准下一步,回到电脑时,一切都已衔接。
电脑需保持在线。仅配对你信任的设备,它们可访问你的全部会话。可随时在设置中撤销设备访问。

点击语音输入,把想法变成文字;开启语音通话,边聊边推进工作。识别与朗读都在你的电脑上完成,一套无需云端语音服务的本地方案。
需在支持的平台安装语音组件。手机音频经隧道中继传回电脑处理,不提供对中继隐藏音频的端到端加密。

把后续工作排好队,让 Loop 逐项推进。新想法随时补充,方向随时调整;需要暂停时就停下,准备好再继续。
快速处理日常工作,或深入解决复杂问题。自由选择适合的模型,也能接入自己的网关,让每个任务都有合适的搭档。
通过 MCP 连接常用工具,用 Skills 留住项目规范和拿手流程。少一些重复交代,让每次任务都用上积累的经验。
打开浏览器,留在终端,或接入自己的应用。 选择适合你的入口。
# 打开本地 Web 工作台
npx @loop-harness/loop web# 配置 DeepSeek,启动交互式 CLI
export LOOP_AI_API_KEY="your-deepseek-api-key"
npx @loop-harness/loop --provider deepseek --model deepseek-v4-pro阅读 CLI 文档// 在环境中设置 DEEPSEEK_API_KEY,再调用 SDK
import {
createAgentSession, createModelRuntime,
messageText, SessionManager,
} from "@loop-harness/loop";
const modelRuntime = createModelRuntime({
provider: "deepseek",
modelId: "deepseek-v4-pro",
apiKey: process.env.DEEPSEEK_API_KEY,
});
const model = modelRuntime.getModel("deepseek", "deepseek-v4-pro");
if (!model) throw new Error("Model not found");
const { session } = await createAgentSession({
modelRuntime, model,
sessionManager: SessionManager.inMemory(),
tools: [],
});
try {
await session.prompt("Hello!");
console.log(messageText(session.state.messages.at(-1)));
} finally {
await session.abort();
session.dispose();
}阅读 SDK 文档Let your code keep up with your ideas. Loop. An open-source coding harness that works locally.
npx @loop-harness/loop web
Your models, pace, tools and devices. Run locally, connect what you need, and pick up from your phone.
Scan to connect your phone and pick up where you left off. Check progress, share an idea or approve the next step. Everything stays in sync when you return to your desk.
Keep your computer online. Pair only devices you trust: they can access all your chats. Revoke access for any device in Settings at any time.

Click Voice input to turn ideas into text, or start a voice call to talk through the work. Speech recognition and playback run on your computer. A local voice solution, without cloud speech services.
Requires voice components on a supported platform. Phone audio passes through the tunnel relay for processing on your computer, without end-to-end encryption that hides it from the relay.

Line up what’s next and let Loop work through it. Add ideas as they come, change direction when you need to, and pause or pick up again on your terms.
Move quickly through everyday work or think deeper on harder problems. Choose the model that fits, or connect your own gateway. Give every task the right partner.
Connect your everyday tools through MCP and turn your team’s know-how into Skills. Spend less time repeating instructions and put proven workflows to work on every task.
Open your browser, stay in the terminal, or embed Loop in your app. Choose the entry point that suits you.
# Open the local Web workspace
npx @loop-harness/loop web# Configure DeepSeek and start the interactive CLI
export LOOP_AI_API_KEY="your-deepseek-api-key"
npx @loop-harness/loop --provider deepseek --model deepseek-v4-proRead the CLI docs// Set DEEPSEEK_API_KEY in your environment, then call the SDK
import {
createAgentSession, createModelRuntime,
messageText, SessionManager,
} from "@loop-harness/loop";
const modelRuntime = createModelRuntime({
provider: "deepseek",
modelId: "deepseek-v4-pro",
apiKey: process.env.DEEPSEEK_API_KEY,
});
const model = modelRuntime.getModel("deepseek", "deepseek-v4-pro");
if (!model) throw new Error("Model not found");
const { session } = await createAgentSession({
modelRuntime, model,
sessionManager: SessionManager.inMemory(),
tools: [],
});
try {
await session.prompt("Hello!");
console.log(messageText(session.state.messages.at(-1)));
} finally {
await session.abort();
session.dispose();
}Read the SDK docs