OpenClaw常见问题100问:从小龙虾是什么到高级配置
关于OpenClaw(小龙虾),你问的、搜的、担心的——都在这里。
一、基础概念(10问)
Q1:OpenClaw是什么? 一个开源的AI智能体系统。Logo是红色龙虾,社区昵称"小龙虾"。
Q2:它能做什么? 管理多个大模型、接入微信等IM平台、自动化日常工作流程、创建多个AI智能体同时干活。
Q3:它和ChatGPT的区别? ChatGPT是大模型产品(只能用自己的模型),OpenClaw是智能体系统(可以接入任何模型)。不是竞品,是互补。
Q4:免费吗? OpenClaw本身免费开源。但大模型的使用需要API Key(各平台有免费额度)。
Q5:需要什么配置? 最低:一台常开的电脑或服务器,2GB以上内存,Node.js ≥ 22.12.0。
Q6:在哪儿用? 平时在微信、飞书、钉钉、Discord里用——不需要打开专门的APP。
Q7:能同时用几个模型? 理论上不限制。你配了几个API Key就能同时调用几个。
Q8:支持中文吗? 完全支持。你可以用中文对话,OpenClaw管理界面也有中文。
Q9:和Claude/Gemini/Copilot的区别? 那三个是AI应用——帮你做特定的事。OpenClaw是系统——让AI融入你所有工作流。
Q10:适合什么人? 想在日常工具中嵌入AI的所有人——从个体户到开发者到企业团队。

二、安装与配置(12问)
Q11:怎么安装?
有Node.js的话终端运行一键安装命令,然后 openclaw onboard 完成配置。
Q12:手机上能装吗? 不能直接在手机上运行OpenClaw服务器。但装好之后可以在手机上的微信/Discord里使用。
Q13:Windows能装吗? 能。Windows/Mac/Linux都支持。
Q14:需要多少硬盘? OpenClaw本身占用很小(几十MB),主要是对话记录和文件会占用空间。建议10GB以上。
Q15:要公网IP吗? 通常不需要。OpenClaw通过WebSocket通信,局域网能用。接入微信需要扫码。
Q16:可以用Docker装吗? 可以,有官方Docker镜像。
Q17:系统提示词(System Prompt)是什么? 你给智能体设定的"角色说明"——定义它是什么身份、用什么语气、有什么能力。
Q18:怎么更新OpenClaw?
openclaw update 或重新运行安装命令。
Q19:能迁移到别的机器吗? 可以。备份配置文件和数据目录,在新机器上恢复。
Q20:有图形界面吗? 有。OpenClaw有Web管理面板,浏览器打开就能用。
Q21:最低Node.js版本? 22.12.0 及以上。
Q22:Mac M芯片能跑吗? 能,原生支持Apple Silicon。
三、大模型API(12问)
Q23:用什么模型最好? 根据场景选。日常→DeepSeek/Gemini,深度写作→Claude,综合→GPT-4o。
Q24:API Key从哪来? 各大模型平台的开发者后台申请。OpenAI/Anthropic/DeepSeek/Google等。
Q25:API费用贵吗? 不贵。DeepSeek新用户送500万tokens(够日常用几个月),Gemini有充足免费额度。
Q26:本地大模型和云端API有什么区别? 本地大模型跑在自己设备上,数据不出设备;云端API用的是远程服务器上的模型,数据会上传。
Q27:可以无API Key就用吗? 不行。OpenClaw需要至少一个大模型"大脑"才能工作。
Q28:API调用失败怎么办? 检查:①Key是否正确复制 ②是否有余额 ③网络是否正常 ④是否为供应商故障。
Q29:怎么切换模型? 在OpenClaw对话界面中直接切换,或在创建智能体时选择默认模型。
Q30:一个API Key能配多个智能体吗? 能。多个智能体可以共用同一个API Key和模型。
Q31:支持哪些国产模型? DeepSeek、通义千问、智谱GLM、文心一言、豆包、月之暗面等。
Q32:API Token是什么? 大模型处理文本的基本单位。中文:1 token ≈ 0.5-1.5个汉字。英文:1 token ≈ 0.75个词。
Q33:怎么控制API开销? 在API平台后台设月度消费上限,日常任务用便宜的模型,重要产出用贵的。
Q34:API Key能共享给别人吗? 不建议。API Key是敏感凭证,共享可能导致费用被消耗、功能被滥用。
四、IM接入(12问)
Q35:能接入微信吗? 能。支持个人微信和企业微信。个人微信接入有一定风控,建议用独立小号。
Q36:怎么接微信? OpenClaw面板→渠道配置→微信→生成二维码→微信扫码→完成。
Q37:微信总掉线? 个人微信接入的不稳定性是平台策略决定的。建议:用独立微信+定期检查+企业场景转企业微信。
Q38:支持Discord吗? 完全支持。创建Discord Bot→获取Token→填入OpenClaw→完成。
Q39:支持飞书/钉钉吗? 支持。在各自开放平台创建应用→填入OpenClaw面板即可。
Q40:能同时接多个平台吗? 能。一个OpenClaw实例可以同时接入微信、飞书、Discord等。
Q41:群里会不会刷屏? 设置触发规则:仅被@时回复,不自动参与所有聊天。
Q42:能接Telegram吗? 能。创建Bot→获取Token→填入OpenClaw。
Q43:私聊和群聊规则能分开设置吗? 能。可以设私聊自动回复,群聊仅@触发。
Q44:能设黑名单吗? 能。某些用户/群可以设为不回复。
Q45:企业微信和个人微信接入有什么区别? 企业微信更合规更稳定,适合团队。个人微信有风控,适合个人。
Q46:微信接入要付费吗? 个人微信接入免费。企业微信部分高级功能可能涉及平台侧费用。
五、智能体管理(10问)
Q47:什么是智能体? OpenClaw中的AI"角色"——有自己的身份、任务、系统提示词和默认模型。
Q48:能创建几个智能体? 理论无限制,取决于你的设备性能。
Q49:多个智能体能同时运行吗? 能。它们可以同时在线、互不干扰、处理不同的任务。
Q50:智能体之间能聊天吗? 可以配置多Agent协作,让智能体之间相互调用。
Q51:怎么分配给不同的人? 通过不同渠道配置,不同的群/账号可以绑定不同的智能体。
Q52:能复制智能体吗? 可以导出配置→导入创建新智能体→修改即可。
Q53:智能体提示词限制? 取决于你使用的模型。主流模型支持几万tokens的提示词。
Q54:能给智能体配知识库吗? 支持。可以给智能体上传文档作为知识来源。
Q55:智能体有长期记忆吗? 有。OpenClaw支持记忆系统,智能体可以记住之前的对话。
Q56:能训练自己的智能体吗? 完全自训练需要F1/G1级别硬件。但可以通过提示词工程+知识库大幅定制行为。
六、定时任务与自动化(8问)
Q57:OpenClaw支持定时任务吗? 支持。Cron模式——可以设置按分钟/小时/天/周的自动任务。
Q58:能做什么定时任务?
- 每天推送天气预报
- 定时检查邮件并通知
- 每日/每周工作总结
- 监控RSS更新并推送
- 定时提醒待办事项
Q59:Skill是什么? OpenClaw的功能扩展包。添加Skill就像给小龙虾装APP。
Q60:怎么装Skill? 社区有Skill市场,找到需要的安装即可。也可以自己写Skill。
Q61:有哪些实用的Skill? 天气查询、邮件处理、RSS监控、文件管理、搜索等。
Q62:能开发自己的Skill吗? 能。OpenClaw有Skill开发文档,支持自定义功能扩展。
Q63:Cron任务失败了会通知吗? 会。面板和日志中可以看到执行状态和失败原因。
Q64:能嵌套任务吗? 能。一个任务完成后可以自动触发下一个。
七、数据安全与隐私(8问)
Q65:我的数据安全吗? OpenClaw所有数据存储在你的设备上——对话记录、配置、知识库全部本地。
Q66:聊天记录会被上传到OpenClaw服务器吗? 不会。OpenClaw没有中心服务器,你的数据只在你自己设备上。
Q67:连接云端API时数据会上传吗? 会。当你调用GPT-4o、Claude等云端模型时,问题和上下文会发送到对应平台。
Q68:怎么保护数据不出域? 使用本地部署的大模型(需要E1及以上硬件),完全离线运行。
Q69:能加密对话吗? OpenClaw本地数据可配合设备级加密(如硬盘加密)。
Q70:API Key存在哪安全? 存在OpenClaw配置文件中(你的设备上)。不要复制Key发到公开平台。
Q71:会被第三方监控吗? 接云端API时数据会经过API平台。纯本地使用不涉及第三方。
Q72:适合处理客户隐私数据吗? 如果数据必须不出域,使用E1+本地大模型方案。A1+云端API方案不适合处理敏感隐私数据。
八、故障排查(12问)
Q73:OpenClaw启动不了? 检查:①Node.js版本≥22.12.0 ②端口是否被占用 ③配置文件是否正确。
Q74:智能体不回复? 检查:①模型配置是否正确 ②API Key是否有效余额 ③网络连接是否正常。
Q75:回复很慢? 可能原因:API服务商限流、网络延迟、设备性能不足。尝试切换模型或检查网络。
Q76:回复是乱码? 检查编码设置。在Web面板中通常正常,终端可能是编码问题。
Q77:微信连上了但收不到消息? 检查:①二维码是否过期 ②微信是否在线 ③OpenClaw服务是否在运行。
Q78:API报错429(请求太频繁)? API频率限制。降低调用频率或升级API套餐。
Q79:API报错401/403(未授权)? API Key无效或过期。重新生成Key并更新。
Q80:端口被占用? 修改配置文件中的端口号,或关闭占用端口的程序。
Q81:更新后不能用了? 尝试备份配置→清理重装→恢复配置。注意看更新日志兼容性说明。
Q82:内存占用太高? 减少同时运行的智能体数量,或减少每个智能体的对话历史长度。
Q83:服务自动停了? 检查是否被系统进程管理工具关闭,或设了定时任务。
Q84:浏览器访问不到管理面板? 确认OpenClaw正在运行,检查HTTP端口,确保同一网络。
九、购买铠盒A1相关(12问)
Q85:A1能代替我的电脑吗? 不能。A1是ARM芯片,4GB内存——是专用AI设备,不是通用计算机。
Q86:A1能本地跑大模型吗? 不能。A1走云端API。需要本地大模型请了解E1及以上。
Q87:A1和自己装OpenClaw有什么区别? A1免安装、0配置、24小时在线、超低功耗。自己装需要技术基础和维护。
Q88:A1需要额外的费用吗? 需要大模型API费用(各平台按量付费,有免费额度)。A1本身是一次性购买。
Q89:A1保修多久? 请访问官网或咨询客服了解最新保修政策。
Q90:A1耗电吗? 不耗电。典型功率5-12W——一个月不到5块钱电费。
Q91:A1声音大吗? 零噪音。全被动散热,没有风扇。
Q92:A1能装其他软件吗? 系统是Ubuntu,理论上可以。但不建议——A1专职跑AI,不要当通用服务器。
Q93:A1能连WiFi吗? 自带WiFi 6模块,但建议用有线(更稳定)。
Q94:A1放哪儿? 随便放。巴掌大(75×75×40mm),无声无热,放桌面/书架/电视柜都行。
Q95:A1和E1怎么选? A1=云端API,便宜。E1=本地跑大模型,性能和隐私更好。看你的需求和预算。
Q96:多台A1能组集群吗? OpenClaw支持多实例部署,可以组Agent集群。
还没找到答案?
访问 OpenClaw 官方文档 获取更详细的技术文档。
或加入 OpenClaw Discord 社区 向开发者和社区求助。
延伸阅读
OpenClaw FAQ: From 'What Is This' to Advanced Configuration
Everything you've asked, searched for, and worried about regarding OpenClaw — answered here.
1. Basic Concepts
Q1: What is OpenClaw? An open-source AI agent system. Its logo is a red lobster, hence the community nickname.
Q2: What can it do? Manage multiple LLMs, connect to messaging platforms, automate workflows, run multiple AI agents simultaneously.
Q3: How is it different from ChatGPT? ChatGPT is an LLM product (locked to one model). OpenClaw is an agent system (works with any model). Complementary, not competitive.
Q4: Is it free? OpenClaw itself is free and open-source. LLM usage requires API keys (free tiers available).
Q5: What hardware is required? Minimum: an always-on computer with 2GB+ RAM and Node.js ≥ 22.12.0.
Q6: Where do I use it? Inside Discord, Slack, Telegram, or other messaging platforms — no separate app needed.
Q7: How many models can I use at once? Practically unlimited. Every API key you configure becomes available.
Q8: Does it support languages other than English? Yes. Full multilingual support. You can use any language OpenClaw and your LLM support.
Q9: OpenClaw vs Claude / Gemini / Copilot? Those are AI applications. OpenClaw is a system that embeds AI across your workflows.
Q10: Who is it for? Anyone wanting AI embedded in their daily tools — from solopreneurs to developers to teams.
2. Installation & Setup
Q11: How do I install?
With Node.js ≥ 22.12.0, run the one-line install command, then openclaw onboard.
Q12: Can I run it on a phone? Not as a server. Once set up, you can use it from your phone via Discord, Telegram, etc.
Q13: Does it work on Windows? Yes. Windows, macOS, and Linux are all supported.
Q14: How much disk space? OpenClaw itself is small (\~tens of MB). Chat history and files need space. Recommend 10GB+.
Q15: Do I need a public IP? Usually no. OpenClaw uses WebSocket communication and works on a local network.
Q16: Docker available? Yes, there's an official Docker image.
Q17: What's a system prompt? The "role description" you give your agent — defining its identity, tone, and capabilities.
Q18: How do I update?
openclaw update or re-run the install command.
Q19: Can I migrate to another machine? Yes. Back up config files and data directories, restore on the new machine.
Q20: Is there a GUI? Yes. OpenClaw has a Web management panel accessible via browser.
Q21: Minimum Node.js version? 22.12.0 or higher.
Q22: Does it run on Apple Silicon? Yes, native support for M-series chips.
3. LLM APIs & Models
Q23: Which model is best? Depends on the task. Daily use → DeepSeek/Gemini. Deep writing → Claude. All-around → GPT-4o.
Q24: Where do I get API keys? From each provider's developer console. OpenAI, Anthropic, DeepSeek, Google, etc.
Q25: Are API fees expensive? No. DeepSeek offers 5M free tokens for new users. Gemini has a generous free tier.
Q26: What's the difference between local and cloud LLMs? Local runs on your device (data stays private). Cloud runs on a provider's servers (data is uploaded).
Q27: Can I use OpenClaw without an API key? No. OpenClaw needs at least one LLM "brain" to function.
Q28: API call failed — what do I do? Check: key correct? Balance remaining? Network ok? Provider outage?
Q29: How do I switch models? Switch directly in the conversation UI or set a default model per agent.
Q30: One API key for multiple agents? Yes. Multiple agents can share one API key and model.
Q31: What Chinese models are supported? DeepSeek, Tongyi Qianwen, Zhipu GLM, ERNIE Bot, Doubao, Moonshot, and more.
Q32: What's a token? The basic unit LLMs process. English: \~0.75 words per token. Other languages vary.
Q33: How do I control API costs? Set monthly spending caps. Use budget models for routine tasks and premium models for important work.
Q34: Can I share my API key? Not recommended. API keys are sensitive — sharing can lead to unauthorized charges and misuse.
4. Messaging Integration
Q35: Does it work with Discord? Fully supported. Create a bot, get the token, paste into OpenClaw.
Q36: Slack? Yes. Create a Slack app, enable bot features, paste the token.
Q37: Telegram? Yes. Create a bot via BotFather, paste the token.
Q38: WhatsApp? Possible with third-party bridges. Check the OpenClaw documentation for supported methods.
Q39: Multiple platforms at once? Yes. One OpenClaw instance can connect to Discord, Telegram, Slack, and others simultaneously.
Q40: Will it spam group chats? Set trigger rules: reply only when @mentioned. Don't auto-reply to every message.
Q41: Can I set different rules for DMs vs groups? Yes. DM auto-reply vs. group @mention-only can be configured separately.
Q42: Can I block certain users? Yes. Blocklists are supported.
Q43: Is there a message rate limit? You can configure rate limits to prevent flooding.
Q44: Does it handle threads? Depends on the platform. Discord threads are supported.
Q45: Can it send images/files? Yes, on most supported platforms.
5. Agent Management
Q46: What's an agent? An AI "role" in OpenClaw — with its own identity, task, system prompt, and default model.
Q47: How many agents can I create? Theoretically unlimited, limited only by your hardware.
Q48: Can multiple agents run simultaneously? Yes. They operate concurrently without interfering with each other.
Q49: Can agents talk to each other? Yes. Multi-agent collaboration is supported.
Q50: How do I assign agents to different people/channels? Different channels or groups can be bound to different agents.
Q51: Can I clone an agent? Export config → import as new agent → modify as needed.
Q52: Is there a prompt length limit? Depends on the model. Most modern models support tens of thousands of tokens.
Q53: Can agents have knowledge bases? Yes. Upload documents as knowledge sources for your agents.
Q54: Do agents have long-term memory? Yes. OpenClaw supports memory systems for persistent context.
Q55: Can I train my own agent? Full training requires F1/G1-level hardware. But prompt engineering + knowledge bases offer extensive customization.
6. Automation & Cron
Q56: Does OpenClaw support scheduled tasks? Yes. Cron-style scheduling: by minute, hour, day, or week.
Q57: What can I automate? Daily weather, email checks, work summaries, RSS monitoring, task reminders.
Q58: What is a Skill? An OpenClaw extension package. Adding a Skill is like installing an app.
Q59: How do I install Skills? Browse the community Skill marketplace or write your own.
Q60: Useful Skills? Weather, email processing, RSS monitoring, file management, web search.
Q61: Can I write my own Skills? Yes. OpenClaw has Skill development documentation.
Q62: Do failed cron jobs notify me? Yes. Check the panel and logs for execution status.
Q63: Can tasks chain together? Yes. One task can trigger the next on completion.
7. Security & Privacy
Q64: Is my data secure? All OpenClaw data is stored on your device — conversations, configs, knowledge bases. Everything is local.
Q65: Does chat history go to OpenClaw servers? No. OpenClaw has no central server. Your data stays on your device.
Q66: What happens when I use cloud APIs? When calling GPT-4o, Claude, etc., your queries and context are sent to that provider's servers.
Q67: How do I keep data fully private? Use locally deployed LLMs (requires E1+ hardware) for fully offline operation.
Q68: Is encryption supported? OpenClaw local data can be protected with device-level encryption (e.g., full-disk encryption).
Q69: Where should I store my API keys? In your OpenClaw config file (on your device). Never post them publicly.
Q70: Can third parties monitor usage? Cloud API usage goes through the provider. Pure local use involves no third party.
Q71: Is it suitable for sensitive customer data? If data must stay on-premises, use the E1+ local LLM solution. A1 + cloud APIs are not suitable for sensitive/regulated data.
8. Troubleshooting
Q72: OpenClaw won't start? Check: Node.js ≥ 22.12.0, port availability, config file validity.
Q73: Agent doesn't respond? Check: model config, API key validity/balance, network connection.
Q74: Slow responses? Possible: API rate limiting, network latency, insufficient hardware. Try switching models.
Q75: Garbled text in terminal? Encoding issue. The Web panel usually displays correctly.
Q76: Discord bot connected but not responding? Check: bot token valid, permissions correct, bot online, trigger rules configured.
Q77: Error 429 (too many requests)? API rate limit hit. Reduce call frequency or upgrade your API plan.
Q78: Error 401/403 (unauthorized)? API key is invalid or expired. Regenerate and update.
Q79: Port already in use? Change the port in config or stop the conflicting program.
Q80: Broken after update? Back up config → clean reinstall → restore config. Check changelog for breaking changes.
Q81: High memory usage? Reduce concurrent agents or shorten conversation history per agent.
Q82: Service stopped unexpectedly? Check if the system killed the process. Check for scheduled shutdown tasks.
Q83: Can't access the Web panel? Confirm OpenClaw is running, check the HTTP port, ensure same network.
9. KAIHE A1 & Hardware
Q84: What is the KAIHE A1? An entry-level Agent Computer with OpenClaw pre-installed. Plug and play.
Q85: Can the A1 replace my PC? No. The A1 is an ARM device with 4GB RAM — a dedicated AI appliance, not a general-purpose computer.
Q86: Can the A1 run LLMs locally? No. The A1 uses cloud APIs. For local LLMs, explore the E1 and above.
Q87: A1 vs DIY OpenClaw installation? A1 eliminates installation, configuration, and maintenance. DIY requires technical skills.
Q88: Any ongoing costs with the A1? LLM API fees (pay-per-use, free tiers available). The A1 hardware is a one-time purchase.
Q89: What's the warranty? Check the official website or contact support for current warranty details.
Q90: Is it power-hungry? No. 5-12W typical — under $1/month in electricity.
Q91: Is it noisy? Completely silent. Fully passive cooling, no fan.
Q92: Can I install other software? Ubuntu is the underlying OS, so technically yes. Not recommended — the A1 is purpose-built for AI.
Q93: Does it have WiFi? Built-in WiFi 6, but wired Ethernet is recommended for stability.
Q94: Where do I put it? Anywhere. Palm-sized (75×75×40mm), silent and cool. Desk, shelf, media cabinet — all fine.
Q95: A1 vs E1 — which to pick? A1 = cloud APIs, affordable. E1 = local LLMs, more power and privacy. Depends on your needs and budget.
Q96: Can I cluster multiple A1s? OpenClaw supports multi-instance deployment. You can create agent clusters.