[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-openclaw-build-train-personal-ai-agent-en":3,"tags-openclaw-build-train-personal-ai-agent-en":30,"related-lang-openclaw-build-train-personal-ai-agent-en":40,"related-posts-openclaw-build-train-personal-ai-agent-en":44,"series-ai-agent-834cbbbd-ba53-48b7-b1c8-7ae11f13aa0d":81},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":10,"image_url":11,"keywords":12,"language":18,"translated_content":10,"views":19,"is_premium":20,"created_at":21,"updated_at":21,"cover_image":11,"published_at":22,"rewrite_status":23,"rewrite_error":10,"rewritten_from_id":24,"slug":25,"category":26,"related_article_id":27,"status":28,"google_indexed_at":29,"x_posted_at":10,"tweet_text":10,"title_rewritten_at":10,"title_original":10,"key_takeaways":10,"topic_cluster_id":10,"embedding":10,"is_canonical_seed":20},"834cbbbd-ba53-48b7-b1c8-7ae11f13aa0d","OpenClaw: How to build your own AI agent","\u003Cp>Claire Vo says she now chats with nine OpenClaw agents that handle work, code, sales, and family logistics. That is a lot of automation for a tool that only showed up in her life a few months ago. It also explains why OpenClaw has become one of the most talked-about personal AI projects in 2026.\u003C\u002Fp>\u003Cp>OpenClaw is an open-source system that runs locally, accepts instructions from Telegram, WhatsApp, Slack, or the terminal, and can use tools, skills, and scheduled jobs to act on your behalf. In Claire Vo’s write-up for \u003Ca href=\"https:\u002F\u002Fwww.lennysnewsletter.com\u002Fp\u002Fopenclaw-the-complete-guide-to-building\" target=\"_blank\" rel=\"noopener\">Lenny’s Newsletter\u003C\u002Fa>, the pitch is simple: teach it once, then let it keep working.\u003C\u002Fp>\u003Cp>That simplicity hides a lot of moving parts. If you want to get real value from OpenClaw, you need to understand how it is wired, where it should run, and what kind of job you should give it first. The setup is approachable, but it is also easy to make a mess if you treat it like a toy on your main laptop.\u003C\u002Fp>\u003Ch2>What OpenClaw actually is\u003C\u002Fh2>\u003Cp>OpenClaw is a local-first AI agent platform. It is built to receive messages, break work into actions, use tools, and keep going on a schedule. The agent can live on a Mac Mini, a VPS, or another owned machine, which matters because it may touch your files, browser, calendar, and email.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775196966682-r0z3.png\" alt=\"OpenClaw: How to build your own AI agent\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>Claire Vo describes it as more autonomous than the usual chatbot experience. That is the right framing. ChatGPT and Claude are great at answering questions, but OpenClaw is built to do work over time. It can wake up, check a task queue, inspect a workspace, and continue where it left off.\u003C\u002Fp>\u003Cp>The architecture is worth understanding before you install anything. OpenClaw uses a local gateway for incoming messages, separate agents with their own identities, scheduled jobs, and a heartbeat check every 30 minutes. It can also install skills on its own, which is where the power and the risk both come from.\u003C\u002Fp>\u003Cul>\u003Cli>Messages can arrive from Telegram, WhatsApp, Slack, or the terminal\u003C\u002Fli>\u003Cli>Agents keep their own workspaces and instructions\u003C\u002Fli>\u003Cli>Jobs can run on cron schedules and heartbeat checks\u003C\u002Fli>\u003Cli>Skills can add Gmail, Docs, browser, API, and CLI actions\u003C\u002Fli>\u003Cli>The system is meant to run on a machine you own and control\u003C\u002Fli>\u003C\u002Ful>\u003Cp>That local design is a big reason OpenClaw feels different from cloud-only assistants. The tradeoff is obvious: you get more control, but you also carry more responsibility. If the agent has access to your machine, it can reach more of your life than a browser tab ever could.\u003C\u002Fp>\u003Ch2>How to set it up without wrecking your day\u003C\u002Fh2>\u003Cp>The first setup rule is blunt: do not install OpenClaw on a computer you actively use for work or personal life. Claire Vo is very direct about this, and she is right. An agent with file access, browser access, and email access can make a tiny mistake that becomes a huge cleanup job.\u003C\u002Fp>\u003Cp>Her recommended setup options are practical. You can use a hosted version from a startup, run it on a VPS, or use a spare laptop or Mac Mini. She started on an old MacBook Air and later moved to a stack of Mac Minis. The lowest-end Mac Mini she mentions costs about $600 for an M4 model with 16GB of memory and 256GB of storage.\u003C\u002Fp>\u003Cp>If you want the least friction, a hosted service is easiest. If you want more control and do not mind setup work, a VPS can be cheaper. If you want the most hands-on experience, a dedicated Mac Mini is the sweet spot. It is also the option that feels most like owning a tiny robot desk worker.\u003C\u002Fp>\u003Cul>\u003Cli>Hosted OpenClaw: easiest to start, least setup pain\u003C\u002Fli>\u003Cli>VPS: cheaper in some cases, but more technical\u003C\u002Fli>\u003Cli>Mac Mini or spare laptop: higher upfront cost, more control\u003C\u002Fli>\u003Cli>Recommended memory: 16GB if you want a comfortable start\u003C\u002Fli>\u003C\u002Ful>\u003Cp>Before installation, she suggests a fresh admin account, a Gmail address for the agent, and Chrome on the machine. Those steps sound boring, but they reduce the chance that your agent ends up mixed into your daily computer habits. Boring prep is what keeps ambitious automation from becoming a support ticket.\u003C\u002Fp>\u003Cp>The install itself is a one-liner from the terminal: \u003Ccode>curl -fsSL https:\u002F\u002Fopenclaw.ai\u002Finstall.sh | bash\u003C\u002Fcode>. After that, OpenClaw drops you into onboarding. If terminal navigation is new to you, the learning curve is mostly about arrows, spacebar, and enter. That is a good reminder that this is still developer tooling, even if the marketing makes it sound friendly.\u003C\u002Fp>\u003Ch2>Why the onboarding matters more than the install\u003C\u002Fh2>\u003Cp>OpenClaw’s real setup work starts after the installer finishes. You pick a model, authenticate with a provider, choose a chat channel, decide whether to enable web search, and install bundled skills. The defaults matter because they shape how much the agent can do on day one.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775196962346-zr9g.png\" alt=\"OpenClaw: How to build your own AI agent\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>Claire Vo recommends Claude Opus 4.6 or Codex 5.4 as the default model, or simply the strongest model available when you read this. She also recommends using an API key rather than reusing a consumer subscription, which is the safer path if you want fewer surprises with account policies.\u003C\u002Fp>\u003Cp>Telegram is her preferred beginner channel, and that tracks. It is easier to use than the terminal, fast on mobile, and good enough for back-and-forth coordination. Once the agent is alive, you can talk to it like you would talk to a teammate who works odd hours and never needs coffee.\u003C\u002Fp>\u003Cblockquote>“If you’re paying attention to AI news, you’ve probably heard about OpenClaw.” — Claire Vo, Lenny’s Newsletter\u003C\u002Fblockquote>\u003Cp>That quote gets at the real reason this project has traction. OpenClaw is not a demo that lives in a browser window. It is an operating layer for ongoing tasks, and it makes that promise in a way that feels uncomfortably close to practical.\u003C\u002Fp>\u003Cp>The onboarding also creates the agent’s memory files. These live in a workspace folder, usually \u003Ccode>.openclaw\u002F[agent_name]-workspace\u003C\u002Fcode>, and include files such as \u003Ccode>AGENTS.md\u003C\u002Fcode>, \u003Ccode>SOUL.md\u003C\u002Fcode>, \u003Ccode>IDENTITY.md\u003C\u002Fcode>, \u003Ccode>TOOLS.md\u003C\u002Fcode>, and \u003Ccode>USER.md\u003C\u002Fcode>. That setup is clever because it turns personality, boundaries, and instructions into editable documents instead of hidden magic.\u003C\u002Fp>\u003Ch2>What OpenClaw can do compared with other tools\u003C\u002Fh2>\u003Cp>OpenClaw is often compared with hosted assistants, but the comparison is incomplete. A chatbot can draft text. OpenClaw can keep a recurring job alive, pull from web search, inspect a repo, update files, and ship a pull request. That difference matters when you want automation that lasts longer than a single chat session.\u003C\u002Fp>\u003Cp>There is also a growing ecosystem around this style of agent. Claire Vo mentions hosted products like \u003Ca href=\"https:\u002F\u002Fwww.railway.app\" target=\"_blank\" rel=\"noopener\">Railway\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fwww.digitalocean.com\" target=\"_blank\" rel=\"noopener\">DigitalOcean\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fcloud.google.com\" target=\"_blank\" rel=\"noopener\">Google Cloud\u003C\u002Fa>, and \u003Ca href=\"https:\u002F\u002Frender.com\" target=\"_blank\" rel=\"noopener\">Render\u003C\u002Fa> for deployment. For messaging, \u003Ca href=\"https:\u002F\u002Ftelegram.org\" target=\"_blank\" rel=\"noopener\">Telegram\u003C\u002Fa> is the simplest entry point. For model access, \u003Ca href=\"https:\u002F\u002Fopenai.com\" target=\"_blank\" rel=\"noopener\">OpenAI\u003C\u002Fa> and \u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\" target=\"_blank\" rel=\"noopener\">Anthropic\u003C\u002Fa> are the obvious providers.\u003C\u002Fp>\u003Cul>\u003Cli>\u003Ca href=\"https:\u002F\u002Fopenai.com\u002Fchatgpt\" target=\"_blank\" rel=\"noopener\">ChatGPT\u003C\u002Fa> is better for one-off conversations and drafts\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fclaude.ai\" target=\"_blank\" rel=\"noopener\">Claude\u003C\u002Fa> is strong for reasoning and writing, but not an always-on system by itself\u003C\u002Fli>\u003Cli>OpenClaw is built for recurring work, local control, and custom skills\u003C\u002Fli>\u003Cli>Hosted agent products may be easier, but they often hide the wiring\u003C\u002Fli>\u003Cli>OpenClaw asks for more setup and gives back more control\u003C\u002Fli>\u003C\u002Ful>\u003Cp>The risk profile is different too. Claire Vo mentions horror stories such as an inbox wipe and a calendar mistake. That is the honest part of the story, and it matters. Any tool that can act on your behalf can also act badly on your behalf if your instructions, permissions, or checks are sloppy.\u003C\u002Fp>\u003Cp>Still, the upside is hard to ignore. If you want an AI that can keep a project moving, update internal pages, check your schedule, or draft outreach after a lead arrives, OpenClaw is closer to a junior operator than a chatbot. That is why the product has attracted attention well beyond the usual AI hobbyist crowd.\u003C\u002Fp>\u003Ch2>How Claire Vo actually uses it day to day\u003C\u002Fh2>\u003Cp>Claire Vo’s examples are useful because they are ordinary. One agent reads email and calendar items before she checks her phone. Another reminds her husband about a school event. Another drafts a sales email after a prospect reaches out. This is the kind of automation that saves attention, not just time.\u003C\u002Fp>\u003Cp>Her favorite use cases are personal assistant work, weekend coordination, and ongoing business maintenance. Those jobs are repetitive enough to automate, but messy enough that a human still wants oversight. That is a good fit for an agent that can work continuously and ask for clarification when needed.\u003C\u002Fp>\u003Cp>She also emphasizes that the first task should be narrow. Pick one job, train the agent on that job, and expand later. That advice sounds obvious, but it is where most people fail with agent tools. They try to make the system do everything on day one, then blame the tool when the scope was the problem.\u003C\u002Fp>\u003Cp>There is a practical lesson in her setup files too. By storing identity, tone, tools, and user context in Markdown, OpenClaw makes the agent editable. You can inspect the behavior, change the instructions, and improve the system without waiting for a product team to expose a new setting.\u003C\u002Fp>\u003Cp>That is also why OpenClaw feels more like software infrastructure than a consumer app. It asks you to think like a manager, a systems admin, and a product owner at the same time. If that sounds annoying, it is. If that sounds powerful, it probably is.\u003C\u002Fp>\u003Ch2>What to do next if you want to try it\u003C\u002Fh2>\u003Cp>If you are curious, start with a spare machine and one narrow job. Do not begin with email triage across your entire life. Start with something bounded, like weekend planning, a single project reminder flow, or a weekly status report that the agent drafts and you approve.\u003C\u002Fp>\u003Cp>The best prediction here is simple: the first wave of personal AI agents will belong to people who are willing to do setup work and define clear boundaries. OpenClaw will probably not replace your calendar app or your inbox, but it can become the layer that keeps those tools moving without constant manual nudging. The question is whether you want that layer to live on your main computer, or on a box you control and can wipe if needed.\u003C\u002Fp>\u003Cp>If you try it, treat the first week like a pilot, not a commitment. Give the agent one job, inspect its outputs, tighten the instructions, and only then expand its access. That is the difference between a useful assistant and an expensive cleanup job.\u003C\u002Fp>","OpenClaw is an open-source AI agent you can run locally, message from Telegram, and train to handle email, calendars, code, and more.","www.lennysnewsletter.com","https:\u002F\u002Fwww.lennysnewsletter.com\u002Fp\u002Fopenclaw-the-complete-guide-to-building",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775196966682-r0z3.png",[13,14,15,16,17],"OpenClaw","AI agent","Telegram","local automation","personal assistant","en",0,false,"2026-04-03T06:15:46.820275+00:00","2026-04-03T06:15:46.798+00:00","done","9310fdae-5b58-4b4e-b292-aa565a3257e5","openclaw-build-train-personal-ai-agent-en","ai-agent","83ca74a3-1c05-4acb-978f-70b3efff1730","published","2026-04-07T07:41:11.023+00:00",[31,33,35,36,38],{"name":15,"slug":32},"telegram",{"name":13,"slug":34},"openclaw",{"name":14,"slug":26},{"name":16,"slug":37},"local-automation",{"name":17,"slug":39},"personal-assistant",{"id":27,"slug":41,"title":42,"language":43},"openclaw-build-train-personal-ai-agent-zh","OpenClaw：自己打造 AI 代理人","zh",[45,51,57,63,69,75],{"id":46,"slug":47,"title":48,"cover_image":49,"image_url":49,"created_at":50,"category":26},"c5d4bc11-1f4d-438c-b644-a8498826e1ab","claude-agent-dreaming-outcomes-multiagent-en","Claude给Agent加了“做梦”功能","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778868649463-f5qv.png","2026-05-15T18:10:25.29539+00:00",{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":26},"fda44d24-7baf-4d91-a7f9-bbfecae20a27","switch-ai-outputs-markdown-to-html-en","How to Switch AI Outputs from Markdown to HTML","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778743249827-wmsr.png","2026-05-14T07:20:22.631724+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":26},"064275f5-4282-47c3-8e4a-60fe8ac99246","anthropic-cat-wu-proactive-ai-assistants-en","Anthropic’s Cat Wu on proactive AI assistants","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778735465548-a92i.png","2026-05-14T05:10:31.723441+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":26},"423ac8ad-2886-42a9-8dd8-78e5d43a1574","how-to-run-hermes-agent-on-discord-en","How to Run Hermes Agent on Discord","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778724656141-i30t.png","2026-05-14T02:10:35.727086+00:00",{"id":70,"slug":71,"title":72,"cover_image":73,"image_url":73,"created_at":74,"category":26},"776a562c-99a6-4a6b-93a0-9af40300f3f2","why-ragflow-is-the-right-open-source-rag-engine-to-self-host-en","Why RAGFlow is the right open-source RAG engine to self-host","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778674254587-0pxn.png","2026-05-13T12:10:25.721583+00:00",{"id":76,"slug":77,"title":78,"cover_image":79,"image_url":79,"created_at":80,"category":26},"322ec8bc-61d3-4c80-bb9e-a19941e137c6","how-to-add-temporal-rag-in-production-en","How to Add Temporal RAG in Production","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778667085221-0mox.png","2026-05-13T10:10:31.619892+00:00",[82,87,92,97,102,107,112,117,122,127],{"id":83,"slug":84,"title":85,"created_at":86},"03db8de8-8dc2-4ac1-9cf7-898782efbb1f","anthropic-claude-ai-agent-task-automation-en","Anthropic's Claude AI Agent: A New Era of Task Automation","2026-03-25T16:25:06.513026+00:00",{"id":88,"slug":89,"title":90,"created_at":91},"045d1abc-190d-4594-8c95-91e2a26f0c5a","googles-2026-ai-agent-report-decoded-en","Google’s 2026 AI Agent Report, Decoded","2026-03-26T11:15:23.046616+00:00",{"id":93,"slug":94,"title":95,"created_at":96},"e64aba21-254b-4f93-aa21-837484bb52ec","kimi-k25-review-stronger-still-not-legend-en","Kimi K2.5 review: stronger, still not a legend","2026-03-27T07:15:55.385951+00:00",{"id":98,"slug":99,"title":100,"created_at":101},"30dfb781-a1b2-4add-aebe-b3df40247c37","claude-code-controls-mac-desktop-en","Claude Code now controls your Mac desktop","2026-03-28T03:01:59.384091+00:00",{"id":103,"slug":104,"title":105,"created_at":106},"254405b6-7833-4800-8e13-f5196deefbe6","cloudflare-100x-faster-ai-agent-sandbox-en","Cloudflare’s 100x Faster AI Agent Sandbox","2026-03-28T03:09:44.356437+00:00",{"id":108,"slug":109,"title":110,"created_at":111},"04f29b7f-9b91-4306-89a7-97d725e6e1ba","openai-backs-isara-agent-swarm-bet-en","OpenAI backs Isara’s agent-swarm bet","2026-03-28T03:15:27.849766+00:00",{"id":113,"slug":114,"title":115,"created_at":116},"3b0bf479-e4ae-4703-9666-721a7e0cdb91","openai-plan-automated-ai-researcher-en","OpenAI’s plan for an automated AI researcher","2026-03-28T03:17:42.312819+00:00",{"id":118,"slug":119,"title":120,"created_at":121},"fe91bce0-b85d-4efa-a207-24ae9939c29f","harness-engineering-ai-agent-reliability-2026","Harness Engineering: From Bridle to Operating System, The Missing Link in AI Agent Reliability","2026-03-31T06:36:55.648751+00:00",{"id":123,"slug":124,"title":125,"created_at":126},"67dc66da-ca46-4aa5-970b-e997a39fe109","openai-codex-plugin-claude-code-en","OpenAI puts Codex inside Claude Code","2026-04-01T09:21:55.381386+00:00",{"id":128,"slug":129,"title":130,"created_at":131},"7a09007d-820f-43b3-8607-8ad1bfcb94c8","mcp-explained-from-prompts-to-production-en","MCP Explained: From Prompts to Production","2026-04-01T09:24:40.089177+00:00"]