[{"data":1,"prerenderedAt":2687},["ShallowReactive",2],{"blog-post-free-openclaw-agent-openrouter-setup":3,"related-posts-free-openclaw-agent-openrouter-setup":356},{"id":4,"title":5,"author":6,"body":10,"category":334,"date":335,"description":336,"extension":337,"featured":338,"image":339,"meta":340,"navigation":341,"path":342,"readingTime":343,"seo":344,"seoTitle":345,"stem":346,"tags":347,"updatedDate":335,"__hash__":355},"blog/blog/free-openclaw-agent-openrouter-setup.md","How to Run a Free OpenClaw Agent in 5 Minutes Using OpenRouter",{"name":7,"role":8,"avatar":9},"Shabnam Katoch","Growth Head","/img/avatars/shabnam-profile.jpeg",{"type":11,"value":12,"toc":320},"minimark",[13,20,23,26,29,32,35,38,41,46,57,60,63,70,74,82,85,88,94,98,106,109,112,115,119,122,129,135,141,147,150,158,162,165,168,181,184,187,191,194,197,200,203,211,215,218,221,224,227,235,239,242,245,248,256,260,263,266,273,276,280,285,288,293,296,301,304,309,312,317],[14,15,16],"p",{},[17,18,19],"em",{},"No API bill. No credit card. No infrastructure headaches. Here's exactly how we did it.",[14,21,22],{},"Someone dropped a comment on one of our Reddit threads last week that stopped me mid-scroll.",[14,24,25],{},"\"BYOK sounds great but what if I don't want to pay for an API key either?\"",[14,27,28],{},"Fair. Really fair.",[14,30,31],{},"We've been saying \"bring your own API keys\" like it's the generous option. But for someone who just wants to test whether an AI agent is actually useful before spending a dollar, even getting an OpenRouter key feels like one more step in a wall of friction.",[14,33,34],{},"So we tried something. We set up a completely working OpenClaw agent, on BetterClaw's free tier, using only free models from OpenRouter.",[14,36,37],{},"$0 total. Not \"$5 free credits.\" Not \"basically free.\" Zero.",[14,39,40],{},"Here's exactly what we did, what we ran into, and what you should know before you try it.",[42,43,45],"h2",{"id":44},"step-1-get-a-free-api-key-from-openrouter-2-minutes","Step 1: Get a Free API Key from OpenRouter (2 Minutes)",[14,47,48,49,56],{},"Go to ",[50,51,55],"a",{"href":52,"rel":53},"https://openrouter.ai/",[54],"nofollow","openrouter.ai",". Sign up. That's it.",[14,58,59],{},"You now have access to 30+ free models. The ones worth knowing about for agent work: Llama 3.3 70b, DeepSeek R1, and Qwen3 Coder 480b. The base limit is 50 requests per day on most free models, but some go up to 1,000 free requests per day.",[14,61,62],{},"For a daily briefing agent or a lightweight personal assistant? 1,000 requests per day is more than enough. Most real-world agent usage runs 10 to 30 requests per session.",[14,64,65],{},[66,67],"img",{"alt":68,"src":69},"OpenRouter free tier dashboard showing Llama 3.3 70b, DeepSeek R1, and Qwen3 Coder 480b with daily request limits ranging from 50 to 1,000 free requests per day","/img/blog/free-openclaw-agent-openrouter-setup-openrouter-models.jpg",[42,71,73],{"id":72},"step-2-sign-up-for-betterclaw-free-tier-2-minutes","Step 2: Sign Up for BetterClaw Free Tier (2 Minutes)",[14,75,48,76,81],{},[50,77,80],{"href":78,"rel":79},"https://app.betterclaw.io/sign-in",[54],"the BetterClaw app",". No card. No trial countdown. Takes about 2 minutes.",[14,83,84],{},"When it asks for your API key, paste in the OpenRouter key you just generated. Then select one of the free models as your default.",[14,86,87],{},"Which free model should you pick? Honestly, for most agent tasks, Llama 3.3 70b or DeepSeek R1 handle daily briefings, summarization, email triage, and basic research just fine. They're not Claude Sonnet. But for a free agent doing routine tasks, they're more than good enough.",[14,89,90],{},[66,91],{"alt":92,"src":93},"BetterClaw LLM configuration screen showing OpenRouter API key field and free model dropdown with Llama 3.3 70b selected as default for the agent","/img/blog/free-openclaw-agent-openrouter-setup-llm-config.jpg",[42,95,97],{"id":96},"step-3-connect-your-channel-1-minute","Step 3: Connect Your Channel (1 Minute)",[14,99,100,101,105],{},"BetterClaw connects to 15+ platforms out of the box. For a free setup, Telegram is the cleanest option. Takes about 60 seconds to get a bot token from BotFather and paste it into the channel config. For the ",[50,102,104],{"href":103},"/openclaw-telegram-setup","step-by-step Telegram walkthrough",", our setup guide covers BotFather and pairing.",[14,107,108],{},"If you want to connect to Slack, Discord, or WhatsApp instead, the process is similar. The agent starts responding on whatever channel you pick.",[14,110,111],{},"And that's it. You're done.",[14,113,114],{},"Five minutes from nothing to a working AI agent. Cost: $0.",[42,116,118],{"id":117},"whats-the-catch-with-free-models","\"What's the Catch With Free Models?\"",[14,120,121],{},"Here's where we'll be straight with you.",[14,123,124,128],{},[125,126,127],"strong",{},"Free models are slower than paid models."," You'll notice the latency. Not unbearable, but noticeable. Expect 3 to 8 seconds per response instead of under 2.",[14,130,131,134],{},[125,132,133],{},"Complex multi-step tasks get shaky."," If you need your agent to run a 10-step research chain with tool calls at each step, free models stumble. They'll misinterpret instructions, skip steps, or hallucinate a tool result. Single-step and two-step tasks? Totally fine.",[14,136,137,140],{},[125,138,139],{},"Rate limits exist."," You're on shared infrastructure. During peak hours you might get queued. Not often, but it happens.",[14,142,143,146],{},[125,144,145],{},"Quality varies by model and by day."," Some days DeepSeek R1 is sharp. Some days it rambles. You learn which model handles which task better over time. This is the honest part of \"free.\"",[14,148,149],{},"But for a \"try before you spend\" setup or a simple daily assistant use case, free models work better than most people expect. We were surprised.",[14,151,152,153,157],{},"If you're curious about how different models perform on OpenClaw tasks generally, we've done a ",[50,154,156],{"href":155},"/blog/best-llm-for-openclaw-glm-5-1-claude-sonnet-minimax","detailed LLM comparison for OpenClaw use cases"," that goes deeper into benchmarks and tradeoffs.",[42,159,161],{"id":160},"will-i-hit-the-100-task-limit","\"Will I Hit the 100 Task Limit?\"",[14,163,164],{},"With this setup? Probably not in month one.",[14,166,167],{},"Here's the rough math:",[169,170,171,175,178],"ul",{},[172,173,174],"li",{},"1 daily briefing cron: ~30 tasks per month",[172,176,177],{},"1 weekly report cron: ~4 tasks per month",[172,179,180],{},"15 to 20 ad-hoc requests per week: ~70 tasks per month",[14,182,183],{},"Total: roughly 100. Tight, but workable if you're not running 5 daily crons.",[14,185,186],{},"If you find yourself constantly hitting the limit, that's the signal the agent is useful enough to upgrade. Month one on free? You'll be fine.",[42,188,190],{"id":189},"why-we-built-this-option","Why We Built This Option",[14,192,193],{},"We've talked to a lot of people who got interested in OpenClaw after seeing it hit 230,000+ GitHub stars and land on the front page of Hacker News. They followed a setup tutorial, ran into Docker issues or YAML configs, and quietly gave up.",[14,195,196],{},"The OpenClaw maintainer himself once warned: \"if you can't understand how to run a command line, this is far too dangerous of a project for you to use safely.\"",[14,198,199],{},"That's a real barrier. Not everyone needs to cross it just to try an AI agent.",[14,201,202],{},"BetterClaw exists because we think more people should be able to experience what a well-configured autonomous agent actually feels like, without the infrastructure tax. The free tier plus OpenRouter's free models is the lowest-friction version of that we've been able to build.",[14,204,205,206,210],{},"If you've been curious about ",[50,207,209],{"href":208},"/openclaw-hosting","what managed OpenClaw hosting actually includes"," versus spinning something up on a VPS yourself, that page walks through the full comparison.",[42,212,214],{"id":213},"the-moment-it-actually-starts-feeling-useful","The Moment It Actually Starts Feeling Useful",[14,216,217],{},"Here's what we didn't expect to be true: the first time your free-tier agent quietly runs a morning briefing, summarizes your overnight Slack threads, or answers a question without you lifting a finger, something clicks.",[14,219,220],{},"It's not the technology that lands. It's the time.",[14,222,223],{},"Most people who try this setup report the same thing. The agent does something useful. They go back to their day. Then they check the output an hour later and think, \"I would have spent 20 minutes on that.\"",[14,225,226],{},"That's when the \"is this worth paying for\" question answers itself.",[14,228,229,230,234],{},"If you want your OpenClaw agent running in 60 seconds with your own API keys and no usage caps, ",[50,231,233],{"href":232},"/pricing","BetterClaw's Pro plan is $19/month per agent"," (up to 25 agents, each billed at $19/month). Bring your own keys, pick any of 28+ model providers, and the infrastructure is completely managed. No Docker. No YAML. No 2 AM debugging.",[42,236,238],{"id":237},"what-happens-when-youre-ready-to-upgrade","What Happens When You're Ready to Upgrade",[14,240,241],{},"The free tier is a starting point, not a ceiling.",[14,243,244],{},"When you move to Pro at $19/month, you get persistent memory with hybrid vector and keyword search, real-time health monitoring, auto-pause on anomalies, and multi-channel support from a single agent. You can also swap in Claude Sonnet or GPT-4o the moment you want sharper reasoning on complex tasks.",[14,246,247],{},"The upgrade takes about 30 seconds inside the same dashboard.",[14,249,250,251,255],{},"If you've been running agents on a VPS or trying to self-host OpenClaw and the maintenance burden is getting old, take a look at ",[50,252,254],{"href":253},"/compare/self-hosted","how BetterClaw compares to self-hosting and managed alternatives",". The hidden costs of DIY infrastructure add up faster than most people realize.",[42,257,259],{"id":258},"give-it-a-try","Give It a Try",[14,261,262],{},"If you've been on the fence about whether AI agents are actually useful, this is the lowest-stakes test you can run.",[14,264,265],{},"No credit card. No infrastructure decision. No API bill waiting at the end of the month.",[14,267,268,272],{},[50,269,271],{"href":78,"rel":270},[54],"Sign up for BetterClaw's free tier",", grab a free OpenRouter key, and have a working agent in 5 minutes. If it's useful, you'll know. If it's not, you've lost nothing but 5 minutes.",[14,274,275],{},"We handle the infrastructure. You handle the interesting part.",[42,277,279],{"id":278},"frequently-asked-questions","Frequently Asked Questions",[14,281,282],{},[125,283,284],{},"What is a free OpenClaw agent and how does it work?",[14,286,287],{},"A free OpenClaw agent is a fully functional AI assistant built on OpenClaw, the open-source agent framework with 230,000+ GitHub stars, deployed on BetterClaw's free tier and powered by OpenRouter's free model tier. BetterClaw handles the hosting, security, and channel connections. OpenRouter provides access to models like Llama 3.3 70b and DeepSeek R1 at no cost. You get a working autonomous agent at $0.",[14,289,290],{},[125,291,292],{},"How does OpenRouter's free tier compare to a paid API key for OpenClaw?",[14,294,295],{},"OpenRouter's free models are slower (3 to 8 seconds per response vs. under 2 for paid) and less reliable on complex multi-step tasks. For simple daily tasks like briefings, summarization, and research lookups, the quality gap is small. For sophisticated chains with multiple tool calls, a paid model like Claude Sonnet or GPT-4o will perform significantly better. The free setup is ideal for evaluation and lightweight personal use.",[14,297,298],{},[125,299,300],{},"How long does it take to set up a free OpenClaw agent with BetterClaw?",[14,302,303],{},"About 5 minutes total. Getting a free API key from OpenRouter takes roughly 2 minutes. Signing up for BetterClaw's free tier takes another 2 minutes. Connecting your Telegram channel takes about 1 minute. The agent starts responding immediately after setup. No Docker, no YAML, no terminal required.",[14,305,306],{},[125,307,308],{},"Is BetterClaw's free tier actually free, or are there hidden costs?",[14,310,311],{},"The free tier is genuinely free. No credit card required, no trial period, no automatic upgrade. You get 1 agent slot and up to 100 tasks per month at $0. If you use OpenRouter's free models, your total monthly cost is $0. If you exceed the task limit or want persistent memory and multi-channel support, the Pro plan is $19/month per agent (up to 25 agents, each billed at $19/month) with bring-your-own API keys.",[14,313,314],{},[125,315,316],{},"Is it safe to run an OpenClaw agent on a free plan?",[14,318,319],{},"Yes. BetterClaw runs all agents in Docker-sandboxed execution environments with AES-256 encryption for credentials, regardless of which plan you're on. The security architecture is the same on free as on Pro. Your API keys are never stored in plaintext. Given that security researchers have found over 30,000 internet-exposed OpenClaw instances without authentication, using a managed platform with built-in sandboxing is significantly safer than a self-hosted setup, especially for someone new to agent infrastructure.",{"title":321,"searchDepth":322,"depth":322,"links":323},"",2,[324,325,326,327,328,329,330,331,332,333],{"id":44,"depth":322,"text":45},{"id":72,"depth":322,"text":73},{"id":96,"depth":322,"text":97},{"id":117,"depth":322,"text":118},{"id":160,"depth":322,"text":161},{"id":189,"depth":322,"text":190},{"id":213,"depth":322,"text":214},{"id":237,"depth":322,"text":238},{"id":258,"depth":322,"text":259},{"id":278,"depth":322,"text":279},"Guides","2026-04-25","Run a fully working OpenClaw agent for $0 using BetterClaw's free tier and OpenRouter's free models. No credit card. No Docker. Setup takes 5 minutes.","md",false,"/img/blog/free-openclaw-agent-openrouter-setup.jpg",{},true,"/blog/free-openclaw-agent-openrouter-setup","7 min read",{"title":5,"description":336},"Free OpenClaw Agent Setup With OpenRouter in 5 Minutes","blog/free-openclaw-agent-openrouter-setup",[348,349,350,351,352,353,354],"free openclaw agent","openrouter free api key","openclaw free setup","betterclaw free tier","run openclaw for free","openclaw no credit card","free ai agent setup","FrnZDvShwP2WKfVbsHDvkjg_vxPLd9SZnHGZDw081es",[357,1213,2225],{"id":358,"title":359,"author":360,"body":361,"category":334,"date":1195,"description":1196,"extension":337,"featured":338,"image":1197,"meta":1198,"navigation":341,"path":1199,"readingTime":1200,"seo":1201,"seoTitle":1202,"stem":1203,"tags":1204,"updatedDate":1195,"__hash__":1212},"blog/blog/openclaw-ollama-guide.md","OpenClaw + Ollama: What Works and What Doesn't (2026)",{"name":7,"role":8,"avatar":9},{"type":11,"value":362,"toc":1180},[363,368,371,374,377,380,386,389,393,396,408,415,421,424,430,433,477,480,488,494,498,501,506,509,514,517,522,525,530,545,548,554,558,563,566,569,574,577,582,585,590,598,604,608,611,617,623,629,635,647,650,653,794,806,812,816,819,824,827,833,868,871,875,882,899,903,914,927,933,937,940,943,946,966,969,977,980,988,992,995,1000,1003,1008,1011,1016,1019,1076,1082,1090,1094,1097,1100,1103,1106,1109,1116,1118,1123,1129,1134,1137,1142,1160,1165,1168,1173,1176],[14,364,365],{},[17,366,367],{},"We tested every recommended local model. Some chat fine. None reliably call tools. Here's the full picture.",[14,369,370],{},"I spent a Saturday afternoon trying to get Qwen3 8B running through Ollama as my OpenClaw agent's primary model. Zero API costs. Full privacy. The dream setup.",[14,372,373],{},"The model loaded. The gateway started. I typed \"hello.\" It responded instantly. This is going to work.",[14,375,376],{},"Then I asked it to check my calendar. The agent generated a narrative essay about how it would check my calendar if it could, instead of actually calling the calendar tool. I asked it to search the web. Same thing. Beautiful prose about web searching. Zero actual web searches.",[14,378,379],{},"Three hours later, I'd tried four different models, two different API configurations, and one custom provider workaround from a GitHub issue. The chat worked perfectly every time. The tool calling failed silently every time.",[14,381,382,383],{},"Here's what nobody tells you about the OpenClaw Ollama setup: ",[125,384,385],{},"chat and tool calling are completely different capabilities, and local models in 2026 handle the first one well and the second one poorly.",[14,387,388],{},"This guide covers exactly what works, exactly what doesn't, and the specific scenarios where Ollama with OpenClaw is genuinely worth the effort.",[42,390,392],{"id":391},"the-fundamental-problem-streaming-breaks-tool-calling","The fundamental problem: streaming breaks tool calling",[14,394,395],{},"This is the root cause of most OpenClaw Ollama failures, and it's documented in GitHub Issue #5769.",[14,397,398,399,403,404,407],{},"OpenClaw sends ",[400,401,402],"code",{},"stream: true"," on every model request. This is fine for cloud providers like Anthropic and OpenAI, whose streaming implementations properly emit tool call responses. But Ollama's streaming implementation doesn't correctly return ",[400,405,406],{},"tool_calls"," delta chunks.",[14,409,410,411,414],{},"What happens: your local model decides to call a tool (web_search, exec, browser). It generates the tool call in its response. But the streaming protocol drops it. OpenClaw receives empty content with ",[400,412,413],{},"finish_reason: \"stop\""," instead of the tool call. The tool never executes.",[14,416,417,420],{},[125,418,419],{},"The result: your agent can have conversations but can't perform actions."," No file operations. No web searches. No shell commands. No skill execution. The model writes about what it would do instead of doing it.",[14,422,423],{},"This affects every Ollama model configured through OpenClaw. Mistral, Qwen, Llama, DeepSeek local variants. All of them.",[14,425,426,429],{},[125,427,428],{},"OpenClaw + Ollama = chat works. Tool calling doesn't."," This isn't a config problem. It's an architectural mismatch between OpenClaw's streaming requirement and Ollama's tool call implementation.",[14,431,432],{},"The community has proposed a fix: a per-provider config option to disable streaming when tools are present. The suggested code is straightforward:",[434,435,439],"pre",{"className":436,"code":437,"language":438,"meta":321,"style":321},"language-javascript shiki shiki-themes github-light","const shouldStream = !(context.tools?.length && isOllamaProvider(model));\n","javascript",[400,440,441],{"__ignoreMap":321},[442,443,446,450,454,457,460,464,467,470,474],"span",{"class":444,"line":445},"line",1,[442,447,449],{"class":448},"sD7c4","const",[442,451,453],{"class":452},"sYu0t"," shouldStream",[442,455,456],{"class":448}," =",[442,458,459],{"class":448}," !",[442,461,463],{"class":462},"sgsFI","(context.tools?.",[442,465,466],{"class":452},"length",[442,468,469],{"class":448}," &&",[442,471,473],{"class":472},"s7eDp"," isOllamaProvider",[442,475,476],{"class":462},"(model));\n",[14,478,479],{},"As of March 2026, this hasn't been merged into a release. Until it is, local models through Ollama are limited to chat-only interactions.",[14,481,482,483,487],{},"For a detailed breakdown of all ",[50,484,486],{"href":485},"/blog/openclaw-local-model-not-working","five ways local models fail in OpenClaw"," (including discovery timeouts, WSL2 networking, and the CLI vs API confusion), our troubleshooting guide covers each failure mode.",[14,489,490],{},[66,491],{"alt":492,"src":493},"Diagram showing OpenClaw streaming request flow with Ollama tool call being dropped","/img/blog/openclaw-ollama-streaming-bug.jpg",[42,495,497],{"id":496},"what-actually-works-with-openclaw-ollama","What actually works with OpenClaw + Ollama",[14,499,500],{},"The streaming bug kills tool calling. But not everything in OpenClaw requires tools. Here's what genuinely works.",[14,502,503],{},[125,504,505],{},"Basic conversation",[14,507,508],{},"This works perfectly. Ask questions. Get answers. Have discussions. The agent responds through whatever chat platform you've connected (Telegram, WhatsApp, Slack). If all you want is a private chatbot that runs on your hardware, Ollama delivers.",[14,510,511],{},[125,512,513],{},"Memory and context",[14,515,516],{},"Ollama models maintain conversation context through OpenClaw's memory system. The agent remembers previous messages, stores preferences, and builds context over time. This works the same as cloud models for conversational interactions.",[14,518,519],{},[125,520,521],{},"SOUL.md personality",[14,523,524],{},"Your agent's personality configuration works normally with local models. Customize tone, behavior rules, and working context. The model follows the system prompt instructions.",[14,526,527],{},[125,528,529],{},"Model switching mid-conversation",[14,531,532,533,536,537,540,541,544],{},"The ",[400,534,535],{},"/model"," command works with Ollama models. You can switch between local and cloud providers on the fly. Type ",[400,538,539],{},"/model ollama/qwen3:8b"," for a quick local response, then ",[400,542,543],{},"/model anthropic/claude-sonnet-4-6"," when you need tool execution.",[14,546,547],{},"This hybrid approach is actually the best use of Ollama in OpenClaw: local for chat, cloud for actions.",[14,549,550],{},[66,551],{"alt":552,"src":553},"OpenClaw chat working correctly with Ollama local model on Telegram","/img/blog/openclaw-ollama-chat-working.jpg",[42,555,557],{"id":556},"what-breaks-and-why-you-cant-config-your-way-around-it","What breaks (and why you can't config your way around it)",[14,559,560],{},[125,561,562],{},"Tool calling (the big one)",[14,564,565],{},"Every skill that requires the agent to call a tool fails silently. This includes: web search, file read/write, shell command execution, browser automation, email skills, calendar skills, and essentially every skill that makes an agent more than a chatbot.",[14,567,568],{},"The model generates the intent to call the tool. The streaming protocol loses it. OpenClaw never receives the instruction. No error message appears. The agent just produces text instead of action.",[14,570,571],{},[125,572,573],{},"Cron jobs that require actions",[14,575,576],{},"Scheduled tasks that involve tool use (morning briefings that check your calendar, email triage that reads your inbox) fail for the same reason. The cron fires. The model responds. But no tools execute. You get a narrative about what the agent would do, not an actual result.",[14,578,579],{},[125,580,581],{},"Sub-agent parallel processing",[14,583,584],{},"Sub-agents inherit the tool calling limitation. If your main agent spawns workers for parallel tasks, those workers can't execute tools either. The parallelism works. The execution doesn't.",[14,586,587],{},[125,588,589],{},"Browser relay",[14,591,592,593,597],{},"OpenClaw's ",[50,594,596],{"href":595},"/blog/openclaw-browser-relay","browser automation"," requires precise tool calling to click elements, fill forms, and navigate pages. Local models can't generate the structured tool calls needed. Browser relay with Ollama simply doesn't function.",[14,599,600],{},[66,601],{"alt":602,"src":603},"Terminal showing OpenClaw agent generating text about tool use instead of executing it","/img/blog/openclaw-ollama-tool-failure.jpg",[42,605,607],{"id":606},"the-models-the-community-actually-recommends","The models the community actually recommends",[14,609,610],{},"Despite the tool calling limitation, some local models work noticeably better than others for the chat-only use case.",[14,612,613,616],{},[125,614,615],{},"glm-4.7-flash (~25GB VRAM):"," The community favorite. Multiple users in GitHub Discussion #2936 call it \"huge bang for the buck.\" Strong reasoning and code generation. Runs on an RTX 4090, though not entirely in VRAM.",[14,618,619,622],{},[125,620,621],{},"qwen3-coder-30b:"," Good for code-heavy conversations. Requires significant hardware (24GB+ RAM for quantized versions).",[14,624,625,628],{},[125,626,627],{},"hermes-2-pro and mistral:7b:"," Ollama's official recommendations for tool calling. These are the models most likely to work when the streaming fix eventually lands, since they have native tool calling support in non-streaming mode.",[14,630,631,634],{},[125,632,633],{},"Models under 8B parameters:"," Frequent failures on agent tasks even in chat-only mode. Context tracking degrades quickly. Instructions get ignored or misinterpreted. Not recommended for anything beyond basic Q&A.",[14,636,637,638,641,642],{},"🎥 ",[125,639,640],{},"Watch: OpenClaw with Ollama Local Models Setup and Limitations","\nIf you want to see the Ollama configuration in action (including what the tool calling failure actually looks like and which models perform best for chat-only use), this community walkthrough provides an honest demonstration.\n🎬 ",[50,643,646],{"href":644,"rel":645},"https://www.youtube.com/results?search_query=openclaw+ollama+local+model+setup+2026",[54],"Watch on YouTube",[14,648,649],{},"For local models, plan for 30B+ parameters with at least 64K context window. Anything smaller struggles with OpenClaw's system prompts and multi-turn conversations.",[14,651,652],{},"Ollama's own OpenClaw integration docs recommend 64K minimum context. Many popular models default to much less. Set it explicitly in your config:",[434,654,658],{"className":655,"code":656,"language":657,"meta":321,"style":321},"language-json shiki shiki-themes github-light","{\n  \"models\": {\n    \"providers\": {\n      \"ollama\": {\n        \"baseUrl\": \"http://127.0.0.1:11434\",\n        \"apiKey\": \"ollama-local\",\n        \"api\": \"ollama\",\n        \"models\": [{\n          \"id\": \"qwen3:8b\",\n          \"contextWindow\": 65536\n        }]\n      }\n    }\n  }\n}\n","json",[400,659,660,665,673,681,689,705,718,731,740,753,764,770,776,782,788],{"__ignoreMap":321},[442,661,662],{"class":444,"line":445},[442,663,664],{"class":462},"{\n",[442,666,667,670],{"class":444,"line":322},[442,668,669],{"class":452},"  \"models\"",[442,671,672],{"class":462},": {\n",[442,674,676,679],{"class":444,"line":675},3,[442,677,678],{"class":452},"    \"providers\"",[442,680,672],{"class":462},[442,682,684,687],{"class":444,"line":683},4,[442,685,686],{"class":452},"      \"ollama\"",[442,688,672],{"class":462},[442,690,692,695,698,702],{"class":444,"line":691},5,[442,693,694],{"class":452},"        \"baseUrl\"",[442,696,697],{"class":462},": ",[442,699,701],{"class":700},"sYBdl","\"http://127.0.0.1:11434\"",[442,703,704],{"class":462},",\n",[442,706,708,711,713,716],{"class":444,"line":707},6,[442,709,710],{"class":452},"        \"apiKey\"",[442,712,697],{"class":462},[442,714,715],{"class":700},"\"ollama-local\"",[442,717,704],{"class":462},[442,719,721,724,726,729],{"class":444,"line":720},7,[442,722,723],{"class":452},"        \"api\"",[442,725,697],{"class":462},[442,727,728],{"class":700},"\"ollama\"",[442,730,704],{"class":462},[442,732,734,737],{"class":444,"line":733},8,[442,735,736],{"class":452},"        \"models\"",[442,738,739],{"class":462},": [{\n",[442,741,743,746,748,751],{"class":444,"line":742},9,[442,744,745],{"class":452},"          \"id\"",[442,747,697],{"class":462},[442,749,750],{"class":700},"\"qwen3:8b\"",[442,752,704],{"class":462},[442,754,756,759,761],{"class":444,"line":755},10,[442,757,758],{"class":452},"          \"contextWindow\"",[442,760,697],{"class":462},[442,762,763],{"class":452},"65536\n",[442,765,767],{"class":444,"line":766},11,[442,768,769],{"class":462},"        }]\n",[442,771,773],{"class":444,"line":772},12,[442,774,775],{"class":462},"      }\n",[442,777,779],{"class":444,"line":778},13,[442,780,781],{"class":462},"    }\n",[442,783,785],{"class":444,"line":784},14,[442,786,787],{"class":462},"  }\n",[442,789,791],{"class":444,"line":790},15,[442,792,793],{"class":462},"}\n",[14,795,796,797,800,801,805],{},"For guidance on ",[50,798,799],{"href":321},"choosing the right model for your specific use case",", our ",[50,802,804],{"href":803},"/blog/openclaw-model-comparison","model comparison"," covers cost-per-task data across local and cloud providers.",[14,807,808],{},[66,809],{"alt":810,"src":811},"Comparison chart of Ollama local models showing VRAM requirements and capability ratings","/img/blog/openclaw-ollama-model-comparison.jpg",[42,813,815],{"id":814},"the-three-ollama-gotchas-that-waste-hours","The three Ollama gotchas that waste hours",[14,817,818],{},"Beyond the tool calling bug, three configuration issues eat the most time.",[820,821,823],"h3",{"id":822},"gotcha-1-model-discovery-timeout","Gotcha 1: Model discovery timeout",[14,825,826],{},"When OpenClaw starts, it tries to auto-discover Ollama models. If Ollama is slow (common when the model isn't pre-loaded), discovery times out silently. Your gateway starts. Your model is listed. But requests fail.",[14,828,829,832],{},[125,830,831],{},"Fix:"," Pre-load the model before starting OpenClaw:",[434,834,838],{"className":835,"code":836,"language":837,"meta":321,"style":321},"language-bash shiki shiki-themes github-light","ollama run qwen3:8b\n# Wait for \"success,\" then Ctrl+C\nopenclaw gateway start\n","bash",[400,839,840,851,857],{"__ignoreMap":321},[442,841,842,845,848],{"class":444,"line":445},[442,843,844],{"class":472},"ollama",[442,846,847],{"class":700}," run",[442,849,850],{"class":700}," qwen3:8b\n",[442,852,853],{"class":444,"line":322},[442,854,856],{"class":855},"sAwPA","# Wait for \"success,\" then Ctrl+C\n",[442,858,859,862,865],{"class":444,"line":675},[442,860,861],{"class":472},"openclaw",[442,863,864],{"class":700}," gateway",[442,866,867],{"class":700}," start\n",[14,869,870],{},"Or define models explicitly in your config to skip discovery entirely (shown above).",[820,872,874],{"id":873},"gotcha-2-wsl2-networking","Gotcha 2: WSL2 networking",[14,876,877,878,881],{},"If you're running OpenClaw in WSL2 and Ollama on the Windows host (or vice versa), ",[400,879,880],{},"127.0.0.1"," doesn't resolve across the boundary. Your config says localhost. Your curl works. But OpenClaw can't reach Ollama.",[14,883,884,886,887,890,891,894,895,898],{},[125,885,831],{}," Use the actual WSL2 IP from ",[400,888,889],{},"hostname -I",". Or bind Ollama to ",[400,892,893],{},"0.0.0.0"," with ",[400,896,897],{},"OLLAMA_HOST=0.0.0.0:11434 ollama serve",".",[820,900,902],{"id":901},"gotcha-3-the-cli-vs-api-confusion","Gotcha 3: The CLI vs API confusion",[14,904,905,906,909,910,913],{},"GitHub Issue #11283 documents this bizarre behavior: you configure Ollama as a remote API provider with a ",[400,907,908],{},"baseUrl",". OpenClaw should make HTTP API calls. Instead, it tries to execute ",[400,911,912],{},"ollama run"," as a shell command on your local machine. This happens when OpenClaw's model routing falls back to a cloud model that then tries to \"help\" by calling Ollama via CLI.",[14,915,916,918,919,922,923,926],{},[125,917,831],{}," Make sure your Ollama model is explicitly defined in the ",[400,920,921],{},"models.providers"," section with ",[400,924,925],{},"api: \"ollama\""," and is listed in the models array. Don't rely on auto-discovery for remote Ollama.",[14,928,929],{},[66,930],{"alt":931,"src":932},"Terminal showing three common Ollama configuration errors with fix commands","/img/blog/openclaw-ollama-gotchas.jpg",[42,934,936],{"id":935},"the-honest-cost-comparison-ollama-vs-cheap-cloud-providers","The honest cost comparison: Ollama vs cheap cloud providers",[14,938,939],{},"The appeal of Ollama is zero API costs. But \"zero API costs\" and \"zero cost\" are different things.",[14,941,942],{},"Running Ollama on hardware you own means electricity, hardware depreciation, and your time debugging issues. A Mac Mini M4 running 24/7 consumes roughly $3-5/month in electricity. The machine itself costs $600+ and depreciates.",[14,944,945],{},"Meanwhile, cloud providers in 2026 are absurdly cheap:",[169,947,948,954,960],{},[172,949,950,953],{},[125,951,952],{},"DeepSeek V3.2:"," $0.28/$0.42 per million tokens. A full month of moderate agent usage: $3-8/month.",[172,955,956,959],{},[125,957,958],{},"Gemini 2.5 Flash free tier:"," 1,500 requests/day. $0/month for personal use.",[172,961,962,965],{},[125,963,964],{},"Claude Haiku 4.5:"," $1/$5 per million tokens. Moderate usage: $5-10/month.",[14,967,968],{},"And critically: these cloud providers have working tool calling. Your agent can actually do things.",[14,970,971,972,976],{},"For the full breakdown of ",[50,973,975],{"href":974},"/blog/cheapest-openclaw-ai-providers","which cloud providers cost what for OpenClaw",", our provider comparison covers five alternatives that cost 90% less than most people expect.",[14,978,979],{},"The cheapest model isn't the one with the lowest per-token price. It's the one that can do the job. An Ollama model that can chat but can't call tools isn't a cheaper agent. It's a more expensive chatbot.",[14,981,982,983,987],{},"If you want tool calling that works, multi-channel support, and zero Ollama debugging, ",[50,984,986],{"href":985},"/","BetterClaw supports all 28+ cloud providers"," with BYOK and zero configuration. $19/month per agent. 60-second deploy. Every model routes correctly because the streaming issue doesn't exist with cloud APIs.",[42,989,991],{"id":990},"when-ollama-with-openclaw-genuinely-makes-sense","When Ollama with OpenClaw genuinely makes sense",[14,993,994],{},"I'm not going to pretend Ollama is never the right choice. Three scenarios justify the setup.",[14,996,997],{},[125,998,999],{},"Privacy-first deployments",[14,1001,1002],{},"If your data absolutely cannot leave your network, local models are the only option. Government, healthcare, legal, defense: these environments have compliance requirements that no cloud provider can satisfy. The tool calling limitation is real, but for conversational interaction with sensitive data, Ollama delivers complete data sovereignty.",[14,1004,1005],{},[125,1006,1007],{},"Offline and air-gapped environments",[14,1009,1010],{},"No internet? No API calls. Ollama runs entirely locally. If you need an AI assistant in an environment without reliable connectivity, local models are it.",[14,1012,1013],{},[125,1014,1015],{},"Hybrid heartbeat routing",[14,1017,1018],{},"Use Ollama for heartbeats (the 48 daily status checks that cost tokens on cloud providers) and a cloud model for everything else. Heartbeats don't require tool calling. They're simple status checks. Running them locally saves $4-15/month depending on your cloud model pricing.",[434,1020,1022],{"className":655,"code":1021,"language":657,"meta":321,"style":321},"{\n  \"agent\": {\n    \"model\": {\n      \"primary\": \"anthropic/claude-sonnet-4-6\",\n      \"heartbeat\": \"ollama/hermes-2-pro:latest\"\n    }\n  }\n}\n",[400,1023,1024,1028,1035,1042,1054,1064,1068,1072],{"__ignoreMap":321},[442,1025,1026],{"class":444,"line":445},[442,1027,664],{"class":462},[442,1029,1030,1033],{"class":444,"line":322},[442,1031,1032],{"class":452},"  \"agent\"",[442,1034,672],{"class":462},[442,1036,1037,1040],{"class":444,"line":675},[442,1038,1039],{"class":452},"    \"model\"",[442,1041,672],{"class":462},[442,1043,1044,1047,1049,1052],{"class":444,"line":683},[442,1045,1046],{"class":452},"      \"primary\"",[442,1048,697],{"class":462},[442,1050,1051],{"class":700},"\"anthropic/claude-sonnet-4-6\"",[442,1053,704],{"class":462},[442,1055,1056,1059,1061],{"class":444,"line":691},[442,1057,1058],{"class":452},"      \"heartbeat\"",[442,1060,697],{"class":462},[442,1062,1063],{"class":700},"\"ollama/hermes-2-pro:latest\"\n",[442,1065,1066],{"class":444,"line":707},[442,1067,781],{"class":462},[442,1069,1070],{"class":444,"line":720},[442,1071,787],{"class":462},[442,1073,1074],{"class":444,"line":733},[442,1075,793],{"class":462},[14,1077,1078],{},[66,1079],{"alt":1080,"src":1081},"Hybrid model routing diagram showing Ollama for heartbeats and Claude for tool-based tasks","/img/blog/openclaw-ollama-hybrid-routing.jpg",[14,1083,1084,1085,1089],{},"For the full model routing setup, our ",[50,1086,1088],{"href":1087},"/blog/openclaw-model-routing","intelligent provider switching guide"," covers the config patterns.",[42,1091,1093],{"id":1092},"where-this-is-heading","Where this is heading",[14,1095,1096],{},"The streaming + tool calling bug will get fixed eventually. The proposed patch is clean. The community wants it. It's a matter of when, not if.",[14,1098,1099],{},"When it lands, the best local models (glm-4.7-flash, qwen3-coder-30b) will become genuinely useful for agent tasks. Tool calling will work. Skills will execute. The gap between local and cloud will narrow significantly for the subset of tasks that don't require frontier-level reasoning.",[14,1101,1102],{},"But \"narrowing\" isn't \"closing.\" Cloud models like Claude Sonnet and GPT-4o will still outperform local models on complex multi-step reasoning, long-context accuracy, and prompt injection resistance for the foreseeable future. The hardware requirements for running competitive local models (25GB+ VRAM, 64GB+ RAM for larger models) put them out of reach for most users.",[14,1104,1105],{},"The practical future is hybrid. Cloud for the tasks that need it. Local for the tasks that don't. OpenClaw's model routing architecture already supports this. The tooling just needs to catch up.",[14,1107,1108],{},"For now, if you need an agent that can act (not just talk), cloud providers are the reliable path. If you need complete privacy for conversational AI, Ollama works today.",[14,1110,1111,1112,1115],{},"If you want an agent that works with any provider without debugging streaming protocols, ",[50,1113,1114],{"href":232},"give BetterClaw a try",". $19/month per agent, BYOK with any cloud provider or combination. 60-second deploy. The tool calling just works because we handle the model integration layer. You build workflows instead of workarounds.",[42,1117,279],{"id":278},[14,1119,1120],{},[125,1121,1122],{},"Does OpenClaw work with Ollama local models?",[14,1124,1125,1126,1128],{},"Partially. Chat and conversation work correctly with Ollama models through OpenClaw. Tool calling (web search, file operations, shell commands, browser automation, skills) does not work due to a streaming protocol bug documented in GitHub Issue #5769. OpenClaw sends ",[400,1127,402],{}," on all requests, but Ollama's streaming implementation drops tool call responses. Until this is patched, local models are limited to chat-only interactions.",[14,1130,1131],{},[125,1132,1133],{},"How does Ollama compare to cloud providers for OpenClaw?",[14,1135,1136],{},"Ollama offers zero API costs and complete data privacy but lacks working tool calling in OpenClaw. Cloud providers (Claude Sonnet at $3/$15, DeepSeek at $0.28/$0.42, Gemini Flash free tier) have reliable tool calling, larger context windows, and better multi-step reasoning. For agent tasks that require actions (email, calendar, web search), cloud providers are significantly more capable. For private conversational AI, Ollama works well.",[14,1138,1139],{},[125,1140,1141],{},"How do I set up Ollama with OpenClaw?",[14,1143,1144,1145,1148,1149,1152,1153,1156,1157,1159],{},"Install Ollama and pull your model (",[400,1146,1147],{},"ollama pull qwen3:8b","). Pre-load the model before starting OpenClaw to avoid discovery timeouts. Configure your ",[400,1150,1151],{},"~/.openclaw/openclaw.json"," with the Ollama provider, setting ",[400,1154,1155],{},"contextWindow"," to at least 65536. Start the gateway and test. If on WSL2, use the actual network IP instead of ",[400,1158,880],{},". Expect chat to work and tool calling to fail.",[14,1161,1162],{},[125,1163,1164],{},"Is running OpenClaw with Ollama cheaper than cloud APIs?",[14,1166,1167],{},"Not always. Ollama has zero token costs but requires dedicated hardware ($600+ Mac Mini or GPU-capable machine) and electricity ($3-5/month). DeepSeek V3.2 runs a full agent for $3-8/month via API. Gemini Flash has a free tier. When you factor in hardware cost, electricity, and the time debugging Ollama issues, cheap cloud providers often cost less overall. The exception: if you already have capable hardware and need complete data privacy.",[14,1169,1170],{},[125,1171,1172],{},"Which Ollama models work best with OpenClaw?",[14,1174,1175],{},"For chat-only use: glm-4.7-flash (best quality, needs ~25GB VRAM), qwen3-coder-30b (strong for code, needs 24GB+ RAM), and hermes-2-pro or mistral:7b (Ollama's recommended tool calling models, will be first to work when the streaming fix lands). Avoid models under 8B parameters for agent tasks. Set context window to 64K+ minimum in your config.",[1177,1178,1179],"style",{},"html pre.shiki code .sgsFI, html code.shiki .sgsFI{--shiki-default:#24292E}html pre.shiki code .sYu0t, html code.shiki .sYu0t{--shiki-default:#005CC5}html pre.shiki code .sYBdl, html code.shiki .sYBdl{--shiki-default:#032F62}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html pre.shiki code .s7eDp, html code.shiki .s7eDp{--shiki-default:#6F42C1}html pre.shiki code .sAwPA, html code.shiki .sAwPA{--shiki-default:#6A737D}html pre.shiki code .sD7c4, html code.shiki .sD7c4{--shiki-default:#D73A49}",{"title":321,"searchDepth":322,"depth":322,"links":1181},[1182,1183,1184,1185,1186,1191,1192,1193,1194],{"id":391,"depth":322,"text":392},{"id":496,"depth":322,"text":497},{"id":556,"depth":322,"text":557},{"id":606,"depth":322,"text":607},{"id":814,"depth":322,"text":815,"children":1187},[1188,1189,1190],{"id":822,"depth":675,"text":823},{"id":873,"depth":675,"text":874},{"id":901,"depth":675,"text":902},{"id":935,"depth":322,"text":936},{"id":990,"depth":322,"text":991},{"id":1092,"depth":322,"text":1093},{"id":278,"depth":322,"text":279},"2026-03-18","OpenClaw Ollama chat works fine. Tool calling breaks silently. Here's what the streaming bug means, which models perform best, and when cloud is smarter.","/img/blog/openclaw-ollama-guide.jpg",{},"/blog/openclaw-ollama-guide","14 min read",{"title":359,"description":1196},"OpenClaw + Ollama: Local Model Setup & Tool Calling Fix (2026)","blog/openclaw-ollama-guide",[1205,1206,1207,1208,1209,1210,1211],"OpenClaw Ollama","OpenClaw local model","Ollama tool calling OpenClaw","OpenClaw Ollama setup","best Ollama model OpenClaw","OpenClaw offline","OpenClaw local vs cloud","IwnUivkq8vTAToTxv9U7fyh4YIKxJ_sfYyKFE5fKlGU",{"id":1214,"title":1215,"author":1216,"body":1217,"category":334,"date":2207,"description":2208,"extension":337,"featured":338,"image":2209,"meta":2210,"navigation":341,"path":2211,"readingTime":2212,"seo":2213,"seoTitle":2214,"stem":2215,"tags":2216,"updatedDate":2207,"__hash__":2224},"blog/blog/openclaw-setup-guide-complete.md","OpenClaw Setup Guide: Hardware, Installation, and Configuration in the Right Order",{"name":7,"role":8,"avatar":9},{"type":11,"value":1218,"toc":2185},[1219,1224,1227,1230,1233,1236,1240,1243,1249,1255,1261,1264,1272,1275,1281,1285,1288,1291,1304,1307,1412,1415,1433,1436,1451,1457,1460,1466,1470,1473,1479,1485,1491,1497,1508,1511,1575,1578,1584,1588,1591,1594,1599,1630,1636,1642,1645,1659,1663,1666,1669,1674,1679,1696,1706,1711,1737,1742,1808,1814,1825,1830,1848,1856,1869,1875,1879,1882,1886,1889,1941,1945,1948,1954,1957,1961,1964,1967,1979,1983,1986,2021,2024,2031,2037,2041,2044,2054,2060,2070,2074,2077,2109,2114,2117,2120,2123,2130,2132,2136,2139,2143,2150,2154,2161,2165,2172,2176,2182],[14,1220,1221],{},[17,1222,1223],{},"Every other guide skips steps or puts them in the wrong sequence. This one doesn't.",[14,1225,1226],{},"I followed three different OpenClaw setup guides before I got a working agent. The first one skipped the security steps entirely. The second one had me configuring channels before I'd even picked a model. The third one was written for a version that no longer existed.",[14,1228,1229],{},"The order matters. Do things out of sequence and you'll spend hours debugging problems that wouldn't exist if you'd just done Step 3 before Step 5.",[14,1231,1232],{},"This OpenClaw setup guide puts everything in the right order. Hardware first. Then Node. Then model provider. Then your first channel. Then security. Then skills and automation. Each step builds on the previous one. No backtracking.",[14,1234,1235],{},"Whether you're setting up on a Mac Mini, a VPS, or a managed platform, the sequence is the same. The commands change. The logic doesn't.",[42,1237,1239],{"id":1238},"step-1-pick-your-hardware-and-understand-what-youre-actually-choosing","Step 1: Pick your hardware (and understand what you're actually choosing)",[14,1241,1242],{},"You have three paths. Each has a different cost, complexity, and maintenance profile.",[14,1244,1245,1248],{},[125,1246,1247],{},"Path A: Local machine"," (Mac Mini or laptop). $600+ upfront for a Mac Mini M4 with 16GB RAM. Runs on your desk. Always-on requires an app like Amphetamine to prevent sleep. Your personal files and accounts share the same machine as the agent. This is the path most YouTube tutorials show. It's also the one Microsoft's security blog explicitly recommends against.",[14,1250,1251,1254],{},[125,1252,1253],{},"Path B: Cloud VPS."," $5-29/month. Hetzner, Contabo, DigitalOcean, Hostinger, OVHcloud. Isolated from your personal data. Always-on by default. You manage the server, security, Docker, and updates yourself. Minimum specs: 2 vCPU, 2GB RAM (4GB recommended for browser automation).",[14,1256,1257,1260],{},[125,1258,1259],{},"Path C: Managed platform."," $19/month (Better Claw) to $49/month (ClawHosted). Zero infrastructure management. Deploy in under 60 seconds. Security, updates, and monitoring handled for you.",[14,1262,1263],{},"If you choose Path A or B, keep reading. If you choose Path C, skip to the \"What to do after your agent is live\" section at the end.",[14,1265,1266,1267,1271],{},"For a detailed comparison of ",[50,1268,1270],{"href":1269},"/compare/openclaw","self-hosted vs managed OpenClaw deployment",", our comparison page covers the full tradeoff matrix.",[14,1273,1274],{},"Choose VPS over local machine. The isolation alone is worth $5/month. Running an autonomous agent on the same computer where you do your banking is a risk most security researchers consider unacceptable.",[14,1276,1277],{},[66,1278],{"alt":1279,"src":1280},"Three OpenClaw deployment paths comparing local machine, cloud VPS, and managed platform with cost and complexity tradeoffs","/img/blog/openclaw-setup-guide-complete-hardware.jpg",[42,1282,1284],{"id":1283},"step-2-install-nodejs-22-the-requirement-nobody-mentions-first","Step 2: Install Node.js 22+ (the requirement nobody mentions first)",[14,1286,1287],{},"Before you touch OpenClaw, you need Node.js 22 or higher. Not 18. Not 20. Twenty-two.",[14,1289,1290],{},"Most systems come with an older version. Check yours:",[434,1292,1294],{"className":835,"code":1293,"language":837,"meta":321,"style":321},"node --version\n",[400,1295,1296],{"__ignoreMap":321},[442,1297,1298,1301],{"class":444,"line":445},[442,1299,1300],{"class":472},"node",[442,1302,1303],{"class":452}," --version\n",[14,1305,1306],{},"If it's below 22, upgrade:",[434,1308,1310],{"className":835,"code":1309,"language":837,"meta":321,"style":321},"# Using nvm (recommended)\ncurl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.0/install.sh | bash\nsource ~/.bashrc\nnvm install 22\nnvm use 22\n\n# Or using nodesource\ncurl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -\nsudo apt-get install -y nodejs\n",[400,1311,1312,1317,1334,1342,1353,1362,1367,1372,1396],{"__ignoreMap":321},[442,1313,1314],{"class":444,"line":445},[442,1315,1316],{"class":855},"# Using nvm (recommended)\n",[442,1318,1319,1322,1325,1328,1331],{"class":444,"line":322},[442,1320,1321],{"class":472},"curl",[442,1323,1324],{"class":452}," -o-",[442,1326,1327],{"class":700}," https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.0/install.sh",[442,1329,1330],{"class":448}," |",[442,1332,1333],{"class":472}," bash\n",[442,1335,1336,1339],{"class":444,"line":675},[442,1337,1338],{"class":452},"source",[442,1340,1341],{"class":700}," ~/.bashrc\n",[442,1343,1344,1347,1350],{"class":444,"line":683},[442,1345,1346],{"class":472},"nvm",[442,1348,1349],{"class":700}," install",[442,1351,1352],{"class":452}," 22\n",[442,1354,1355,1357,1360],{"class":444,"line":691},[442,1356,1346],{"class":472},[442,1358,1359],{"class":700}," use",[442,1361,1352],{"class":452},[442,1363,1364],{"class":444,"line":707},[442,1365,1366],{"emptyLinePlaceholder":341},"\n",[442,1368,1369],{"class":444,"line":720},[442,1370,1371],{"class":855},"# Or using nodesource\n",[442,1373,1374,1376,1379,1382,1384,1387,1390,1393],{"class":444,"line":733},[442,1375,1321],{"class":472},[442,1377,1378],{"class":452}," -fsSL",[442,1380,1381],{"class":700}," https://deb.nodesource.com/setup_22.x",[442,1383,1330],{"class":448},[442,1385,1386],{"class":472}," sudo",[442,1388,1389],{"class":452}," -E",[442,1391,1392],{"class":700}," bash",[442,1394,1395],{"class":700}," -\n",[442,1397,1398,1401,1404,1406,1409],{"class":444,"line":742},[442,1399,1400],{"class":472},"sudo",[442,1402,1403],{"class":700}," apt-get",[442,1405,1349],{"class":700},[442,1407,1408],{"class":452}," -y",[442,1410,1411],{"class":700}," nodejs\n",[14,1413,1414],{},"Now install OpenClaw:",[434,1416,1418],{"className":835,"code":1417,"language":837,"meta":321,"style":321},"npm install -g @openclaw/cli\n",[400,1419,1420],{"__ignoreMap":321},[442,1421,1422,1425,1427,1430],{"class":444,"line":445},[442,1423,1424],{"class":472},"npm",[442,1426,1349],{"class":700},[442,1428,1429],{"class":452}," -g",[442,1431,1432],{"class":700}," @openclaw/cli\n",[14,1434,1435],{},"Run the onboarding wizard:",[434,1437,1439],{"className":835,"code":1438,"language":837,"meta":321,"style":321},"openclaw onboard --install-daemon\n",[400,1440,1441],{"__ignoreMap":321},[442,1442,1443,1445,1448],{"class":444,"line":445},[442,1444,861],{"class":472},[442,1446,1447],{"class":700}," onboard",[442,1449,1450],{"class":452}," --install-daemon\n",[14,1452,532,1453,1456],{},[400,1454,1455],{},"--install-daemon"," flag sets up OpenClaw to run automatically in the background, even after restarts. Without it, your agent dies every time you close your terminal.",[14,1458,1459],{},"The wizard walks you through model provider selection and your first channel. But here's the thing: the wizard's defaults aren't always the right defaults. The next two steps explain what to choose and why.",[14,1461,1462],{},[66,1463],{"alt":1464,"src":1465},"Terminal showing Node.js version check and OpenClaw CLI installation with onboarding wizard output","/img/blog/openclaw-setup-guide-complete-install.jpg",[42,1467,1469],{"id":1468},"step-3-choose-your-model-provider-this-decision-controls-your-monthly-bill","Step 3: Choose your model provider (this decision controls your monthly bill)",[14,1471,1472],{},"The onboarding wizard asks for your AI provider. This is the single most important cost decision you'll make.",[14,1474,1475,1478],{},[125,1476,1477],{},"If you want the best agent performance:"," Anthropic (Claude). Sonnet 4.6 at $3/$15 per million tokens is the community consensus best balance of quality and price for agent tasks. Set it as primary. Use Haiku ($1/$5) for heartbeats and sub-agents.",[14,1480,1481,1484],{},[125,1482,1483],{},"If you want the cheapest possible setup:"," DeepSeek V3.2 at $0.28/$0.42 per million tokens. 10x cheaper than Claude. Genuinely capable for standard tasks. Tool calling is less precise on complex chains.",[14,1486,1487,1490],{},[125,1488,1489],{},"If you want free:"," Google Gemini 2.5 Flash through Google AI Studio. Free tier: 1,500 requests/day. No credit card needed.",[14,1492,1493,1496],{},[125,1494,1495],{},"If you want one key for everything:"," OpenRouter. Access 200+ models. Auto-routing picks the cheapest capable model per request. Small markup (under 5%).",[14,1498,1499,1500,800,1503,1507],{},"For the full pricing breakdown and ",[50,1501,1502],{"href":803},"which models cost what for specific agent tasks",[50,1504,1506],{"href":1505},"/blog/openclaw-api-costs","comparison covers real cost-per-task data"," across four providers.",[14,1509,1510],{},"Set up model routing from the start. Don't run everything on your primary model:",[434,1512,1514],{"className":655,"code":1513,"language":657,"meta":321,"style":321},"{\n  \"agent\": {\n    \"model\": {\n      \"primary\": \"anthropic/claude-sonnet-4-6\",\n      \"heartbeat\": \"anthropic/claude-haiku-4-5\",\n      \"subagent\": \"anthropic/claude-haiku-4-5\"\n    }\n  }\n}\n",[400,1515,1516,1520,1526,1532,1542,1553,1563,1567,1571],{"__ignoreMap":321},[442,1517,1518],{"class":444,"line":445},[442,1519,664],{"class":462},[442,1521,1522,1524],{"class":444,"line":322},[442,1523,1032],{"class":452},[442,1525,672],{"class":462},[442,1527,1528,1530],{"class":444,"line":675},[442,1529,1039],{"class":452},[442,1531,672],{"class":462},[442,1533,1534,1536,1538,1540],{"class":444,"line":683},[442,1535,1046],{"class":452},[442,1537,697],{"class":462},[442,1539,1051],{"class":700},[442,1541,704],{"class":462},[442,1543,1544,1546,1548,1551],{"class":444,"line":691},[442,1545,1058],{"class":452},[442,1547,697],{"class":462},[442,1549,1550],{"class":700},"\"anthropic/claude-haiku-4-5\"",[442,1552,704],{"class":462},[442,1554,1555,1558,1560],{"class":444,"line":707},[442,1556,1557],{"class":452},"      \"subagent\"",[442,1559,697],{"class":462},[442,1561,1562],{"class":700},"\"anthropic/claude-haiku-4-5\"\n",[442,1564,1565],{"class":444,"line":720},[442,1566,781],{"class":462},[442,1568,1569],{"class":444,"line":733},[442,1570,787],{"class":462},[442,1572,1573],{"class":444,"line":742},[442,1574,793],{"class":462},[14,1576,1577],{},"This single config change saves 50-80% on API costs compared to running everything on Sonnet.",[14,1579,1580],{},[66,1581],{"alt":1582,"src":1583},"Model provider comparison chart showing Anthropic, DeepSeek, Gemini, and OpenRouter pricing tiers","/img/blog/openclaw-setup-guide-complete-providers.jpg",[42,1585,1587],{"id":1586},"step-4-connect-your-first-channel-start-with-telegram-seriously","Step 4: Connect your first channel (start with Telegram, seriously)",[14,1589,1590],{},"The wizard offers multiple channels. Pick Telegram first. Always Telegram first.",[14,1592,1593],{},"Why? It's the fastest to set up (under 10 minutes), has the simplest authentication flow, and debugging is straightforward. Once your agent responds on Telegram, you know the core pipeline works. Then add other channels one at a time.",[14,1595,1596],{},[125,1597,1598],{},"Telegram setup:",[1600,1601,1602,1617,1624,1627],"ol",{},[172,1603,1604,1605,1608,1609,1612,1613,1616],{},"Open Telegram. Search for ",[125,1606,1607],{},"@BotFather",". Send ",[400,1610,1611],{},"/newbot",". Give it a name and username (must end in ",[400,1614,1615],{},"_bot","). Copy the bot token.",[172,1618,1619,1620,1623],{},"Search for ",[125,1621,1622],{},"@userinfobot",". Click \"Start.\" Copy your numeric user ID.",[172,1625,1626],{},"The wizard asks for both. Paste them in.",[172,1628,1629],{},"Send your bot a message. If it responds, you're golden.",[14,1631,1632],{},[66,1633],{"alt":1634,"src":1635},"Telegram BotFather conversation showing bot creation flow and first successful agent response","/img/blog/openclaw-setup-guide-complete-telegram.jpg",[14,1637,1638,1641],{},[125,1639,1640],{},"After Telegram works, add other channels."," WhatsApp requires Meta's Business API (budget 30-60 minutes). Slack needs OAuth configuration with specific scopes. Discord wants a bot token from the Developer Portal.",[14,1643,1644],{},"Each channel is an independent authentication flow. If one fails, it doesn't affect the others.",[1646,1647,1649],"callout",{"type":1648},"video",[14,1650,1651,1654,1655],{},[125,1652,1653],{},"Watch: Complete OpenClaw Installation and First Channel Setup","\nIf you want to see this entire installation flow in action (from Node installation through the onboarding wizard to your first Telegram response), this community walkthrough covers each step with real terminal output so you can follow along.\n",[50,1656,646],{"href":1657,"rel":1658},"https://www.youtube.com/results?search_query=openclaw+setup+guide+installation+telegram+2026",[54],[42,1660,1662],{"id":1661},"step-5-security-hardening-the-step-most-guides-save-for-later-and-users-never-do","Step 5: Security hardening (the step most guides save for \"later\" and users never do)",[14,1664,1665],{},"Here's what nobody tells you about the OpenClaw setup process: the default configuration is not secure. The installer gets you running. It doesn't get you safe.",[14,1667,1668],{},"This step takes 15-20 minutes. Skipping it puts your API keys, your connected accounts, and your server at risk. Researchers found 30,000+ internet-exposed OpenClaw instances without authentication. Don't be one of them.",[14,1670,1671],{},[125,1672,1673],{},"The minimum security checklist:",[14,1675,1676],{},[125,1677,1678],{},"Bind gateway to localhost:",[434,1680,1682],{"className":835,"code":1681,"language":837,"meta":321,"style":321},"openclaw configure\n# Select \"Local (this machine)\"\n",[400,1683,1684,1691],{"__ignoreMap":321},[442,1685,1686,1688],{"class":444,"line":445},[442,1687,861],{"class":472},[442,1689,1690],{"class":700}," configure\n",[442,1692,1693],{"class":444,"line":322},[442,1694,1695],{"class":855},"# Select \"Local (this machine)\"\n",[14,1697,1698,1699,1702,1703,898],{},"Verify: ",[400,1700,1701],{},"ss -tlnp | grep 18789"," should show ",[400,1704,1705],{},"127.0.0.1:18789",[14,1707,1708],{},[125,1709,1710],{},"Set file permissions:",[434,1712,1714],{"className":835,"code":1713,"language":837,"meta":321,"style":321},"chmod 700 ~/.openclaw\nchmod 600 ~/.openclaw/openclaw.json\n",[400,1715,1716,1727],{"__ignoreMap":321},[442,1717,1718,1721,1724],{"class":444,"line":445},[442,1719,1720],{"class":472},"chmod",[442,1722,1723],{"class":452}," 700",[442,1725,1726],{"class":700}," ~/.openclaw\n",[442,1728,1729,1731,1734],{"class":444,"line":322},[442,1730,1720],{"class":472},[442,1732,1733],{"class":452}," 600",[442,1735,1736],{"class":700}," ~/.openclaw/openclaw.json\n",[14,1738,1739],{},[125,1740,1741],{},"If on a VPS, configure the firewall:",[434,1743,1745],{"className":835,"code":1744,"language":837,"meta":321,"style":321},"sudo ufw default deny incoming\nsudo ufw default allow outgoing\nsudo ufw allow 22/tcp\nsudo ufw limit 22/tcp\nsudo ufw enable\n",[400,1746,1747,1763,1777,1788,1799],{"__ignoreMap":321},[442,1748,1749,1751,1754,1757,1760],{"class":444,"line":445},[442,1750,1400],{"class":472},[442,1752,1753],{"class":700}," ufw",[442,1755,1756],{"class":700}," default",[442,1758,1759],{"class":700}," deny",[442,1761,1762],{"class":700}," incoming\n",[442,1764,1765,1767,1769,1771,1774],{"class":444,"line":322},[442,1766,1400],{"class":472},[442,1768,1753],{"class":700},[442,1770,1756],{"class":700},[442,1772,1773],{"class":700}," allow",[442,1775,1776],{"class":700}," outgoing\n",[442,1778,1779,1781,1783,1785],{"class":444,"line":675},[442,1780,1400],{"class":472},[442,1782,1753],{"class":700},[442,1784,1773],{"class":700},[442,1786,1787],{"class":700}," 22/tcp\n",[442,1789,1790,1792,1794,1797],{"class":444,"line":683},[442,1791,1400],{"class":472},[442,1793,1753],{"class":700},[442,1795,1796],{"class":700}," limit",[442,1798,1787],{"class":700},[442,1800,1801,1803,1805],{"class":444,"line":691},[442,1802,1400],{"class":472},[442,1804,1753],{"class":700},[442,1806,1807],{"class":700}," enable\n",[14,1809,1810,1813],{},[125,1811,1812],{},"Disable SSH password authentication"," (VPS only):",[14,1815,1816,1817,1820,1821,1824],{},"In ",[400,1818,1819],{},"/etc/ssh/sshd_config",", set ",[400,1822,1823],{},"PasswordAuthentication no",". Restart sshd.",[14,1826,1827],{},[125,1828,1829],{},"Run the built-in security audit:",[434,1831,1833],{"className":835,"code":1832,"language":837,"meta":321,"style":321},"openclaw security audit --deep\n",[400,1834,1835],{"__ignoreMap":321},[442,1836,1837,1839,1842,1845],{"class":444,"line":445},[442,1838,861],{"class":472},[442,1840,1841],{"class":700}," security",[442,1843,1844],{"class":700}," audit",[442,1846,1847],{"class":452}," --deep\n",[14,1849,1850,1851,1855],{},"For the complete 10-step hardening process, our ",[50,1852,1854],{"href":1853},"/blog/openclaw-security-risks","OpenClaw security risks guide"," covers every documented vulnerability and the specific config to address each one.",[1646,1857,1859],{"type":1858},"cta",[14,1860,1861,1862,1865,1866],{},"If you'd rather not manage any of this yourself, ",[50,1863,1864],{"href":985},"Better Claw handles security natively"," with Docker sandboxing, AES-256 encryption, and anomaly detection built in. ",[125,1867,1868],{},"$19/month per agent. BYOK. Zero security config needed.",[14,1870,1871],{},[66,1872],{"alt":1873,"src":1874},"Terminal showing OpenClaw security audit output with gateway binding and firewall configuration","/img/blog/openclaw-setup-guide-complete-security.jpg",[42,1876,1878],{"id":1877},"step-6-skills-cron-jobs-and-making-it-actually-useful","Step 6: Skills, cron jobs, and making it actually useful",[14,1880,1881],{},"Your agent is running. It responds on Telegram. It's secured. Now make it do something worth the setup time.",[820,1883,1885],{"id":1884},"configure-your-soulmd","Configure your SOUL.md",[14,1887,1888],{},"This file in your workspace defines your agent's personality and context. Give it your name, your preferences, your work context. The more specific, the better.",[434,1890,1894],{"className":1891,"code":1892,"language":1893,"meta":321,"style":321},"language-markdown shiki shiki-themes github-light","# About the User\nName: [Your name]\nRole: [Your role]\nCommunication style: Concise, direct, no fluff.\nTimezone: [Your timezone]\n\n# Agent Behavior\nDefault to brief responses unless I ask for detail.\nAlways confirm before sending emails or modifying files.\n","markdown",[400,1895,1896,1902,1907,1912,1917,1922,1926,1931,1936],{"__ignoreMap":321},[442,1897,1898],{"class":444,"line":445},[442,1899,1901],{"class":1900},"surfw","# About the User\n",[442,1903,1904],{"class":444,"line":322},[442,1905,1906],{"class":462},"Name: [Your name]\n",[442,1908,1909],{"class":444,"line":675},[442,1910,1911],{"class":462},"Role: [Your role]\n",[442,1913,1914],{"class":444,"line":683},[442,1915,1916],{"class":462},"Communication style: Concise, direct, no fluff.\n",[442,1918,1919],{"class":444,"line":691},[442,1920,1921],{"class":462},"Timezone: [Your timezone]\n",[442,1923,1924],{"class":444,"line":707},[442,1925,1366],{"emptyLinePlaceholder":341},[442,1927,1928],{"class":444,"line":720},[442,1929,1930],{"class":1900},"# Agent Behavior\n",[442,1932,1933],{"class":444,"line":733},[442,1934,1935],{"class":462},"Default to brief responses unless I ask for detail.\n",[442,1937,1938],{"class":444,"line":742},[442,1939,1940],{"class":462},"Always confirm before sending emails or modifying files.\n",[820,1942,1944],{"id":1943},"set-up-your-first-cron-job","Set up your first cron job",[14,1946,1947],{},"A morning briefing is the best first automation. Set it to run at 6:00 AM:",[1949,1950,1951],"blockquote",{},[14,1952,1953],{},"\"Check my calendar for today, summarize any priority emails from overnight, and check the weather. Send the summary to Telegram.\"",[14,1955,1956],{},"This runs daily without prompting. You wake up to useful information.",[820,1958,1960],{"id":1959},"install-skills-carefully","Install skills carefully",[14,1962,1963],{},"The ClawHub marketplace has 13,700+ skills. It also had 824+ malicious ones (roughly 20% of the registry at one point). Cisco found a skill performing data exfiltration without user awareness.",[14,1965,1966],{},"Before installing any skill: read the source code, check the publisher's reputation, search for the skill name in GitHub issues. Start with skills maintained by the OpenClaw core team.",[14,1968,1969,1970,1974,1975,1978],{},"For a curated list of ",[50,1971,1973],{"href":1972},"/blog/best-openclaw-skills","community-vetted OpenClaw skills"," that are safe and genuinely useful, our ",[50,1976,1977],{"href":1972},"skills guide"," ranks the best options.",[820,1980,1982],{"id":1981},"set-cost-and-safety-limits","Set cost and safety limits",[14,1984,1985],{},"On every skill and cron job:",[434,1987,1989],{"className":655,"code":1988,"language":657,"meta":321,"style":321},"{\n  \"maxContextTokens\": 4000,\n  \"maxIterations\": 15\n}\n",[400,1990,1991,1995,2007,2017],{"__ignoreMap":321},[442,1992,1993],{"class":444,"line":445},[442,1994,664],{"class":462},[442,1996,1997,2000,2002,2005],{"class":444,"line":322},[442,1998,1999],{"class":452},"  \"maxContextTokens\"",[442,2001,697],{"class":462},[442,2003,2004],{"class":452},"4000",[442,2006,704],{"class":462},[442,2008,2009,2012,2014],{"class":444,"line":675},[442,2010,2011],{"class":452},"  \"maxIterations\"",[442,2013,697],{"class":462},[442,2015,2016],{"class":452},"15\n",[442,2018,2019],{"class":444,"line":683},[442,2020,793],{"class":462},[14,2022,2023],{},"Set daily spending caps on your API provider. A runaway agent loop can burn through $37 in six hours (documented community incident) or $3,600 in a month (another documented case).",[14,2025,2026,2027,2030],{},"For the complete picture of ",[50,2028,2029],{"href":1505},"how API costs accumulate and how to cap them",", our cost guide covers five specific optimizations.",[14,2032,2033],{},[66,2034],{"alt":2035,"src":2036},"OpenClaw workspace showing SOUL.md configuration, cron job setup, and skills marketplace","/img/blog/openclaw-setup-guide-complete-skills.jpg",[42,2038,2040],{"id":2039},"what-to-do-after-your-agent-is-live","What to do after your agent is live",[14,2042,2043],{},"Once you've completed all six steps, your agent is running, secured, and doing useful work. Here's the maintenance rhythm:",[14,2045,2046,2049,2050,2053],{},[125,2047,2048],{},"Weekly:"," Run ",[400,2051,2052],{},"npm update -g @openclaw/cli"," to stay current on patches. The project had three CVEs in a single week in early 2026. Check your API provider dashboard for unexpected cost spikes.",[14,2055,2056,2059],{},[125,2057,2058],{},"Monthly:"," Review your model routing. New models launch frequently. What was the cheapest option last month may not be this month. DeepSeek, Gemini Flash, and Haiku pricing all shifted in 2026.",[14,2061,2062,2065,2066,2069],{},[125,2063,2064],{},"Ongoing:"," Monitor your gateway logs at ",[400,2067,2068],{},"/tmp/openclaw/openclaw-[date].log",". Set up a simple health check (a cron job that pings a monitoring service if the gateway is running).",[42,2071,2073],{"id":2072},"the-honest-time-estimate","The honest time estimate",[14,2075,2076],{},"For a developer comfortable with command line, VPS, and Docker:",[169,2078,2079,2085,2091,2097,2103],{},[172,2080,2081,2084],{},[125,2082,2083],{},"Steps 1-2"," (hardware + install): 30-60 minutes",[172,2086,2087,2090],{},[125,2088,2089],{},"Step 3"," (model provider): 15 minutes",[172,2092,2093,2096],{},[125,2094,2095],{},"Step 4"," (first channel): 10-15 minutes (Telegram) to 60 minutes (WhatsApp)",[172,2098,2099,2102],{},[125,2100,2101],{},"Step 5"," (security): 15-20 minutes",[172,2104,2105,2108],{},[125,2106,2107],{},"Step 6"," (skills + cron): 30-60 minutes",[14,2110,2111],{},[125,2112,2113],{},"Total: 2-4 hours for a production-ready, secured agent.",[14,2115,2116],{},"For someone learning as they go, double that. Budget a full weekend.",[14,2118,2119],{},"For the managed path: under 2 minutes from signup to a live agent. No steps 1, 2, or 5. Model provider and channel configuration still take the same time because those are account-level decisions regardless of hosting.",[14,2121,2122],{},"The setup isn't hard. It's just longer than the README suggests. And the order matters more than any individual step.",[14,2124,2125,2126,2129],{},"If you've gone through this guide and decided the infrastructure isn't how you want to spend your time, ",[50,2127,2128],{"href":232},"give Better Claw a try",". $19/month per agent, BYOK, 60-second deploy. We handle steps 1, 2, and 5 entirely. You handle the parts that are actually interesting: choosing your model, connecting your channels, and building workflows.",[42,2131,279],{"id":278},[820,2133,2135],{"id":2134},"what-hardware-do-i-need-for-an-openclaw-setup","What hardware do I need for an OpenClaw setup?",[14,2137,2138],{},"Minimum: 2 vCPU, 2GB RAM, 10GB storage. Recommended: 4GB RAM for browser automation. You can run on a local Mac Mini ($600+ upfront), a cloud VPS ($5-29/month), or a managed platform ($19/month). Security researchers recommend VPS or managed over local machines because of the isolation between your personal data and the autonomous agent.",[820,2140,2142],{"id":2141},"how-does-self-hosted-openclaw-compare-to-managed-platforms-like-better-claw","How does self-hosted OpenClaw compare to managed platforms like Better Claw?",[14,2144,2145,2146,2149],{},"Self-hosted gives you full control but requires 2-4 hours of initial setup, ongoing security patching, server maintenance, and Docker management. ",[50,2147,2148],{"href":985},"Better Claw"," deploys in under 60 seconds with built-in Docker sandboxing, AES-256 encryption, and anomaly detection. Both use BYOK for API costs. The tradeoff is control vs convenience, and $5-10/month in VPS costs vs $19/month for zero maintenance.",[820,2151,2153],{"id":2152},"how-long-does-openclaw-installation-take-from-scratch","How long does OpenClaw installation take from scratch?",[14,2155,2156,2157,898],{},"For an experienced developer: 2-4 hours for a fully configured, secured, multi-channel agent. For beginners: 4-8 hours spread across setup, troubleshooting, and security hardening. The biggest time sinks are WhatsApp Business API configuration (30-60 minutes), security hardening (15-20 minutes), and debugging Node version or ",[50,2158,2160],{"href":2159},"/blog/openclaw-not-working","Ollama discovery issues",[820,2162,2164],{"id":2163},"how-much-does-it-cost-to-run-openclaw-after-setup","How much does it cost to run OpenClaw after setup?",[14,2166,2167,2168,2171],{},"API costs with smart model routing (Sonnet primary, Haiku heartbeats): $15-50/month. Without routing (everything on Opus/GPT-4o): $80-200/month. Hosting: $5-19/month (VPS vs managed). The first cost optimization to implement is ",[50,2169,2170],{"href":1505},"model routing",": assign cheap models to heartbeats and sub-agents. This single change saves 50-80%.",[820,2173,2175],{"id":2174},"is-openclaw-safe-to-install-on-my-personal-computer","Is OpenClaw safe to install on my personal computer?",[14,2177,2178,2179,898],{},"OpenClaw's own maintainer warned that it's \"far too dangerous\" for users unfamiliar with command-line security. Microsoft recommends running it only in fully isolated environments. CrowdStrike published enterprise risk advisories. If you install on a personal machine, the agent has access to your files, accounts, and system. A VPS ($5/month) or managed platform ($19/month) provides isolation that protects your personal data from agent errors or ",[50,2180,2181],{"href":1853},"security compromises",[1177,2183,2184],{},"html pre.shiki code .s7eDp, html code.shiki .s7eDp{--shiki-default:#6F42C1}html pre.shiki code .sYu0t, html code.shiki .sYu0t{--shiki-default:#005CC5}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html pre.shiki code .sAwPA, html code.shiki .sAwPA{--shiki-default:#6A737D}html pre.shiki code .sYBdl, html code.shiki .sYBdl{--shiki-default:#032F62}html pre.shiki code .sD7c4, html code.shiki .sD7c4{--shiki-default:#D73A49}html pre.shiki code .sgsFI, html code.shiki .sgsFI{--shiki-default:#24292E}html pre.shiki code .surfw, html code.shiki .surfw{--shiki-default:#005CC5;--shiki-default-font-weight:bold}",{"title":321,"searchDepth":322,"depth":322,"links":2186},[2187,2188,2189,2190,2191,2192,2198,2199,2200],{"id":1238,"depth":322,"text":1239},{"id":1283,"depth":322,"text":1284},{"id":1468,"depth":322,"text":1469},{"id":1586,"depth":322,"text":1587},{"id":1661,"depth":322,"text":1662},{"id":1877,"depth":322,"text":1878,"children":2193},[2194,2195,2196,2197],{"id":1884,"depth":675,"text":1885},{"id":1943,"depth":675,"text":1944},{"id":1959,"depth":675,"text":1960},{"id":1981,"depth":675,"text":1982},{"id":2039,"depth":322,"text":2040},{"id":2072,"depth":322,"text":2073},{"id":278,"depth":322,"text":279,"children":2201},[2202,2203,2204,2205,2206],{"id":2134,"depth":675,"text":2135},{"id":2141,"depth":675,"text":2142},{"id":2152,"depth":675,"text":2153},{"id":2163,"depth":675,"text":2164},{"id":2174,"depth":675,"text":2175},"2026-03-17","The OpenClaw setup guide that puts steps in the right order. Hardware, Node 22, model provider, Telegram, security, skills. 2-4 hours to production.","/img/blog/openclaw-setup-guide-complete.jpg",{},"/blog/openclaw-setup-guide-complete","16 min read",{"title":1215,"description":2208},"OpenClaw Setup Guide: Install in the Right Order (2026)","blog/openclaw-setup-guide-complete",[2217,2218,2219,2220,2221,2222,2223],"OpenClaw setup guide","OpenClaw installation","install OpenClaw 2026","OpenClaw hardware requirements","OpenClaw Telegram setup","OpenClaw security config","OpenClaw getting started","RZ_vqA3S-yoNBoPjFFBpQ_hGI1bmbC9ZDF4aTZ6Zs2Y",{"id":2226,"title":2227,"author":2228,"body":2229,"category":334,"date":2670,"description":2671,"extension":337,"featured":338,"image":2672,"meta":2673,"navigation":341,"path":2674,"readingTime":2675,"seo":2676,"seoTitle":2677,"stem":2678,"tags":2679,"updatedDate":2670,"__hash__":2686},"blog/blog/openclaw-whatsapp-setup.md","How to Connect OpenClaw to WhatsApp (Without the API Headache)",{"name":7,"role":8,"avatar":9},{"type":11,"value":2230,"toc":2651},[2231,2236,2239,2242,2246,2249,2255,2261,2267,2273,2276,2282,2286,2289,2295,2301,2320,2326,2332,2338,2349,2355,2359,2362,2366,2369,2372,2376,2379,2383,2386,2390,2393,2399,2403,2406,2409,2413,2419,2425,2431,2435,2438,2441,2447,2453,2459,2466,2470,2473,2476,2479,2486,2492,2496,2499,2505,2511,2517,2528,2536,2540,2543,2546,2549,2556,2562,2573,2575,2580,2583,2588,2591,2596,2599,2604,2607,2612,2615,2619],[14,2232,2233],{},[17,2234,2235],{},"Forget the unofficial API guides and QR code nightmares. OpenClaw connects to WhatsApp natively. Here's the simple path and the advanced one.",[14,2237,2238],{},"If you've been reading about WhatsApp bots and unofficial APIs and QR code sessions that expire every 20 minutes, stop. OpenClaw connects to WhatsApp natively through the chat interface for most setups. Here's how it actually works.",[14,2240,2241],{},"I spent a full Saturday trying to set up an OpenClaw WhatsApp connection using an unofficial API library before I discovered the native method existed. That Saturday is gone forever. Yours doesn't have to be.",[42,2243,2245],{"id":2244},"the-native-whatsapp-connection-this-is-what-most-people-need","The native WhatsApp connection (this is what most people need)",[14,2247,2248],{},"The native connection links your WhatsApp account directly to your OpenClaw agent. You message the agent through WhatsApp like you'd message a friend. The agent responds in the same chat. No API keys. No webhook URLs. No session management.",[14,2250,2251,2254],{},[125,2252,2253],{},"Step 1: Make sure your OpenClaw agent is running and responsive."," Send a test message through the web interface first. If the agent responds there, the gateway is healthy and ready for channel connections.",[14,2256,2257,2260],{},[125,2258,2259],{},"Step 2: Start the WhatsApp connection from OpenClaw."," In the OpenClaw interface, navigate to the channel connection flow and select WhatsApp. OpenClaw will generate a QR code for WhatsApp Web pairing.",[14,2262,2263,2266],{},[125,2264,2265],{},"Step 3: Scan the QR code from your WhatsApp app."," Open WhatsApp on your phone, go to Linked Devices, and scan the QR code. This is the same process as linking WhatsApp Web to your phone. You're giving OpenClaw access as a linked device.",[14,2268,2269,2272],{},[125,2270,2271],{},"Step 4: Send a test message."," Open WhatsApp and send \"hello\" to the agent chat. If you get a response, you're connected.",[14,2274,2275],{},"The whole process takes about 3-5 minutes. Most of that is waiting for the QR code to appear and getting your phone out to scan it.",[14,2277,2278],{},[66,2279],{"alt":2280,"src":2281},"OpenClaw WhatsApp native connection setup showing QR code pairing process through Linked Devices","/img/blog/openclaw-whatsapp-native-connection.jpg",[42,2283,2285],{"id":2284},"what-works-once-youre-connected","What works once you're connected",[14,2287,2288],{},"Once your OpenClaw WhatsApp setup is complete, the agent works through WhatsApp with the same capabilities as any other channel.",[14,2290,2291,2294],{},[125,2292,2293],{},"Text messages and responses."," Type naturally. Ask questions. Give instructions. The agent responds in the same chat thread. Conversations feel like texting a knowledgeable friend.",[14,2296,2297,2300],{},[125,2298,2299],{},"Voice notes."," This is where WhatsApp genuinely shines for OpenClaw. Send a voice note and the agent processes the audio, transcribes it, and responds in text. You can ramble for two minutes about what you need, and the agent extracts the actual request and acts on it. This is especially useful when you're walking, driving, or just don't feel like typing.",[14,2302,2303,2306,2307,2309,2310,2309,2313,2309,2316,2319],{},[125,2304,2305],{},"All OpenClaw commands work."," The slash commands (",[400,2308,535],{},", ",[400,2311,2312],{},"/memory",[400,2314,2315],{},"/status",[400,2317,2318],{},"/new",") work identically in WhatsApp. Type them in the chat and the agent processes them.",[14,2321,2322,2325],{},[125,2323,2324],{},"Skills execute normally."," Web search, calendar, file operations, and any installed skills work through WhatsApp the same way they work through the web interface or Telegram. The agent receives your message, processes it through the skill pipeline, and sends the result back to WhatsApp.",[14,2327,2328,2331],{},[125,2329,2330],{},"Memory persists across platforms."," If you started a conversation on Telegram and switch to WhatsApp, the agent remembers everything. Same persistent memory, same context, different app. This cross-platform memory is one of the reasons WhatsApp is popular with users who also connect Telegram or Discord to the same agent.",[14,2333,2334,2337],{},[125,2335,2336],{},"Cron jobs deliver to WhatsApp."," Set up a morning briefing and the agent sends it to your WhatsApp at 7 AM. No browser to open. No app to check. The information arrives in the same place as your other messages.",[14,2339,2340,2341,800,2345,2348],{},"For a broader look at ",[50,2342,2344],{"href":2343},"/blog/best-openclaw-use-cases","the best workflows to run through an OpenClaw agent",[50,2346,2347],{"href":2343},"use cases guide"," covers the setups that provide the most value across all channels.",[14,2350,2351],{},[66,2352],{"alt":2353,"src":2354},"OpenClaw WhatsApp features showing voice notes, text chat, slash commands, and cross-platform memory","/img/blog/openclaw-whatsapp-capabilities.jpg",[42,2356,2358],{"id":2357},"whatsapp-specific-things-to-know","WhatsApp-specific things to know",[14,2360,2361],{},"WhatsApp isn't Telegram. A few things behave differently, and knowing them upfront saves confusion.",[820,2363,2365],{"id":2364},"message-formatting","Message formatting",[14,2367,2368],{},"WhatsApp supports basic formatting (bold with asterisks, italic with underscores, monospace with backticks) but doesn't render full Markdown. If your agent generates responses with headers, tables, or complex formatting, they'll appear as plain text in WhatsApp. The content is the same. The visual presentation is simpler.",[14,2370,2371],{},"This rarely matters for conversational interactions but can look cluttered if your agent generates structured reports or formatted data. If formatted output matters to you, Telegram or Discord render Markdown more completely.",[820,2373,2375],{"id":2374},"message-length-limits","Message length limits",[14,2377,2378],{},"WhatsApp has a per-message character limit (roughly 65,000 characters, which is generous). Your agent's responses will almost never hit this limit. But if the agent generates a very long response (a detailed research report, for example), WhatsApp may split it across multiple messages. The content is complete. It just arrives in chunks.",[820,2380,2382],{"id":2381},"media-and-file-sharing","Media and file sharing",[14,2384,2385],{},"The agent can receive images, documents, and voice notes through WhatsApp. Whether it can process them depends on the skills you have installed. Voice note processing works natively. Image analysis requires a vision-capable model. Document processing requires a file reading skill.",[820,2387,2389],{"id":2388},"whatsapp-business-vs-personal-account","WhatsApp Business vs personal account",[14,2391,2392],{},"Both work with the native OpenClaw connection. You don't need a WhatsApp Business account for personal agent use. If you're building a customer-facing bot and want the Business features (business profile, catalogs, auto-replies), a WhatsApp Business account adds those on the WhatsApp side. The OpenClaw connection works the same either way.",[14,2394,2395],{},[66,2396],{"alt":2397,"src":2398},"OpenClaw WhatsApp formatting and media support showing text, voice notes, and file handling differences","/img/blog/openclaw-whatsapp-quirks.jpg",[42,2400,2402],{"id":2401},"when-the-native-connection-isnt-enough","When the native connection isn't enough",[14,2404,2405],{},"Here's where most people get it wrong. They read about the unofficial WhatsApp API and assume they need it. Most don't.",[14,2407,2408],{},"The native connection covers personal use, family use, small team use, and even modest customer-facing scenarios. It breaks down only when you need specific capabilities that WhatsApp Web pairing can't provide.",[820,2410,2412],{"id":2411},"what-the-unofficial-api-approach-gives-you","What the unofficial API approach gives you",[14,2414,2415,2418],{},[125,2416,2417],{},"Multiple agents on one WhatsApp number."," The native connection is one agent per linked account. If you need to route different conversations to different agents based on topic or customer segment, you need API-level access.",[14,2420,2421,2424],{},[125,2422,2423],{},"Automated outbound messages at scale."," The native connection is reactive. Someone messages you, the agent responds. If you need to proactively message hundreds of customers (order updates, marketing campaigns, appointment reminders), you need the WhatsApp Business API.",[14,2426,2427,2430],{},[125,2428,2429],{},"Persistent sessions without phone dependency."," The native connection relies on your phone being online (same as WhatsApp Web). If your phone goes offline, the agent loses the connection. API-based setups run independently of your phone.",[820,2432,2434],{"id":2433},"the-real-risks-of-the-unofficial-api-path","The real risks of the unofficial API path",[14,2436,2437],{},"Stay with me here. This matters.",[14,2439,2440],{},"The unofficial WhatsApp API libraries (like Baileys, whatsapp-web.js, and similar projects) reverse-engineer WhatsApp's protocol. They work. But Meta explicitly prohibits this in their Terms of Service.",[14,2442,2443,2446],{},[125,2444,2445],{},"QR code sessions expire."," The linked device session needs periodic re-authentication. If you don't re-scan the QR code, the connection drops. Some community setups report sessions lasting days. Others report expiry within hours. It's unpredictable.",[14,2448,2449,2452],{},[125,2450,2451],{},"Phone number flagging and banning."," Meta actively detects unofficial API usage. Numbers using unofficial clients get flagged and can be temporarily or permanently banned. Losing your primary phone number to a WhatsApp ban is a real risk, and one that multiple community members have reported.",[14,2454,2455,2458],{},[125,2456,2457],{},"No support or recourse."," If your account gets banned for unofficial API usage, Meta's support won't help. The ban is for violating their terms. You agreed to those terms when you signed up.",[14,2460,2461,2462,2465],{},"For the ",[50,2463,2464],{"href":1853},"broader security considerations of running OpenClaw",", our security guide covers the risks across the entire stack, not just WhatsApp-specific ones.",[820,2467,2469],{"id":2468},"who-should-actually-consider-the-api-path","Who should actually consider the API path",[14,2471,2472],{},"Business-scale operations that need proactive outbound messaging to hundreds or thousands of customers. Companies that need the official WhatsApp Business API (which is separate from the unofficial libraries) for compliance and reliability. High-volume customer support operations where the native connection's session dependency isn't acceptable.",[14,2474,2475],{},"For everyone else, the native connection works. Don't overcomplicate it.",[14,2477,2478],{},"The native WhatsApp connection handles 90% of use cases. The unofficial API adds complexity, instability, and real ban risk. Only go down that path if you specifically need outbound messaging at scale or phone-independent sessions.",[14,2480,2481,2482,2485],{},"If managing WhatsApp connections, session stability, and re-authentication isn't how you want to spend your time, ",[50,2483,2484],{"href":985},"Better Claw handles WhatsApp as a pre-configured channel"," from the dashboard. $19/month per agent, BYOK. Connect your WhatsApp, pick your model, deploy in 60 seconds. The connection management is handled so you don't wake up to a dropped session.",[14,2487,2488],{},[66,2489],{"alt":2490,"src":2491},"OpenClaw WhatsApp native vs unofficial API approach comparison showing risks and use cases","/img/blog/openclaw-whatsapp-api-comparison.jpg",[42,2493,2495],{"id":2494},"telegram-vs-whatsapp-for-openclaw-which-should-you-use","Telegram vs WhatsApp for OpenClaw: which should you use?",[14,2497,2498],{},"This isn't a feature comparison. It's a practical recommendation based on your situation.",[14,2500,2501,2504],{},[125,2502,2503],{},"If you just want to chat with your agent personally:"," Telegram is easier to set up and has no phone dependency for the connection. The native Telegram connection is simpler than WhatsApp's QR-code-based linking. If you don't already use WhatsApp heavily, start with Telegram.",[14,2506,2507,2510],{},[125,2508,2509],{},"If your daily life already runs through WhatsApp:"," Use WhatsApp. The whole point of OpenClaw is meeting you where you already communicate. If every conversation you have is in WhatsApp, adding your AI agent there means you never leave the app. The voice note feature makes WhatsApp especially good for people who prefer talking over typing.",[14,2512,2513,2516],{},[125,2514,2515],{},"If you want to share the agent with family or a small team:"," WhatsApp group chats are more natural for non-technical people. Your family already knows how WhatsApp works. Creating a group and adding the agent (via the native connection) is straightforward. Telegram requires people to install a new app if they don't already have it. The friction difference matters for non-technical users.",[14,2518,2519,2522,2523,2527],{},[125,2520,2521],{},"If you want a customer-facing business bot:"," This is a different conversation entirely. For customer-facing bots at scale, you need the official WhatsApp Business API (not the unofficial libraries), proper compliance, and infrastructure that doesn't depend on your personal phone. For the ",[50,2524,2526],{"href":2525},"/blog/openclaw-agents-for-ecommerce","full ecommerce agent setup including WhatsApp customer support",", our ecommerce guide covers the architecture.",[14,2529,2530,2531,2535],{},"For the companion guide on connecting OpenClaw to Telegram, our ",[50,2532,2534],{"href":2533},"/blog/openclaw-telegram-setup","Telegram setup post"," covers the native connection, BotFather setup, and when you need a dedicated bot.",[42,2537,2539],{"id":2538},"the-part-most-whatsapp-guides-skip","The part most WhatsApp guides skip",[14,2541,2542],{},"Here's what nobody tells you about running your agent on WhatsApp long-term.",[14,2544,2545],{},"WhatsApp has 2.7 billion monthly active users. It's the dominant messaging platform in most of the world outside the US. When you put your OpenClaw agent on WhatsApp, you're putting it on the platform where most of your customers, family, and contacts already spend their time.",[14,2547,2548],{},"That's powerful. It's also a responsibility. Every message your agent sends appears in the same app as messages from your partner, your kids, your boss. A badly configured agent that sends late-night notifications or gives wrong information doesn't just annoy a user. It damages trust in a space that's personal.",[14,2550,2551,2552,2555],{},"Write your ",[400,2553,2554],{},"SOUL.md"," carefully. Set response hours if appropriate. Define escalation rules. Test the agent with friends before exposing it to customers. WhatsApp conversations feel more personal than Telegram or Discord. Your agent's tone should match.",[14,2557,532,2558,2561],{},[50,2559,2560],{"href":1269},"managed vs self-hosted comparison"," covers how different deployment approaches handle multi-channel management, including the WhatsApp-specific connection considerations.",[14,2563,2564,2565,2568,2569],{},"WhatsApp is available as a pre-configured channel on ",[50,2566,2567],{"href":232},"BetterClaw",". You connect your account from the dashboard, no config files or API keys. ",[50,2570,2572],{"href":78,"rel":2571},[54],"One click and your agent is live on WhatsApp.",[42,2574,279],{"id":278},[14,2576,2577],{},[125,2578,2579],{},"How do I set up OpenClaw with WhatsApp?",[14,2581,2582],{},"The fastest method is the native connection through OpenClaw's chat interface. Start the WhatsApp connection in OpenClaw, scan the QR code from your WhatsApp app (same process as linking WhatsApp Web), and send a test message. The whole process takes 3-5 minutes. No unofficial API libraries, no webhook configuration, no API keys needed for personal use.",[14,2584,2585],{},[125,2586,2587],{},"How does WhatsApp compare to Telegram for OpenClaw?",[14,2589,2590],{},"Telegram has an easier native connection (no QR code dependency), better Markdown rendering, and a dedicated bot system for multi-user access. WhatsApp has a larger user base (2.7B+ monthly active users), voice note support that works naturally with OpenClaw, and is the default messaging app in most markets. For personal use, either works. For reaching non-technical users or customers, WhatsApp wins because people already have it installed.",[14,2592,2593],{},[125,2594,2595],{},"Does OpenClaw WhatsApp work with voice notes?",[14,2597,2598],{},"Yes. Send a voice note through WhatsApp and OpenClaw processes the audio, transcribes it, and responds in text. This makes WhatsApp especially useful for hands-free interaction. You can ramble a two-minute request while walking and the agent extracts the actual ask and acts on it. Voice note processing works natively without additional skills or configuration.",[14,2600,2601],{},[125,2602,2603],{},"Does the OpenClaw WhatsApp connection cost anything extra?",[14,2605,2606],{},"No. The WhatsApp connection itself is free. The costs of running an OpenClaw agent are hosting ($12-29/month depending on self-hosted VPS or managed platform) and AI model API fees ($5-30/month depending on model and usage). WhatsApp adds zero additional cost. On managed platforms like BetterClaw ($19/month per agent), WhatsApp is included as one of 15+ pre-configured channels.",[14,2608,2609],{},[125,2610,2611],{},"Is connecting OpenClaw to WhatsApp safe?",[14,2613,2614],{},"The native connection (WhatsApp Web pairing) is as safe as using WhatsApp Web on your computer. The unofficial API route carries real risks: Meta detects unauthorized API usage and can flag or ban your phone number. Multiple community members have reported temporary and permanent bans. For personal and small-scale use, stick with the native connection. For business-scale operations, use the official WhatsApp Business API through proper channels, not unofficial libraries.",[42,2616,2618],{"id":2617},"related-reading","Related Reading",[169,2620,2621,2627,2633,2639,2645],{},[172,2622,2623,2626],{},[50,2624,2625],{"href":2533},"OpenClaw Telegram Setup Guide"," — Companion guide for the other major messaging channel",[172,2628,2629,2632],{},[50,2630,2631],{"href":2343},"Best OpenClaw Use Cases"," — Workflows that work especially well through chat channels",[172,2634,2635,2638],{},[50,2636,2637],{"href":1853},"OpenClaw Security Risks Explained"," — Broader security considerations for any deployment",[172,2640,2641,2644],{},[50,2642,2643],{"href":2525},"OpenClaw Agents for Ecommerce"," — Customer-facing bots at scale",[172,2646,2647,2650],{},[50,2648,2649],{"href":1269},"BetterClaw vs Self-Hosted OpenClaw"," — Multi-channel management across deployment options",{"title":321,"searchDepth":322,"depth":322,"links":2652},[2653,2654,2655,2661,2666,2667,2668,2669],{"id":2244,"depth":322,"text":2245},{"id":2284,"depth":322,"text":2285},{"id":2357,"depth":322,"text":2358,"children":2656},[2657,2658,2659,2660],{"id":2364,"depth":675,"text":2365},{"id":2374,"depth":675,"text":2375},{"id":2381,"depth":675,"text":2382},{"id":2388,"depth":675,"text":2389},{"id":2401,"depth":322,"text":2402,"children":2662},[2663,2664,2665],{"id":2411,"depth":675,"text":2412},{"id":2433,"depth":675,"text":2434},{"id":2468,"depth":675,"text":2469},{"id":2494,"depth":322,"text":2495},{"id":2538,"depth":322,"text":2539},{"id":278,"depth":322,"text":279},{"id":2617,"depth":322,"text":2618},"2026-04-06","OpenClaw connects to WhatsApp natively via QR code. No unofficial APIs needed. Here's the setup, the WhatsApp-specific gotchas, and when you need more.","/img/blog/openclaw-whatsapp-setup.jpg",{},"/blog/openclaw-whatsapp-setup","11 min read",{"title":2227,"description":2671},"OpenClaw WhatsApp Setup: Connect in 5 Minutes","blog/openclaw-whatsapp-setup",[2680,2681,2682,2683,2684,2685],"OpenClaw WhatsApp setup","connect OpenClaw to WhatsApp","OpenClaw WhatsApp integration","OpenClaw WhatsApp bot","OpenClaw WhatsApp not working","OpenClaw WhatsApp guide 2026","lJfJ25-3S5EzaYP9aBqBCVDO0B-917pe7b2jrHLqnYU",1777113709008]