Run OpenClaw Locally With Qwen 3.5: Free Personal AI Agent Setup
Prefer to watch? Watch on YouTube · Subscribe for new builds · Read the transcript
In this video I explain why I think local AI agents hit a tipping point, and I go through the three things that came together to get us there: a mainstream guide to personal AI agents, a fully local OpenClaw setup with Qwen 3.5, and AI chips built for smartwatches. It's for people who aren't technical and want a personal AI agent without paying monthly API bills, or who want to know when cloud models are actually worth the money.
I recorded this in March 2026. These tools change fast, so check the current docs before you set anything up.
What I cover in this video
- 0:00 Why I think local AI agents hit a tipping point
- 0:47 Every.to's guide to setting up your first personal AI agent with OpenClaw
- 3:04 Running OpenClaw with Qwen 3.5 on Telegram, fully local
- 3:51 What hardware you need: an M1 MacBook with 16GB for the 9B model
- 6:52 The cloud camp vs the local camp, and the hybrid approach I use
- 9:55 Qualcomm's Snapdragon Wear Elite chip and Meta's Malibu 2 watch
- 12:11 How agents went from the cloud to your computer, then your phone, then your wrist
- 15:13 What I would do if I were starting from scratch
The three things that came together
1. A major publication told everyone to get a personal agent
Every.to, one of the biggest AI newsletters around, published a full guide called setting up your first personal AI agent, and it uses OpenClaw. It goes through demos and workflows and shares what they call hard-won lessons from building agents that run around the clock. This lines up with Jensen Huang saying at his keynote that pretty much every business needs an OpenClaw strategy. When a mainstream tech publication says personal AI agents are ready for normal people, I take that as a sign we're past the experiment stage.
You don't need a computer science degree for this. I don't have one. I've spent the past year using these tools and building projects. If you have expertise in a certain area, package it up as a skill and hand it to your agent. Only you know that industry the way you do, and that becomes your competitive moat.
2. OpenClaw plus Qwen 3.5, fully local
Someone on the LocalLLaMA subreddit posted a full guide to running OpenClaw with Qwen 3.5 on Telegram, completely locally. Your data never leaves your machine. There are no API calls, no cloud servers, and no monthly bills. Ollama runs Qwen 3.5 on your own hardware, and you talk to it through Telegram like a personal assistant. The post got a lot of traction because people were surprised how easy it was to set up.
A year ago, running a local model smart enough to work as an agent basically required a GPU setup costing around 10 grand. Qwen 3.5 changed that:
- The 9 billion parameter version runs on an M1 MacBook with 16GB of RAM, so a laptop from about four years ago.
- Another LocalLLaMA user ran Qwen 3.5 on an M1 Pro with 16GB and said it replaced their whole two-model agentic setup.
- The 35 billion parameter version runs on 64GB of RAM, which means a Mac Studio or a maxed-out MacBook Pro. That's still consumer hardware.
If you have an old computer sitting in a closet, put it to use. I have several sitting in the corner.
3. AI chips built for your wrist
Qualcomm announced the Snapdragon Wear Elite chip at Mobile World Congress. It's designed to run AI models on smartwatches, with an NPU (neural processing unit) built into the watch chip, so you don't need a phone or internet connection. Qualcomm's CEO said 2026 is the year AI agents go from concept to everyday reality. Meta is also working on a smartwatch code named Malibu 2, built around what they call personal super intelligence. Garmin is moving in this direction too, and there's a wave of AI apps that add personalized insights, diet recommendations, and training plans to watches people already own.
Cloud vs local: the hybrid approach I use
My whole content pipeline runs on a Mac Mini behind me: 12 AI agents working 24/7 on content creation, email management, community management, and research. I use Claude models through Anthropic's $200 Max plan. Claude Opus is my daily driver and Sonnet handles the smaller tasks. For what I'm doing, it's worth it.
Most people don't need that much power. Right now the AI world splits into two camps. The cloud camp pays monthly for models from OpenAI, Anthropic, and Google Gemini. The local camp runs open source models like Qwen, Llama, and Mistral, and those keep getting better with every release. MiniMax 2.7 showed up on OpenRouter with benchmarks that look pretty good.
You don't have to pick one. I teach a hybrid setup:
- Cloud models for the work that matters, like content creation, business logic, and anything customers see. When I need a video script, it goes to Claude Opus and costs me maybe a dollar per script or less.
- Local models for routine work like reminders, simple note taking, basic research, and web scraping. My morning briefing, where my agent checks the weather and my calendar, doesn't need Opus. It could run on a local model and save me money every day.
- Claude checking the work. Let Qwen or another open source model do the grunt work, then have Claude check it every hour. Did the coding finish? Did the landing page finish? Does the code look right?
Think of it like having a senior employee and an intern. You don't pay senior employee rates for someone to sort the mail. For more on how agents fit together, see my AI agents page.
Why agents keep moving closer to you
AI agents started in the cloud, and you had to go to a website to use them. Then they moved to your computer, which is where OpenClaw is now. Then to your phone through Telegram and WhatsApp. Next is your wrist. Each step gives the agent more context about your life. An agent on your computer only knows what's on your computer. One on your phone knows your location, messages, and calendar. One on your wrist knows all of that, plus your heart rate, sleep patterns, and activity level. Add the ability to book appointments, send messages, or control your smart home, and you have something that can save you real time.
How I would start from scratch
- Set up OpenClaw on whatever machine you have: Mac, Windows, or Linux. If you don't want to pay for API calls yet, use Ollama with Qwen 3.5. It's free.
- Connect a messaging app. Telegram is the easiest and the one I prefer. WhatsApp works too, and so do Discord and iMessage. Now you have a personal AI agent on your phone.
- Start small. Have it summarize articles, remind you about tasks, and check your email. Get comfortable having an AI assistant.
- Upgrade once you see the value. Get a cloud API key or a subscription, switch to Sonnet or Opus, and start building real automations: content pipelines, business workflows, and customer support systems.
Starting local means you can learn how AI agents work without spending anything. You can experiment, break things, and rebuild them for free, then add cloud models on top when you're ready.
Key takeaways
- OpenClaw with Qwen 3.5 through Ollama gives you a fully local personal agent with no API bills, and your data stays on your machine.
- The 9B Qwen 3.5 model runs on an M1 MacBook with 16GB of RAM. The 35B model needs 64GB.
- A hybrid setup works well: local models for routine tasks, and Claude for high value work and for checking the local model's output.
- Chips like Qualcomm's Snapdragon Wear Elite point to agents running on your wrist with more context about your day.
- Start free and local, get comfortable, and upgrade to cloud models when you see the value.
FAQ
Can you run OpenClaw completely locally for free?
Yes. You can run Qwen 3.5 through Ollama, point OpenClaw at it, and connect it to Telegram. There are no API calls or cloud servers, and your data never leaves your machine.
What hardware do I need to run Qwen 3.5 with OpenClaw?
The 9 billion parameter version of Qwen 3.5 runs on an M1 MacBook with 16GB of RAM. The 35 billion parameter version needs 64GB of RAM, so a Mac Studio or a maxed-out MacBook Pro.
Should I use local models or cloud models like Claude for my AI agent?
I recommend a hybrid setup. Use local models for routine tasks like reminders, note taking, and basic research. Use cloud models like Claude Opus for content, business logic, customer-facing work, and checking the local model's output.
Which messaging app is best for connecting to OpenClaw?
Telegram is the easiest and the one I prefer. WhatsApp, Discord, and iMessage also work.
Chapters
Transcript
Read the full transcript
0:00I'm going to tell you why local AI agents just hit a tipping point in this whole AI industry and no one seems to be talking about it. I'm not talking about some far off future thing. I'm talking about right now, today, March 18th, people are running full AI agents on their laptops, on their phones, on their smartwatches. This is a straight out of a sci-fi movie. No cloud, no monthly API bills, just your hardware doing the thinking for you. I'm going to back up and break down what is actually happening because these three things converged this month that changed basically everything in this whole AI space. If you're not paying attention to this, you're going to be left behind. That's plain and simple. I hope you're subscribed to the channel in that case because we're releasing AI updates about
0:47six times a day. So, the first thing to do, which is one of the biggest, is this AI newsletter, one of the biggest out there, just published a full guide called setting up your first personal AI agent. It will be linked down below. Check it out. Using OpenClaw, by the way. And this is not some random Reddit post. This is a major publication telling their entire audience, "Hey, you need a personal AI agent running 24/7." And that's kind of goes back to what uh Jensen Huang said at the keynote the other day that pretty much every business needs a OpenClaw strategy. So, confirming that in this big article. You know, in the article, they walked through some demos, some workflows, and they called hard-won lessons from
1:33building agents that run around the clock. The fact that a mainstream tech publication is now saying personal AI agents are ready for normal people, that's a pretty big sign. And if you're not doing it already, what's holding you back? It's free to get it installed. Get OpenClaw going and build some things, break some things. You know, that tells me we've basically crossed that tipping point, we've crossed that line. This is no longer like an experiment mode. This isn't a demo or a tutorial. Full on people are making tons of money building AI automations. I go back to saying like if you have domain expertise or you're really skilled in a certain area, package that up as a skill, hand it off to your agent, and let them do the work. And that's going to create a great
2:18competitive moat for you because you only know that particular skill and that particular industry. So, you know, this isn't just for developers anymore. This is just for regular people like you and me. I don't have a 10-year tech background with a computer science degree. You know, I've just been using these tools for the past year and building cool projects. And that's what I want to tell you, don't feel like you need to be a super technical person to build your first successful product. So, regular people are just setting these up and they're going crazy. They're getting real value from them and they're making money. I guess the second thing, too, is somebody on the local Llama subreddit post, he posted a full guide on running OpenClaw with Qwen 3.5 on Telegram completely locally. And when I say
3:04completely locally, I mean your data never leaves your machine, which is really nice. No API calls, no cloud servers, no monthly bills, just a Llama running Qwen 3.5 on your own hardware, connected to Telegram, responding to like a personal assistant. And people who value security, that should be a really nice glimpse into the future. You know, no API calls, no dependency on cloud servers, being able to talk to a great model like Qwen 3.5 and have a personal assistant at your fingertips. This post got a ton of traction because people were shocked at how easy it actually is to set up and get it going. A year ago, running a local model that was actually smart enough to be useful and having it as an agent was basically impossible and useless unless you had
3:51some crazy GPU setup that cost like 10 grand. But with Qwen 3.5, it pretty much changed that. The 9 billion parameter version runs on literally an M1 MacBook with 16 gigs of RAM. So, a laptop from four years ago, basically, can now run a full local AI agent. Crazy. And someone else on local Llama ran Qwen 3.5 on an M1 Pro with 16 gigs and said it replaced their entire two-model agentic setup. One model doing the work of two. And the bigger versions, the 35 billion parameter model runs on 64 gigs of RAM. That's pretty good. And that's a Mac Studio or a maxed-out MacBook Pro. Still consumer hardware, still something you can buy at the Apple Store or Best Buy.
4:36We're not talking about server racks or data centers. We're talking about a computer that you probably have in your closet sitting somewhere. Put it to use. I have several that are just sitting in the corner. Real quick, before we keep going, if you're watching this and you want to actually build with some of these tools, not just watch videos about them, then you're going to want to check out our community down below, Shipping School. We have a full Claude code course, a full OpenClaw course, and four live bootcamps every single week where we actually help you get set up from scratch. Like actually set this thing up, not just watch a tutorial and figure it out by yourself. And we also provide one-on-one coaching, so you could book a call with me, we could share screens, and I can help you get Claude code or OpenClaw running on your machine. That's
5:21it, no fluff. I built this community because watching YouTube's only gets you so far. We launched it just three days ago and we have over 55 members. You need people around you who are actually building, people who hold you accountable, and coaches who can help you when you get stuck. I'll put the link in the description down below. Get in now before the price goes up. And I'm going to tell you why this matters for you. Right now, I run the entire content pipeline on a Mac Mini behind me. 12 AI agents working 24/7. Mainly focused towards content creation and email management, helping me manage the community, and doing a lot of research. Uh you know, I'm using Claude models, but that's fine. I use the max plan through Anthropic. Claude Opus is the
6:06daily driver, Sonnet handles the smaller tasks, and those are the API calls. And they're worth every penny because for what you get for $200 is insane. And for what I'm doing, uh the quality is pretty good. So, not everyone needs that level of power. You know, if you're just getting started and you want a personal assistant that can help you organize your day, answer questions, maybe manage your calendar, you know, send you simple reminders, you can literally do that for free now. Download Qwen 3.5, get it going. $0. Just download a Llama, pull the Qwen 3.5 model, point to OpenClaw, and connect to Telegram or WhatsApp or Discord or whatever you want to use. You can use iMessage. And that's it. You have a personal AI agent in your pocket. And I'm going to tell you why I think this is super important and you need to pay attention right now. For the bigger picture, the AI industry right now is
6:52split into basically two types of camps. You've got the cloud camp, like me, people who use OpenAI and Anthropic's models and Google Gemini. They want you paying monthly for access to their models. That's how they're making money. And look, those models are amazing and they're great and they're worth it. I use Claude every single day and it's literally the backbone of my entire content business, my school community, my resale business, my personal life, our kids' homeschooling, our kids' chore chart, my wife's personal development and diet plan. Like it's literally woven through my whole family. But you know, and then you got the local camp. You know, that's the cloud camp, me, and then you got the local camp. There there's no reason why you couldn't build out the same type of level of infrastructure doing something locally.
7:38You know, open-source models like Qwen and Llama Mistral and these models are getting better every day and at a rate that honestly scares me because with every single update, these get better and better. MiniMax 2.7 just dropped on OpenRouter and the settings and the parameters and the benchmarks are looking pretty good. So, this should scare cloud companies, by the way, that as this technology gets better, you're going to be able to have less demanding hardware to be able to run these models. You know, every time an open-source model gets good enough, that's one more reason for people not to pay for API calls. This is where it gets more interesting. You don't have to choose one or the other. What I do and what I teach in Shipping School is more of like a hybrid approach. You know, use the
8:23powerful cloud models for the stuff that matters, your content creation, your business logic, or like your customer-facing stuff. Like a good setup is like maybe have Qwen or some of these open-source models do a lot of the grunt work and then have like Claude check the work. You know, check the work every hour. Hey, did that coding finish? Did the landing page finish? Check the code if it even looks good. That will save you a ton of money in the long run. You know, if you're using local models for the routine tasks, that's a good way to go. Your reminders or simple like note-taking or like simple basic research or web scraping, that's perfect. That way you get the best of both worlds. You get top-tier intelligence where you need it and then the free local processing for everything else. It's basically like having a senior employee and an intern
9:09at the same time. Like you don't need to pay a senior employee rates for someone to sort your mail. It just makes sense. And let me get specific about what this looks like in practice. So, in my setup, when I need a video script written, that goes to Claude Opus. It's one of the best models for script writing and outlines. And I believe it's the best model in the world, but hey, I'm biased. Don't stone me. Um and it's very, you know, creative, personable, and it cost me maybe a dollar per script or less. You know, totally worth it. But when I need my agent to check the weather and look at my calendar or send me like the daily briefing in the morning on what's going on, that doesn't need to be Opus. That could have run on a local model and save me money every single day. And for someone who's just starting out and
9:55doesn't have the budget for API calls or you can't afford the $200 a month plan on with Anthropic, the local approach means you can just learn everything about AI agents without spending a dime. You can experiment, you can break things, rebuild it all for free, and not have to be so worried about spending so much money. And, you know, when you're at ready to level up, just upgrade. You know, you can add the cloud models on top. You know, that's probably the smartest way to go about doing it. The third thing that happened and this one is pretty crazy is Qualcomm just announced the Snapdragon Wear Elite chip at Mobile World Congress. And this chip is designed specifically to run AI models on smartwatches. This is insane. On your wrist, having an AI agent, I
10:41don't even know what to think about that. But, we're talking about an NPU, a neural processing unit, built into a watch chip that can run AI locally. No phone needed. No internet needed. Your watch just syncs for you. Qualcomm CEO basically just said, you know, 2026 is the year of AI agents move from concept to everyday reality, and that's on your watch. And he's not wrong because Meta is also working on a smartwatch code name Malibu, too, that's built around what they're calling personal super intelligence. Wait till it makes its way into your smart glasses. So, a watch that knows your habits, your schedule, your preferences, and proactively like helps you throughout the day. Even Garmin's getting into it. They are, you know, building similar, you know, watches out
11:26there. You know, they're all integrating this, basically. So, there's a new wave of AI apps that are turning your existing like Garmin watch into like an AI wearable, basically. It takes like the health and fitness tracking and adds personalized insights. I like that. You know, it adds diet recommendations, training plans, AI-powered agents running your whole food pipeline, basically. I actually am a big fan of that. So, it's not just big tech companies making these new devices, existing devices are getting smarter, too. You know, I know some of you are thinking like, "What do I even care about this?" But, there is actually a lot of people that wear smartwatches. So, here's why I think this is important for the trajectory and people to learn what's even going on. AI agents lived in the cloud, and you had to go to a
12:11website to use them. Then they moved to your computer, and that's where we are kind of right now with OpenClaw. And now they are moving to things like your phone and your watch with local models, you know, using Telegram and WhatsApp application. And now, they're moving to your wrist. So, every step of AI brings this closer to you and integrates into your personal life. More personal, and it's always there, more useful. And the closer it gets, the more value it does provide because it has more context about your life. Here's why I like this for leveling up skills. You're going to hear me in these videos all the time talking about you need to learn skills and you need to learn them as fast as possible. What's going to keep you from being replaced at your job and your startup failing is learning new skills as fast as possible. And how do you do
12:57that? You work with very smart large language models, and you just help have them help you learn new skills as you go. You know, this is a huge thing that we teach in our community is learning how to use these tools are going to level up your skills greatly and going to make you irreplaceable as you learn to build whatever you're trying to build. So, if you think about it this way, you know, an AI agent on your computer only knows what's on your computer. An AI agent on your phone only knows your location, your messages, your calendar. That's great. But, an AI agent on your wrist knows all of that plus your heart rate, your sleep patterns, your activity level. Then, the more context it gets, the more AI has, and then the better it can help you. So, you know, this this isn't about a cool
13:42gadget. This isn't, you know, more of like a Twilight Zone type thing. This is about AI that actually understands you as a person. And when you combine that context with the ability to take action and like actually book appointments and to send messages or to control your smart home. Now, you're talking about something that's genuinely life-changing. It will save you more time. If you're struggling to keep up with content, well, I'm about to save you about 40 days' worth of work. [music] I built something called Content Machine. It's 10 AI agents that run on the OpenClaw orchestration, and they handle everything. Scripts, [music] thumbnails, X posts, blogs, outreach, clips, newsletters, all of it. So, I went from 1,000 subscribers to 4,000
14:28subscribers on YouTube in 7 days using this exact system. Every single morning I wake up, [music] and the content's already done. I spend maybe 15, 20 minutes reviewing and approving them, and I move on with my day. It works for any niche, fitness, finance, real estate, marketing, whatever you're building, and it is 100% completely customizable to your [music] use case. So, you get the mission control dashboard, all of the cron jobs, everything I've built over the last 40 days helping me gain more and more people to subscribe and join the community. So, you plug in your own thing, and it molds it to you. It learns how you talk, and it writes [music] so it doesn't sound like AI slop. $97 one time. It's not a subscription. I'll put the link down below, and you'll
15:13thank me later. I'm going to circle back to what this really matters for you watching this, you know, because I know my audience, you're builders, you're entrepreneurs at heart, you're people who want to create something, and I love that. So do I. We have that in common. Can we hang out? Maybe you want to build a SaaS product. Maybe you want to automate your existing business. Maybe you just want to stop doing repetitive tasks and get your time back. Whatever that is, understanding this local AI movement is going to be a pretty big one for you. Here's what I would do if I were starting from scratch today. The first step would be setting up OpenClaw. Like, come on now. Whatever machine you have, Mac, Windows, Linux, it doesn't really matter. If you don't want to pay for API calls yet, just use a llama with Quinn 3.5. It's free. Step two, connect
15:58your messaging app to it. Telegram is the easiest. WhatsApp is also good, but I prefer Telegram. You know, and then you would have a personal AI agent you can talk to on your phone, and that's pretty insane. The third thing would be starting small. Have it like summarizing articles for you. Have it remind you about tasks. Have it check your email, things like that. Just get comfortable with having an AI assistant. I guess the next step would be once you see the value, that's when you can upgrade. You know, get a cloud API key or go through the subscription. Switch to Sonus or Sonet or Opus, and then you can start building real automations, content pipelines, business workflows, customer support systems. That's when things get really powerful. And if you want to skip the learning curve and just have someone show you exactly how to do all of this,
16:43that's literally why we started the community. With over 90 members in 9 days, we are growing insanely fast. We do four live calls every single week. I also do one-on-one to help you take the next step in this journey. You know, we're building real stuff, and we're real people, not just a bunch of hey gen avatars in there. We're real people with real problems that we want to solve, and I just love the community so much, and I truly have a passion for helping people take the next step along in their journey. So, if you're just getting started, we'll help you. We have a beginners call happening in 7 hours where people are like, "Hey, I discovered OpenClaw. I have all of these ideas, but I don't know how to get it installed." We also have a deep dive call. We have a ship and roast call where, you know, maybe you have something shipped and something broke.
17:29We will help critique it and help you take the next step in the journey. So, we're building real automations. People are making real money, and we do the four live calls for a reason, but we're looking to build it out to five calls here in the next week, and then meet people where they're at, as well as doing several calls throughout the day. Our whole goal is to have this huge network of people where we build together. This isn't just some course that you watch. This is a thing where we can actually build together live, and that's the whole goal. And that's what makes it different than any other community. We're being more for more thoughtful about actually working with people during the live calls. You know, so like, for example, like if you're stuck at 2:00 a.m. trying to figure out why your agent isn't responding, you can
18:14just send me a message, and I'll get back to you. I'll help you figure it out. And, you know, the thing that's really what I want to take away from all this is like, you know, we are at that inflection point right now. You know, the combination of open-source models is getting good enough to be useful, and platforms like OpenClaw make it easy to set up these agents. The hardware is getting more powerful to run on AI locally, and it can run on just your laptop or your watch in the future. That combination is going to change how everyone works within the next 12 months. And the people who start now, and the people who learn how to set this up and customize their own AI agent, those are the people who are going to have a massive advantage. I learned this the hard way, and a year ago I was still doing everything manually, you know, writing all the outlines, posting to social media one at a time, manually checking my
19:01email and calendar every morning. Now I have AI agents do that for me while I sleep, and I just get a morning briefing every day. And my output has gone up, you know, at least 5 to 10x while my actual work time has gone down. I'm not saying that to flex or anything. My mission statement for my personal life and business with the use of AI is, "How can I get more time back?" I want to spend time with my kids and my family, and and I'm sure you do, too, if that resonates with you. You know, I've been working 40 to 60 hours a week since I became an entrepreneur. It's time to now start handling have AI agents handle the the hard heavy lifting stuff so I could focus on making content and enjoying my life, right? So, that's what happens when you let AI handle the repetitive
19:46stuff. So, you know, emails, stuff like that, let AI do that. You can focus on the creative stuff and the stuff that actually moves the needle for you. So, whether you start a free local setup or go all-in with cloud models, I just say just start. Don't sit on the sidelines watching everyone else build while you wait for the perfect moment that does not exist. The perfect moment was like four years ago. The next best moment is now, guys. And, you know, if you want to be around like-minded people, I'll put the link down below. We have 90 members in nine days. Everyone's learning and growing from all walks of life, from non-technical entrepreneurs people who have set up AWS infrastructure and their database. Check it out, link below. Get in there before the price goes up, and we'll see in the next one. Have a blessed day.
Ready to start building with AI?
Join the Shipping Skool community free. Paid members get live bootcamp calls every week and weekly-updated courses.
Join Shipping SkoolKeep reading
She Runs 9 AI Agents That Manage Her Entire Life. Here Is Her Exact Playbook.
A startup founder scaled from one AI agent to nine, managing her entire life and two businesses. Here is her exact playbook, mistakes included.
OpenClaw Memory Fix: 4-Layer Setup With Supermemory and Neon
To stop OpenClaw from forgetting, I use four memory layers: bootstrap markdown files, daily memory logs, Supermemory for semantic memory, and Neon (Postgres) for structured data. Every agent searches both systems before a run and writes back to both after it. Lossless Claw and QMD add more on top.
OpenClaw CRM: How DenchClaw Runs Your Sales Pipeline Locally
DenchClaw is a local CRM framework built on OpenClaw. You manage your pipeline by talking to the agent, which can import leads, change views, and run browser-assisted sales tasks. Your data stays in a DuckDB file on your own machine. Start with narrow permissions and clear stage definitions.
