Open Source Friday: OpenClaw with Hannes Rudolph

Vídeo em Inglês, publicado pelo canal GitHub.

About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.

Video summary

The video features a discussion about OpenClaw, a popular open-source always-on AI agent project. The guest, Hannis Rudolph, community manager and maintainer at the OpenClaw Foundation, demonstrates how the tool interacts with a computer, manages files, handles tasks like checking emails and organizing documents, and discusses the project's philosophy, community collaboration, and security aspects.

Key points

• OpenClaw is described as an always-on open-source agent that has access to a computer environment, enabling it to perform tasks like browsing, checking emails, and managing local files.
• The OpenClaw Foundation operates as a nonprofit organization aimed at democratizing the always-on agent space.
• Hannis Rudolph demonstrates a personal instance of OpenClaw (named Mousie) performing tasks such as creating daily dashboards from emails and organizing document folders.
• The project relies on users bringing their own AI models (including local models or provider-based options) rather than supplying its own proprietary AI.
• OpenClaw incorporates features like memory cores, auto skill learning, and server-side compaction to manage long-running tasks and context.
• Community contributions and open-source collaboration are emphasized as core drivers for the project's evolution, with support available via Discord and community forums.

Topics

OpenClawAI agentsOpen sourceAutomationComputer useLocal modelsCommunity collaborationTask management

Video transcript

Transcript from available captions; it may contain errors.

Heat. Heat. Heat. Heat. N. Heat. Heat. Heat. Heat. Yeah, white go. Welcome, welcome everybody. I am super excited to be here for yet another open source Friday. My name is Dr. Sarah Kaiser and I am um turning off the reverb. I am really excited for our guest this week. Um, I probably like many folks watching here have been trying out the huge ecosystem of AI uh developer tooling and we have an just kind of an iconic project for this week. So without further ado, uh I'm going to bring in my friend Hus who is a maintainer on OpenClaw. >> Hello. How are you? >> Uh doing great. doing great. How's your Friday? >> My Friday so far so good. Um, I'll be honest, I'm a little nervous being on someone else's broadcast podcast. So, to everybody, my name is Hannis Rudolph and I am a community manager, Devril, and maintainer at the OpenClaw uh foundation. So, nice to meet you. We uh it folks are pretty nice here. It's all good. >> Yeah. I just saw you guys have you guys had Peter from from Clawbot back in the day um >> on on there and I thought, "Oh man, now I've got to, you know, o uh I bet he he did real well." I something I thought about that yesterday, day before um Anthropic came out with their their openclaw type offering. Always on agent, I guess, is the category. And they missed a real opportunity to call it Claudebot. I mean, oh no. Um, yeah, for folks in chat, uh, let us know where you're joining from. Uh, I know I'm on the West Coast, so this is kind of a nice morning. I just got my coffee. Uh hopefully that'll be enough for >> I've got I've got two. >> Yeah, >> we're we're on to hydrating with just some water. But uh >> there's water in this coffee. I promise you it's it's not just the beans. >> That's true. >> Um awesome. So, uh I feel like uh a number of folks here might know what OpenClaw is, but can you give us like the TLDDR of what it is? >> What is OpenClaw? OpenClaw is the I think first of or one of the more popular um always on open- source um agents. always on agent is the category where basically you you could you could text it or open it up in an app or web web browser and say hey do this or do that and it and it does that thing. I know you can always you know that's what you do with chat GPT uh but this is different. It has a computer it has access to. So when you are saying you know um check my email um or check my email every morning and tell me what people are saying it doesn't simply you know run a program to do it. It it will quite often go and check your email with with a mouse. Uh no email is not a good example in this case because you can easily check it without being thought of as spam. But let's say I wanted to check my LinkedIn posts every day. There's not a way to check your LinkedIn posts um through um a tool, but OpenClaw could just go and open the browser and scroll through it like I would the same sort of way I would and then give me a report and go, "Okay, here's what's going on." And it would probably just say there's lots of people posting motivational stuff. But um >> that is a lot of my LinkedIn, too. Yeah. So, yeah, I I don't know if that that gives you good insight into what Open CLY is. Um, its relevance is that I think it became the most most starred um repo. It passed uh Linux overnight practically. It just, you know, people really seem to have an appetite for this and um there's there's it it's spawned a whole um category because we now have Grockbot, we have Muse, we have Anthropics, I'm going to call it Cloudbot, um we have Dots, u and all these types of things. and they are largely inspired by what we did and are still doing. And that's the whole point. We're not they didn't copy us. We gave this formula, you know, these tools often not as polished so that companies could use them and the the technology could be advanced and that also means free offerings and all that. They all get this and so, you know, we're super interested in the open- source side of this. We're run by the Open Claw Foundation, which is a nonprofit. We're not a for-profit in any way. Um, and our goal is to democratize this always on agent space. So, yeah, that's us. I think it's really awesome uh as developers as we're trying to kind of understand how AI tools are going to work for us, how we're going to work with them to have a project like OpenCloud that can kind of be where everybody works stuff out in public collaboratively >> and then yeah, as you say, if other people want to kind of refer refine and do their own productization, like it's building off of kind of But the essence of open source is we're all collaborating to build things together. So yeah, it's it's been a really cool project. I've been following it mostly since it started. Um I used it a lot on my my personal computer. So um yeah. >> So it's sort of cool. We're going both directions. We have companies like uh like Meta that have built Muse that is a really cool tool and it um it's based off of Open Claw in a serious way. Like I don't know if it still does but it had um we pioneered something called the soul.md file that had very specific wording. This is basically the the soul of your your agent. And if you look at, you know, dig into your Muse setup, it has the exact wording from our file. So, it was pulled straight over and they absolutely built their thing from the ground up um inspired by OpenClaw. And so that's one side that's really cool and we are honored and we hope we can continue to put ideas out there that they can take advantage of to build awesome tools for people cuz everybody, you know, I mean, people talk trash about all the big companies, but you know what? I I love using my Facebook. It it's it's handy. It has its purposes. So, uh, and then we have the other side, which is Microsoft um, and Microsoft Scout Autopilot, I guess it's is what it's called. And that isn't simply inspired by OpenClaw. That's built on top of Open Claw. >> And so we're seeing um these aren't the only two examples, but we're seeing inspiration on both sides, you know, copy copycats and then building on top of. And we love both. So copycat's not a negative term in this >> uh cover like a music cover. >> Yeah. >> Um we've got a lot of good questions in chat actually already. Um but I want to make sure before we get distracted by those. Do you have a demo you want to show us >> because I'm guessing some of those might be addressed by that. >> Okay. No, I do. I do. I don't know. >> Sorry. That was on the homework sheet. >> Yes, I know. I know. Um, you guys are like, "You don't have to, but you have to do a demo." So, okay, let's let's share it. We'll share my personal claw >> here. You can load it up screen. Its name is Mousie. It is um named after I I used Chatubt to bring to life one of my children's stuffed animals and give it claws. sort of looks creepy. I need to fix that. >> So, >> can you um zoom in like bigify the fonts at all? >> All right, cool. Now, yep, we're good. >> Okay. So, um it just looks like JGPT, doesn't it? Just another another one of these tools. Uh right now I have going on I have I'm transferring uh RSPS. Those are those are savings for my children's um education. I'm transferring them from one company to another. I'm doing some paperwork filing orthodontic claims, but those aren't my uh my the demo. So the reason why I only see three things here is because when I'm done with them, I actually archive them. M >> so this sort of acts as long as these are open I have something I need to do. >> Um and that that's part of my demo is showing how I use it, how I keep track of it. One thing that throws me off is I'm I like the idea of using chat GPT where I ask questions and then I have all these chats that are open and it's like I just had a question and now it's gone. Like um I don't know what I have left to still do. So in this case, let's say school emails. My kids have lots of I get emails all the time. So, I'm gonna I'm going to uh dashboard create a dashboard. um create a dashboard that refreshes daily with my um school activities based on emails that come in. have it refresh each day at 5:00 p.m. Okay, there's one apparently. Oh man. Okay, so I haven't enabled that. Um, enable it then. >> Okay, so um >> I always love when I work with model and it says I can't do that Dave and I just say M do it though. >> I'm going to start another one off while that one works. Organize my iCloud uh documents folder. >> I don't know about you, but but my documents folder. Yeah, exactly. I thought one day this would be cool to deal do. Um >> I've enabled it. Oh, well that was cool. Um that was not staged. I was sort of my my heart sank when that happened. So I was like, "Oh no, we we've been um making changes. I'm running uh beta and we're making changes adding, you know, making sure things are secure. We want to give it access to computer, but we we don't want to give a bad actor access. So there's this balance between yolo and so we yeah, we we need to make this work and that sometimes the agent is a bit restricted and so we're fixing that. We don't want it to be the case. And in this case, it it wasn't one of those. So, I'm glad. Okay. So, we have that going. Oh my. Your iCloud document makes his personal paperwork. I'd recommend a light tidy. yada yada yada. Um H. Okay. Sure. So, how do you go about doing your own risk analysis when like you know I have my tax documents and things like that in say a folder like that. Um, when do you like what's your cut off for like this is something I'm okay to say sure to versus uh um I don't want it I'm worried about it making mistakes and I don't want it to touch that. Well, in this case, um, it's running a model GPT 6.1 soul that is going to its behavior. It's going to shy away from reading documents that might have um things it shouldn't see in there. Now, that doesn't help when you actually want it to read those. So, there's two approach to this. One is to use a provider you trust and you know turn off the settings you know don't train with my account that type of thing but that isn't enough for a lot of people and that's totally understandable and that's where it comes uh comes into using local models so we've talked about Grockbot and Muse a little bit and those are great they use their AI and OpenClaw doesn't use our AI because we don't have AI you use whatever AI you select In this case, I'm using my chat GBT subscription, but you can use a local model. And the price for this type of thing is coming down. It I I would not say you can simply um download a local model and have open claw work seamlessly. Uh I would advise uh doing some research into it. We did a podcast episode yesterday. Um, you can find it on our YouTube channel about uh we had Nvidia as a guest. Um, and uh we talked about the I think the DJX Spark laptop that was announced last Thursday. It was a collaboration with Microsoft. So you guys and Nvidia and we talked about this problem when to use the local model how to you know how to get on that path. So I don't know if that answers your question or not. In this case, I know this I'm confident this model's not going to be reading my documents. But that being said, it it could. So, I'm not cons, you know, I I I >> I know I don't have any >> Everybody has their own kind of personal threshold. So, I'm always curious learning to uh understand where people's are cuz um I'm unfortunately I think a very paranoid person because I've worked in security. So, um I'm always trying to assess is my threshold too high? Is it you know kind of where is everybody else at? >> Yeah. Um >> so I'll let you keep going. I'm going to tell it to dig deeper and um organize subfolders. Um we have personal business projects archive. Um archive is not oops archive is not actually archive but just where I dumped things and never organized them. That's like my temp inside of temp inside of actually temp inside me in all caps. >> I Yeah. And then you're like, what? Why? Why do I delete? What? This is a long last changed 2012. Um, >> so I just used archive because technically if I'm not touching it anymore, I guess it's archived. Um, I'm keeping it. Uh, but yeah, it's not. So, anyways, I'm I'm getting it to work on that. Um, this isn't as exciting because I didn't show you a before and after of what that folder looked like, but if anybody um has a great mess of documents and folders, they can get their their agent to sort through this. That's not exactly the most groundbreaking uh implementation or use of this technology, but if you think about it, you can apply this to all sorts of things. you could uh get it to go through documents for you um and actually help you figure out which ones you want to keep, which ones you don't, which ones apply to certain things. My brother uses uh it for renaming, you know, going through and renaming all these documents. Now, again, this is something you can do with other tools quite often. We have, you know, we're not not the only guy in town, but um yeah, this can easily be done. sorting files is something that can easily be done with a local model. So, >> okay, >> so per >> perhaps relatedly while it's thinking, um, we had a great question from Aribba. Uh, says, "I'm curious how open claw hands long handles longunning tasks that require an AI to remember context, make decisions, and recover without uh interaction." Uh we have several aspects. One is our memory core which we are working on improving. It um gives the model access to past or other sessions. In this case, I have it. So I could say what did I uh talk to you about yesterday and it could pull a summary of the individual sessions because those sessions all exist in log files. Um and so it and and on top of that sorry we have our memory core which embeds you know important details um and then we have skills auto skill learning so for procedural activities and um we also for a longunning task we have I'm using compaction in this case it is um server side compaction because openai's GPT compaction is quite effective So on top of the compaction where it you know basically summarizes the relevant points once you hit max context to to zip it down. This session can also look at itself to look for keywords in its past history if that makes any sense. >> Oh yeah yeah. So experiment >> and it works very well. >> There there are no right answers here. It's just what are we trying to see to make it work? Um >> compaction is works great. Um so is there support for compaction apart from from the provider in openclaw or >> yes >> yeah well local model obviously then it would openclaw can handle the compaction >> the cool thing is that um so openclaw is not for people that want a one-click solution. we are pushing the bounds of this technology. And it's not that we don't want to make it simple. It's that we're so busy adding to and refining the um methods, the technology that it's not always simple. We And so as we simplify it, we then add more stuff that makes things complicated. So, that being said, if you're a tinkerer and you decide to use OpenC Claw, it's it's not it's not impossibly difficult by any means. It look, it's easier than using Linux, I swear. Um, and this is that that that is not a shot at Linux. No, no, by any means. Which, by the way, it's really easy to use Linux if you can just get your claw to connect to it with remote desktop um or SSH. So anyways, um it the cool thing is that you can change you can create plugins or even modify source code to get it to behave in a way you want it to or to have a feature that you want to. So, and it it can be a bit daunting with 4 million lines of code, but if you if you have uh a bit of a budget and a subscription for um Claude or or you know, chat GBT, etc. Then you can just um get it to do the research, dig in, make the changes, trigger the rebuild. So, um, here where we are on the dashboard. Enabled your school activities dashboard. Um, okay. Where's my dashboard? Uh, dashboards. Okay. This is top secret confidential information. Okay. Okay. Technically, it's personal. You might see kids' names, teachers names. I'm not Oh, hey, there's my hudolfphmail in case you're wondering. That's my email address. Um, not that hard to find if you Google. So, uh, they go to Joseph Mdemco. Please don't dox me. I'm doxing myself. Um, this stuff's not hard to find, okay? So, I'm not I'm not giving you their phone numbers. Um, anyways, here we're listing various things that came up. Uh, other things that are coming up now. I have links to the actual email. So, if I click this, it'll take me right to the Gmail. >> And this will refresh every day. Now, it's still real rough. So, let's uh unexpand this. And we could we could make Okay, this is ugly. Make it easy to look at. Okay. So, we're going to tell that to do its thing and we're going to go look at our organization. Okay. So, see here, while we were working, it learned about organizing for a future organization. And now it has this >> Okay. >> Yeah. Now, these skills, if they aren't used for a long period of time, it it archives them because if if you don't use it, let's say for 30 days, um I don't know what the default. I think that's the default, but you can most likely there's a setting for it. If there's not, I'm sorry I said there was. Um I have a hard time keeping up on how much it I use this software like extensively since uh January 15th, and I can't even keep up to all the possible things that change. So anyways, um it created a skill and so next time um it will use this but also as we continue to go it will update that. Can you inspect or drill into what the memory is? Um in this case the skill yeah there is uh skills uh skill workshop. Which one was that? See the UI could definitely use some love. >> How to organize. >> Uh oh yeah. Um I don't know. >> It seems like there is >> there there is >> I feel your pain about uh using dog food and betas of things. Uh I mean I am regular. >> See it keeps track of how many uses there are. >> Yeah. Yeah. >> Okay. This is sorted by uses. So we're going to be see unused for 23 days. Um, >> oh here's not highlighted. >> You did >> yellow. >> Ah, okay. >> Yeah, there we go. >> There we go. Um, this here should also slide down with it, but it doesn't. Again, we're working on the UI every day to make this better. Cool thing is open source, you can just go make a poll request >> and >> anybody in chat wants to help fix this. >> Yeah, make it so that this box here uh slides down with us because that's sort of annoying. Anyways, here is what it does inspect histories times it was used it doesn't have any dependencies I could archive it and then we can also see archived ones verify multi-room hotel options oh when we were going for vacation >> we had to we were trying yeah I was getting it to do the research um often what I'll do is I'll I'll give it a task like I think of it as I would I would go and do it this way. So, I'll simply get it to um take the steps I would and I will instruct it often in small steps because once I'm done getting it to do that thing like I'll go, okay, do this, then do this. I might give it one or two steps, then I'll say, okay, we'll learn this as a skill if it hasn't already learned it. And now that thing that I did is reproducible like it do it again next week. Let's see. I'm going to create another one while we do this. Let's go um slashdashboard. Um create a dashboard for my Okay. The bills I have. I I've always had a hard time keeping track of all the emails. I want to archive stuff, but then I don't want to archive it. I got to pay that one. That one's autopay, but I still want to review that one. Right. You set up autopay, you get an extra 5% off, but now you're not reviewing the bill. And I'm like, but I need to make sure that I look at at least glance over it because they're taking my money. So, um, okay, this here, um, has done its organizing. We could go deeper into that, but this is just touching the surface. I wanted to with this I wanted to show you guys that it has full access to the computer. >> I have this on a uh its own computer by the way or you can put it in a VM uh virtual machine parallels or something like that. Um and then hey school dashboard that looks better. >> You look good. >> Yeah. >> Okay. Due today Friday $4 soda shop pre-orders. I don't look at this. I don't even know what that is. Um, nobody told me about that. Oh, look at that study reminder email. That's much easier. So, as you can see, it's pretty easy that I can start >> um monkeying around to get this to do what I want it to do or look how I want it to look. >> So, someone just asked a really fun kind of relevant question here. So, if we're having OpenCloud go through all the emails and stuff to check this, um, what if it were to read like an email that tried to give it malicious instructions kind of what is the what are we talking for security here? Now, obviously, whoever is hosting, if you're using a hosted model, then that's kind of partly on them. Is there things built into OpenClaw that can help? there are some things open but a lot of that is is the models um models so when I'm using entropic or open AI models and I go pay my bills it often goes well like I'm I can't do that um you can get it too uh but they have restrictions because they train the models to not do these things and that's not by any means a hard security you I wouldn't use that as hard security. I would um keep an eye on what your model what you are giving your um claw access to. In the case of passwords, let's say I give it access to you can give it access to a vault that have specific passwords for it that you can easily reset and this type of thing. So don't give it access to you know deep um you know your your most important passwords if you don't feel comfortable with that and it largely depends on the model you're using. It's pretty hard to stop um everything but we we haven't had issues with this. Now one thing to note is um when it comes to the security your model if you use a model that just simply obeys then you know to everything it doesn't have judgment then you're more likely to be victim of a prompt injection. That being said when the model reads the email it knows that it's greeting the email. There are it's specifically given instructions that this is it's basically reading this email that these are not instructions. Instructions come from when you type them in the box. >> So there are internal prompts uh and and framing that the model is only supposed to be doing the thing that you're telling it to do. Anything outside of that it isn't authorized to do. So if your model's disobedient to the aggressive prompting, then it's hard to stop it if you've given it the tools. It's sort of like, you know, I my kids uh help me when I'm doing work around the house construction stuff. And when I give them a hammer, I give them instructions. At the end of the day, if they want to swing that into their kneecap, I I can't stop them. I can give them knee pads. I can Right. So, I don't give it to like a one-year-old. Um, >> yeah, >> if that makes any sense. >> No, I I think that does. Um, perhaps another kind of general behavior question. Uh, does OpenClaw continuously monitor and log bugs and fix those? Like I guess maybe the kind of generalization here is what are some of the built-in skills or maybe some that you personally find super interesting or cool? >> There are a whole bunch of built-in skills. Um OpenCloud does not continuously monitor bugs and fix them. Um the idea is that you run a build from um us and then you can improve it through uh making a poll request or jumping into one of our community threads and and giving your input. Now, if you want to just make changes to it, then you can fully do that and and you could have multiple claws set up where one where they talk to each other and um you could set that you could set it up to do that, but out of the box it doesn't do that. Uh we found that when openclaw edits it itself it's like itself itself not another open claw it often conflicts with security that's built in because again now prompt injection if someone has something where they want to get open claw to edit itself to make it not secure so that they can get in that would be bad. So there are some hard guard rails >> and we are working on adding I know we should have a slashfeedback. >> Um >> okay >> we don't have that right now. We we've been reluctant to add any sort of tracking or connection to people once they download it. You know we we don't so we do have some minor telemetry. I believe it's opt in. Um but yeah, we we it the open- source space and the nature of our project is that often our users might be more interested in um anonymity than others. And and we don't want to take that away or make it too difficult for them because we we don't we're not here to make a political statement about what you should and shouldn't do. We're here to give you tools and you do what you do. >> Which does make it all the more important for like open source projects like this that aren't getting that kind of feedback, file bugs, reply comments on things because otherwise like having maintained things myself like you have nothing. It's crickets. Like is it working? Is it you know we just pushed these new features? Is anybody even trying them? Um, you know, lots of big companies have telemetry built in and, you know, dashboards and everybody knows what's happening, but um, open source projects that take this approach don't have any feedback generally or very little. >> Yeah, >> Alexander also had a >> that makes sense. >> Great idea of using this to create a personal Jarvis setup. a number of people have >> here >> definitely seems like search it on GitHub or somewhere and you might find an open source implementation you can build off of >> oh gosh same thing um that is not readable so ugly I talked to my your agent will learn your communication style your preferences as you go through these memory mechanisms You can give it, you can implement your own skills. You can say remember this, always do this and it will add it to um stuff like the agents.mmd file and so on to make sure that it always does that. And again, it will only follow that if the model itself is good at adhering to these things. But the more instructions you give, do this, don't do that, the more it's likely you are to have um instructions that conflict. And some models will stop, some models will just choose one path. Um, someone we've kind of you're showing some examples of this now, but someone was asking how does OpenClaw impact you, Hannis's open day uh day-to-day life. >> H, >> it's helping you parse your email. Are there any other kind of like cool ways or things that you're like, "Oh my gosh, I can do this." >> I actually get it to pay my bills. Um people people don't like that. I have it I I've wrangled the model to go into my bank account and pay the bills. Um my model my my claw I don't have any operations where it's just running in the background unless I tell it to do something. And so when I tell it to do something I'm watching it in the sense that it's over on a screen. I have uh you know four screens and it's over up here um doing its thing. And so I'm keeping an eye on it, but then I don't have to go in. So I'll start by going, "Okay, what are my bills?" And then I'll say, "Okay, yeah, pay those bills or whatever. Pay them on this date." Um, what else do I use it for here? I'm having it file um claims with our uh health care provider to get our you know the or the the orthodontist for my kids braces doesn't auto connect to the um healthcare provider. So I've got to file the receipts every month. And so I'm just getting it to do that. And so what I'm setting up is I'm doing it in this chat over here. And I'll eventually just get it to automatically do it every month. once it does it without um running into any um walls. Uh another thing is I'm having it helping helping me manage my investments. Goes in, looks at performance and then gives me options. I I don't just say manage my investments. I have a conversation. I think a lot of people um want to just tell it sort of like if you had a robot you said like go clean my house like >> what does that even mean um what is a clean house >> that needs specified >> yes and so you've you need to have a conversation and the cool thing is with open clause that that conversation um helps it next time so next time you say go and do this thing it goes and look it's look so if I say order me pizza. It will suggest Pizza Hut and say, "Well, you ordered this last time." >> Mhm. >> So, I don't have to go to Pizza Hut website anymore to order pizza. I just I'm outside with the kids and I open Telegram and send a message or the web UI u the mobile app even and say, "Order me pizza." And it won't just order it. It will ask me what I want. And I'll say, "Uh, you know, for my family and it will go, okay, well, I would suggest, you know, three largest because there's seven of you." um this kind of thing. Somebody said they confirm payments before paying. >> Nick said uh they really liked your kind of uh visual checks like and I think this is really important when you are doing potentially spicy things with agents is to like you were saying part of how you derisk it is you watch it. Um, but you know, if somebody isn't doing that, you know, having kind of built-in checks like this and nice ways to check on those, uh, I think is a really great idea. >> Yeah. I I started with a don't ever make a payment without, you know, explicit confirmation. >> And then I started getting annoyed when I was like, "Okay, pay my bills." And it was like, "Okay." It would stop and be like, "Okay, I have this queued up. Can I submit it?" Like, "Okay." And then next. and like five times for my five main utilities or you know internet whatever and I just kept forgetting to go back to that chat because I was doing stuff. So now I just give it the go-ahhead for all five at the same time. >> Makes sense. Um >> which I'm not recommending. I'm not taking any risk if your AI goes roan. Yeah. Yeah, someone's like, "If I leave them to run autonomously, uh, let's see, I may get older and forget about them running." Kind of like all the people with the PHP servers that they didn't know they had running recently. >> Yes, that you know, most of my the autonomous refreshing a dashboard for the kids email or for that kind of stuff. But I don't know. I like I see my agent as a tool that I use, not as my robot slave. Um I I don't think we're there where we just go look, take care of my personal affairs. I'm going to live my life. I don't know if we'll ever be there. I have, you know, that's a long discussion but >> it's a philosophical uh truth, >> right? But I know that it's really easy. I I don't like going through my emails. Uh I can now actually tell that dashboard to archive the emails after it adds them, >> which cleans up my inbox and now I can see what's left over. Um and I can finally get my inbox to zero kind of thing, >> which is something that I never have the discipline to do. And so start by like routing these things into well I have my bills, I have my, you know, spam, I have my and just actually getting it taken care of. I'm always like, well, I need to and archive that because I need to like deal with it and I never deal with it. So yeah, this is this is a mech suit for you. This is not your robots robot slave. Not saying you can't experiment with getting it to do things totally autonomously. I just um I don't know at your own risk, I guess. >> So, as we're kind of getting closer to time, I wanted to ask you kind of some more meta questions, which were uh to start with, how did you get into working on OpenClaw? Like we we talked a little bit about your interesting path there, but uh >> well, I um here one second. I'm just going to do one more thing here. I'm going to say uh make me a Flappy Birds dashboard. Play Flappy Birds. Now it knows it's not making one about it's it's a game. I can play it. Okay. How did I get into it? Enable it. I really need to figure out why this is make the dashboard. Okay. Um I'll troubleshoot that after. This this is an old agent so it has like upgrade, you know, we we change things, we fix things. I have workarounds. So sometimes starting from scratch helps. Okay. How did I get into this? Um I was building software for my small business. I have a small renovations and handyman business. Um, five employees typically and I was building a tool that made it we discovered that we could um get leads no problem, get 40 leads in a day. But how do you talk to 40 people and document it all and then actually the ones that ended up using us get that information and move it to the employees that have to go do the work? Someone wants their pictures hung, right? Someone wants their baseboards fixed or the cocking ground, their bathtub. How do we make sure that their specific concerns about, you know, making sure the door is closed so the cat doesn't get let out? How do we make sure that gets transferred all the way along? And I would be writing this stuff down and then putting it into a sauna. Um, probably the only handyman on earth that uses a sauna. Um, and I didn't want to do that anymore. So, I thought, okay, what if I just like typed the notes and had them automatically entered? And that whole journey, what about when people text me? What if I just ingest the text messages? Well, what about when I want to make a bill? So, all these things led me to using um uh trying to build software for myself. And because of this, I ended up at uh and and I do have a background in in software development. I've I've tinkered with it for I'm 42. I started learning PHP when I was 14 or 16, something like that. And um so I was building that software and I ended up working with Rue Code. I worked at Root Code and then we shut down our um coding agent and went with uh you know in-house something called remote. It's another open source project. I don't get into that. And then I ended up working at OpenClaw. This is a tool that I've been using for six months before I started over six months before I started at OpenClaw. And so I thought, that's, you know, let me talk to them. And I reached out and here I am. It's sort of a backwards p. I was just trying to fix people's stuff and that's how I got here. I mean it sounds like it's very on brand. You are trying to address anything from small to very large uh annoyances and problems. >> Well, and you know what's cool is the things that I was doing with my small business, I'm still doing them. I actually have a claw that is if if I get a text message from a customer um somebody we they go to our website, we have a Google ad running and they message us and go, "Hey, um the cocking around my bathtub needs to be replaced." My claw will say, "Hey, my name is Hannis Rudolph um with Rudolph Enterprises. What's your name?" And then it'll tell me and I'll say, "Could you send me some pictures of what you're working with?" And the re it's not AI slop response because this is exactly what I typed hundreds of times. the exact I was copy and pasting it. So I'm like, well, why don't I just get the agent to do this and if it falls outside of this path where I ask them for their name, I ask for photos, I give them pricing, and then I pass them off to uh my business partner, then it flags it and I can jump in. And so this is coming right to my iPhone, my iMessage, and it automatically updates the step, all the progress right in uh Asana, our CRM in this case. That's what we're using it for. So, I'm still doing the thing that I started. I'm doing both things and it really doesn't take very much work to run the handyman thing anymore. So, another an aside of how far you can get if you tinker hard enough. >> So, yeah. What we got on the dashboard here? >> Here. Let's Okay, there. This is all in open climb. I'm using it right. And we could say, let's change that to a lobster or something. Let's add sounds. That kind of thing. I don't know how I'm supposed to pause when I'm clicking. Ah, got it. So, anyways, any more questions? Any more questions from the people here? >> Um, >> or you? >> See, I'm looking through chat here. Um, while I do that, what was the last thing that you pushed or like kind of your most recent contribution thing? Uh, you know, which obviously we all know that doesn't need to be code, but kind of just a snapshot. >> I believe it is when I'm using the mobile web UI and I hit the enter button. I don't want I want it to to do a line break. I don't want it to submit. >> One because I always uh hit the enter button. I have large hands and I hit it by accident. Another one is because I often give it multi-line instructions and I need a way to give another line. And so on the mobile web web UI then you actually have to hit the send button. And this is similar. I looked at how some other apps like Jet GPT do it and that's their default as well. So that was >> uh the changes that I make are usually um as I'm doing Devril a lot of the time most the time are usually like oh this annoys me and they're topical enough that I can make them without um you know having to deep dive too hard. So yeah, that and I think uh the one before that was automatically the browser whenever my my claw was browsing the internet it would pop up in the sidebar and I just didn't like that. I'd rather that it just sort of showed it in line, you know, in the in the chat. >> Yeah. Um, so perhaps uh most interestingly for a lot of folks who have kind of thrown out great ideas in chat for how they want to get started using this or trying stuff out, how would somebody get started contributing to OpenClaw? >> An excellent way um is to go to community.openclaw.ai. This is a discourse page where we have um lots of volunteers and maintainers that are looking to help and give you input, answer any questions. And also if you have a stomach for mayhem, then Discord, you can go to our it's it's also you can find our Discord page through that community.openclaw.ai. But one way is to get in Discord. We have a channel called claw tributors where you can jump in and ask some questions. There's always a maintainer um in there. Well, should be there is always one as far as uh I see uh we have people working around the clock to keep that manned and yeah we can answer questions about uh the specific improvement you want to make and if you are looking for support discourse uh through community atopenclaw.ai AI. But our Discord has a support channel that has a claw in it that will answer your questions and work help you work through things. His name is Krill. >> I I do really like the lobster theme. Uh because I learned recently the main mascot is Molt. Is that right? >> Multi. >> Multi. It's a great name. >> Yeah. Yeah. Every day, Multi Multi actually makes music in our maintainers channel. Um, AI slot music about the work we're doing and it's it's they're usually three or four minute silly songs. It's >> Multi has quite a personality. It's it's Peter, our uh our founders's personal agent, his personal claw. >> Makes sense. Um, someone was asking, uh, what is your like top most favorite like feature setting? Something that you'd be like this, this is it. >> Uh, there's just so much. I mean, I my example I always give for VS Code because I'm a VS Code user is VS Code pets because I love having uh tons of including crabs and lobsters hang around with me while I'm coding. >> Well, I mean, we we have our own little pets. We have uh the built-in Where is it? Appearance. There's the lobster decks. >> Builtin pets. >> Well, there's the lobster decks. So, these are all ones that have appeared. They appear down here and they jump up and down. >> Um, yeah, different different lobsters um that don't really look like lobsters. They're called blobsters, but that isn't my favorite feature. I probably the my favorite feature would be uh browser use. And I know that seems sort of lame, but it makes it so that when I want to order on Amazon, when I want to do anything, it just when I want to send a Discord message, it just goes and does it not through like breaking the terms of service in the sense of like making API calls through the back end using my personal token that it's not supposed to do. It just goes and like me just goes and clicks and goes into that message and types the thing that I said and I just find that super super useful because everything that I do on a computer is basically accessed through the web these days. >> Yeah. Or an app on the computer. It can also access that. I want to change how my settings I want to do this. But the the way it uses computer use on a box that isn't mine cuz I have a hard time using um chat GPTs um and and claw desktop and stuff when it's on my computer. It's I'm trying to use my computer. Don't get in my way. Right. Like I guess if I want to type something and then walk away, but I don't I'm not doing that. So yeah, >> I I don't think that's lame. Uh it's definitely something I've gotten a lot of >> things that you know maybe should have MCP server. Like why why is there not like a great LinkedIn API? >> You know what? >> Like it feels like that would be something that would be very useful. But short of that, I also have my agents click around and interact with LinkedIn for me. So, >> yeah, it's definitely uh >> I think there would be even more spam. Um, >> well, there is that. >> Okay. Oh, hey, side note at Rabbit Trail. Recently, LinkedIn, I think, added this uh when you hit the context menu on a post, there's a there's an option that you can select that says seems like AI slop. M >> so maybe this will help clean up LinkedIn because I like LinkedIn but I get sort of I don't know demoralized when I go through there. I'm like seriously I that's slopp like I want to read what people actually have to say. I don't So yeah I think we need to do more of that. And on the screen here I'm showing you one of our blobsters in the lobster decks is Leonardo Dainci. >> That's great I think. >> Is there one of you? No, I put seven in here recently. Um, okay. >> My wife my wife's ideas uh this my daughter came up with this one. You can't see its name yet, but this is Taylor Pinch. >> Um, you can see and this is uh Claw 2D2. >> Solid. So, and we have Shelvis and um yeah, >> I I think it's a great community building thing to have like a good mascot andor kind of theming to your project that people can kind of get excited about. >> Well, and these guys here as uh as they visit today, >> right, you you get to they visit, they pop up here, then you get to see have access to them. See, there's one. See, and if you click him enough, he gets angry at you. He's He's frowning at me. >> I think we are getting pretty close to time. Um, I think I've gotten a lot of the questions. There's plenty more. And if you have more questions for Hannis and the rest of the OpenClaw team, you can definitely check out um the community forums that we were talking about earlier. Uh you can check out the actual GitHub repo for the project openclawclaw. Pretty pretty easy that one. Um but yeah, thanks everybody for joining and hanging out with us. Uh happy Friday, happy weekend. >> One one more thing, >> Saturday morning. >> If anybody sends me a Discord message, I'm my user ID is H Rudolfph. You can go on to our find our Discord page through um any of these links she's sharing. um then I will respond to them. I respond to every single DM people send. So you can ask me direct or you can just hop on and ask in general, but we're always available. >> Seems like a very y'all seem pretty nice. Like I wouldn't feel very uh intimidated by dropping a message in your Discord. >> I don't know. Some some projects seem like really formal and like I get the purpose for having a bunch of structure like that, but like one of the things I one of the reasons I really like doing open source Friday stuff is because I get to see a human that is tied to this project. I know that there's actually somebody there I am talking with, not just a bot. >> Yes. >> Yeah. >> Human aspect of open source is really important to me. >> Well, we understand that. Um, people are often contributing outside of their regular hours and they're not always pros in this area and so following a bunch of strict rules after they get off work is not what inspires them. What inspires them is an idea they had while they were in the shower or driving home or working their 9 to5. And we tap into that by making it easy and not saying, "Oh, well, you shouldn't, you know, you didn't check that box off." Um, we don't want to be those guys because these big companies who are utilizing the ideas that we help foster, they could pay an engineer a million dollars a year that couldn't come up with the idea that you as a user came up with while you were driving home from work. So, we don't want to say no to those ideas. So, please bring them to us. Bring us your crazy ideas. We'll try to shoot them down as gently as possible and hopefully not shoot them down at all. help you figure out how to implement it. >> Yes, we don't like to shoot people down. I was just joking about that part. >> So, last question for you. Yes. >> What is your favorite ice cream flavor? >> Vanilla. That's I know that's super lame but >> there are so many options. >> That That's the problem is it's so confusing and then they're not all available. Vanilla is always available. So functionally it I can always get my favorite ice cream. >> I guess thinking about it like that then you're never you're never disappointed. >> It's a math equation for me. I'm a little square. All I wear is black shirts. I I have like 16 of these. So >> makes sense. Well, thanks everybody again for joining today. Uh thanks Hannis for coming and sharing about your project kind of uh >> how y'all are working with the open source ecosystem. This is a really amazing project. Super nice people. If you are looking to get into open source, this is a great place to start both tool-wise and contributing wise. >> Thank you for all the kind words and thank you for having me today. >> Yeah. Well, uh next week is GitHub Universe. Um >> nice. I forget if we have a show next week, but if not, I'll catch you on the next one. >> Bye everybody. >> Have a good weekend. Heat. Heat.

Assistir ao vídeo original no YouTube

Assuntos: Assistente pessoal, Assistente pessoal. Publicado no YouTube em 2026-10-24.

Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.