OpenClaw 3.8 Update: What's New (Release Notes Breakdown)

0:00And yet another OpenClaw 3.8 update. I just upgraded today this morning and I'm already using three of the new features in production. This is not a demo video. It's not any kind of tutorial. It's just me telling you how I actually run OpenClaw every single day. Real agents doing real work. So, I'm just going to get right into this. I don't want to believe any fluff. It's just get right into the meat and potatoes. I have 10 cron jobs running right now through OpenClaw. Atlas, the chief of staff agent, handles my morning briefing things and evening check-ins. The Rizza runs every morning at 6:00 a.m., scrapes Reddit, Hacker News, and X. It does use a Brave Search API for the trending AI topics, and then it generates all of the content related. So, we got Inspected

0:45Deck who creates the daily Twitter posts and handles more of the podcast outreach. I also have newsletter agents that help, you know, send out the daily blog, get my newsletter ready so I could proofread it. I also have a Skill Scout that auto installs new capabilities. So, this is my entire kind of content and business engine that I have automated that I used to spend at several hours a day doing, and now it's running through OpenClaw right behind me locally 24/7. So, when 3.8 just dropped, it was overwhelming because we just saw a drop a couple days ago, 3.7, which shipped some amazing features. But, I want to talk about why 3.8 is a significant upgrade and why you should waste no time getting it installed. So, I didn't just

1:30read the release notes and move on. I upgraded immediately this morning because I saw a couple features that were relevant to what we have going on in our OpenClaw setup, and it directly improves the whole system. So, let me walk you through the big ones and I'll show you exactly how I'm using them right now. The biggest feature for me is the new Brave Search LLM. It's like an uh it uses context mode. So, before this update, the agents would just do the web search, and then they would go back and send raw links and text snippets, which was useful, but it did limit the search. So, now with the LLM context mode enabled, Brave Search actually extracts and processes the full page context and returns it with proper grounding data. I enabled this immediately because the

2:16content pipeline agent, The Rizza, scrapes the web every morning looking for trending AI topics. Better search results means better video scripts, better content creation, better copywriting, and I can already see the difference. Instead of getting fragments like OpenAI announces new, followed by a cut-off snippet, I'm getting full context of what actually happened. So, why it matters and how it connects to other developments and in our project specifically as well. So, this matters because The Rizza doesn't just find topics, it has to understand them well enough to write the scripts, write the content, create the outreach letters in my voice. So, when you're feeding an agent better context from web searches, you just get better outputs. It's that simple. Have you ever tried prompting an

3:01LLM? And if it's a terrible prompt, you're going to get a terrible output, which I would consider, if you're not already reverse prompting. So, ask the LLM to generate the prompt that you could feed back to the LLM because whatever you're using, Claude or GPT, it's going to produce a much better prompt than you ever could at this point. So, this new update gives us a much better, more robust copywriting, better information, better input, so the outputs can be solid. So, the old way would miss like the nuances or connections between different sources, it would constantly mess up. The new way gives my agent the full picture to work with. And here's what's really cool about this is the grounding data means that I can trace back exactly where

3:46specific information came from. So, if The Rizza writes something about a new AI model that just dropped and its capabilities. I could see it pulled it from that actual research paper, not just some blog post speculation. That's huge for accuracy. I tested this yesterday with a couple of searches about Apple's AI plans and the old system returned like eight different article titles and short snippets. So, the new LLM context mode gave me a comprehensive summary of what Apple's actually working on, pulled from multiple sources with clear attribution. Night and day difference. If you're running any kind of research or content agents, this feature alone makes the 3.8 upgrade worth it. The quality of information your agents can work with just got

4:32significantly better and it's amazing to me because just two days ago I made a video talking about how 3.7 was one of the biggest upgrades I have seen and 3.8 is a sneaky one. They just dropped this morning and I'm already seeing the benefits. So, the second feature that I implemented immediately is the new backup system. Open Claw now has a built-in CLI commands for backup and verification. So, I set up like a weekly cron job that runs every Sunday at 3:00 a.m. and auto backs up the entire Open Claw configuration. This matters more than you might think. My Open Claw JSON file contains 10 different cron job configurations and all of my Telegram bot settings, API keys for Claw, Brave

5:17Search, Nano Banana plugin configurations, the custom skills we've even built. Losing that file would hurt a bit since we've put in hundreds of hours getting this all built out. I'd have to rebuild this entire agent infrastructure from scratch. Yikes. So, I created a simple bash script that runs Open Claw backup create and with a timestamp and it stores it in the cloud storage and sends me a Telegram notification when it's all done. Takes about 30 seconds. Now I sleep better knowing that I can restore everything if something goes wrong and it will go wrong. Just know that. If you haven't got Open Claw set up before, you know the frustrations of getting it set up on how to tailor it to your specific needs. Things will break. Welcome to the Open Claw life. I haven't actually tested the backup restore process yet,

6:02but the backup creation is working perfectly. The CLI tells you exactly what got backed up and verifies the file integrity. Super clean implementation. What's funny is I was actually manually exporting my config every few weeks before this. Now it just happens automatically every Sunday. So, one of those features you don't realize you need until you actually have it. Could you imagine losing everything that you already started building? That's a nightmare. So, I'm going to talk about the Telegram fixes because these were the real bugs I was hitting every day. There were two main issues that we were facing. First, the DM deduplication was broken and agents would send me duplicate replies. Annoying. Second, there was a cron delivery bug where cron jobs were report as delivered, but the messages never actually hit my Telegram

6:50or they would hit the Telegram on my PC and not on my phone and vice versa. Forget context. Super annoying. I experienced both of these first hand. The duplicate replies were getting to me, but they weren't critical. The new cron delivery bug was actually painful because I rely on my morning and evening briefings from Atlas just so I can know what's up to date and what's going on in Mission Control. So, I'd wake up expecting the daily briefing and nothing. The log showed it delivered successfully, but my Telegram was empty. I literally had to manually fix the delivery targets on my morning and evening briefing cron jobs. Yesterday, I went through it and had to fix it all because of this bug. So, I had to go into the cron configuration, update the Telegram

7:36target settings, test the delivery, make sure it was actually reaching my phone. Took about 20 minutes to fix both agents. no big deal. But with the 3.8 drop this morning, both bugs are resolved. My morning briefing came in through easily and perfectly this morning, and the evening briefing worked last night. Clean delivery, no duplicates, and reliable notifications. When you're running a business through AI agents, reliability matters. When you're running When you're running a business through AI agents, reliability matters more than speed, in my opinion. These weren't just showstopper bugs, but they were actually death by a thousand cuts type issues. Not major, but annoying. There's also a new talk mode silence timeout feature, which I haven't used, but it is nice to have. This lets

8:23you configure how long open claw waits for silence before auto sending a voice conversation. So, I'm not just using talk mode yet, but this is useful if you do voice interaction with your agents. This is something I could see myself doing in the future since I use whisper flow throughout my entire workflow. You can't set it to wait You can set it to wait 2 seconds, 5 seconds, whatever works for your speaking pace. I might experiment with this for like the Atlas interactions that has Opus 4.6 on the back end. So, you know, it might be nice to have as I've been doing that with Claude for some time in some of my Claude projects here on the PC. So, sometimes I have like a good back and forth, and you know, it's something that I would want to consider moving over to

9:09Mission Control. And you know, I would talk about strategy, you know, the priorities, maybe new ideas that I have, and that just is so much more helpful than just typing everything. But for now, my workflow is all text-based through Telegram and Mission Control dashboard. I say I spend probably 70% of the time communicating with. When I use the Telegram, I'm That's when I'm like out and about, you know, I could be in town, and you know, like this morning, the update came out, and I just said, "Hey, 3.8 update is out." I copied um over the repository. It scanned it and then I asked it how it would be relevant for our use case and it instantly um you know, implemented the Brave Search LLM and upgraded that and uh automatically scheduled that cron job for the backup

9:55to happen every Sunday. So, you know, things like that, but Telegram isn't my main way I communicate with OpenClaw. It's definitely just through the local dashboard here. There's also an ACP providence with session trace IDs for multi-agent setups. This is basically tracking and logging for when you have multiple agents working together. I'm not using this yet, but it's relevant as my agent crew grows. Right now my agents mostly work independently. There is a generate scripts, inspected deck handles like the social post, the newsletter, the blog posts. Atlas does the morning and evening briefings. They don't really collaborate on tasks, but I am building towards a more complex workflow where the agents might hand off work to each other. When that happens, being able to trace which agent did what and when will

10:41be valuable for debugging and future optimization. The other fixes are more technical, but worth mentioning. The 2E light theme auto detection, which I don't use since I'm always using dark mode, but the browser relay WSL2 support for Windows running Linux subsystems, configuration secrets persistence fix, which probably explains why I occasionally had to re-enter API keys after updates. Super annoying, but this did happen several times a couple weeks ago and I thought I was losing my mind. None of these None of these affect my daily usage, but they do make OpenClaw more stable and more accessible for different setups. I think everyone who's running OpenClaw should not just copy everyone else's setup. Tailor it to what you're actually

11:26trying to automate. Focus on one task first, automate that well before you move on to the next thing. It's like it's so tempting to like say, "I wanted to automate my my email, my Google Calendar. I wanted to write scripts. I wanted to do content." It's like one step at a time because if you try to get all this going at once, you're going to run into a lot of issues and a lot of things failing and you might just give up because it's super annoying and I know you're probably have used Claude before or Claude code and you're like, "I could have just done this in Claude code in under 10 minutes." But I think if you spend the time investing in to getting build out building out an automation within Open Claw, um it is worth it because that's something that now you don't have to do anymore. So, the persistence fix especially is one of those things that you only notice when

12:12it's broken. Here's what's running Open Claw in production actually looks like. That's why I'm explaining in all of my videos how it relates to me with each of these updates. So, like for example, every morning I wake up to a fresh uh daily briefing, video scripts, updates, trending news that happened on X, uh Reddit, Hacker News. I got all the social media posts ready and generated and a detailed briefing. So, what my agents actually accomplished yesterday and what's planned for today is nice to have at least a foundation of where to start, a bit of information and doesn't mean I'm going to choose to run with everything, but it does mean that I'm ahead of the [clears throat] game by having all this and and not having to manually do it myself. So, and then not

12:57having to generate that manually is a time saver for me and I'm all about saving time. So, my entire YouTube content pipeline runs through Open Claw. We're connected to the back-end APIs. It's checking click-through rates, which thumbnails performed, where people dropped off. It's pretty incredible with now over 500 subscribers in the last 5 days. Welcome to the channel, guys. We're going to be dropping updates every single day, three to five videos. So, if you like AI news, I'd recommend subscribing because I'm going to be dropping these videos every couple hours. So, the really cool thing is every thumbnail we run through Nano Banana 2 on kai.ai. It's like 8 cents per generation. So, we generate three thumbnails per video and it does the

13:44testing over time on what's better, what's worse, and kind of optimizes the content. Pretty cool. So, some people pay, you know, content engineers, content marketers for this information, thousands of dollars a month and this is something we just built out for less than $50. So, very cool. The agents are mainly using Claude. We use Sonnet 45 46 and Opus 46 for a lot of the coordination and copywriting. And then the smaller cron jobs are ran on Gemini through Open Router just because the token usage is basically nothing compared to like an Opus. So, the Brave search is really cool. Nano Banana's integrated for all like logo design, image creation. You could use something like Firecrawl API if you

14:30don't want to use Brave. So, the agents get posts ready for Twitter. I review them. I would say 90% of them I don't use, but the 10% that I do use are actually really well written in my voice. So, they also handle the newsletter content and even reach out to podcast hosts for guest appearances, people that are building in public specifically on Twitter. And now with the 3.8 update, it's more reliable, more capable, better search results, automated backups, a cleaner Telegram delivery. These aren't flashy features, but they do make the system I depend on every day work a bit better. Like I said, over 270,000 stars on GitHub and over 100 developers working on this project daily. Pretty

15:16incredible. The thing that people don't really understand about Open Claw is it's not just a cool AI tool that you plug and play and everything is working. I made a billion dollars, yay. And I never want to come off portraying that to you. It's takes probably over 100 hours to tailor it to what you're trying to do, how is it going to serve you, how is it going to buy back your time. And if you're trying to use it to make money, that's going to take time. So, don't believe the hype that you could just get this installed on day one and start making thousands of dollars. You need to figure out how it's going to help you save time, help automate some of those redundant tasks, and how it's going to help you grow your business. It's infrastructure. So, when you're running a business through AI agents, reliability and consistency matter more than anything than fancy features. You

16:02need to know more of You need to know your morning briefing will be there when you wake up. You need to know your content pipeline won't break because of the search API hiccup. That's what 3.8 delivers, more stability, better data quality, proper backup systems. It's the kind of update that makes you confident to build bigger and more complex agent workflows. If you want to actually learn how to build a simple workflow that can help you buy back your time, you're going to want to join us in Shipping School. I'll put a link to it down below. Three bootcamps a week, as well as a full content course on how to get Open Claw up and running. So, that's where I walk you through the building process, Q&A, maybe we can roast your app, help you get your startup off the ground. Our goal is to help you go from zero to one

16:48as fast as possible. is why we also have a Next.js starter kit that you could plug and play that has authentication, database, and payments all built in to get you going. So, I really just want to show you the tools that I use daily and how I'm using them to grow my businesses. So, hands-on help is important, but also holding you accountable to what you're trying to do. It's not just a community where we talk about stuff. We're actually holding you accountable to actually ship features. So, that's what you get AI agents, ideas shipped and running all down below linked. You'll see it. This isn't just about watching tutorials and, you know, those are great, but I think the people that actually do instead of think, those are the ones that are going to be, you know, breaking things and shipping ugly

17:33landing pages, but that's where the iteration process comes and that's what it's all about. Building together, troubleshooting the real issues, and creating systems that actually work in production. So, that's what I talked about in this video and those are the things that I'm doing. So, check it out link down below. I'll help you go from an idea to a deployed AI agent crew or app. We'll see you in the next one, guys. Be blessed.