OpenClaw 3.7 Update: What's New (Release Notes Breakdown)

0:00So, Open Claw 3.7 just dropped, and there are some massive updates that you need to know about. These aren't just some incremental upgrades. There are some huge things happening inside of this update that I'm going to be going over in this video. So, they don't just fix a few bugs and call it a day. When they release an update, there is usually some substantial security fix or a new model that you could use within Open Claw. So, we're talking about features that fundamentally change how you build and deploy AI agents in production. So, I run five agents on my Mac Mini every single day. Atlas is the chief of staff who manages everything and kind of corresponds and who I chat with inside of Mission Control. The Rizza scrapes and writes the YouTube scripts, does a lot of the content, the email marketing.

0:45Inspector Deck generates all my Twitter posts. And I've got agents for thumbnails and newsletters and blog posts. So, these agents run 24/7 behind me all the time locally on a Mac Mini, and they're running my entire content engine. So, when Open gets a major update like this, it directly affects the daily operations. So, I'm going to break it all down what's new and why you should care. This is, if you're trying to build something, you're trying to get a startup off the ground, or you're just trying to automate to buy back some of your time, you're going to want to learn what is going on. So, GPT 5.4 was probably the biggest thing to integrate into Open Claw. Otherwise, you kind of had to go and do a workaround. So, now 5.4 is available. Gemini 3.1 Flash is

1:30also available. Another cool thing is Gemini 3.1, which a lot of people don't talk about, not really a big news release. It is fast. So, 3.1 Flash Light is incredibly fast. So, the responses come back in milliseconds. So, when you're running agents that are constantly processing information and making decisions, speed matters a lot. We use 2.5 Flash for a lot of the cron through open router. So, these cost These cost like 90% less in the token usage compared to like a Sonnet 4.6. So, this does matter a lot. The thing is about running AI agents in production is that every second counts. When the Rizza is scraping trends at 6:00 a.m. and

2:15needs to have scripts ready for me when I wake up, it's all set and ready. I can't have it taking 5 minutes to process each trend. So, with Gemini flashlight support, those same operations that used to take minutes now happen in seconds. And here's what's really cool about it. Open Claw's model support, you can mix in and match models with the same agent conversation. So, you might use GPT 5.4 for complex reasoning tasks and then switch to Gemini flashlight for quick data processing, similar to my setup and how I have it all layered in. I use Opus 4.6 as the brain as the coordination agent Atlas, so I can have, you know, really good conversations that are more

3:01personable. And then other agents through open router are, you know, we have Codex for a lot of the coding. We use Opus 4.6 for the script writing and Sonnet handles out a lot of the smaller cron jobs. And then the the super small, you know, security audits and security updates are ran by Gemini. So, you can layer it kind of how you want to. So, the agent handles the switching seamlessly based on like what task it's working on. So, let's talk about the pluggable context engines. This is a huge release and I mean absolutely massive for anyone who is serious about actually building with Open Claw and AI agents. So, traditionally when you build with an AI agent, you're stuck with

3:46whatever memory system the platform gives you. Maybe it stores conversations in a simple database like a memory markdown file, or maybe it has some kind of like basic rag retrieval, but you really can't customize how the agent handles memory or context. Open Claw just changed that completely. With pluggable context engines, third-party developers can build completely custom memory systems. So, I use super memory.ai to help as like a third backup to pushing all our memory and all our conversations to the cloud every couple hours, so it could retrieve that information at any time in addition to keeping our memory markdown file below 2,000 20,000 lines, because I noticed

4:31that it will compact over time and lose context of what we're accomplishing, the mission, the skills that we're learning, and that can be really annoying. So, you want your agent to store memories in a knowledge graph. Build a context engine for it. You want vector embeddings with custom similar scoring. So, you could build a context engine. You want memories that decay over time like human memory? You could build that, too. This opens up so many possibilities, and when I first got going with Open Claw, that was the most frustrating thing was the persistent memory. It would just constantly forget what I was doing, who I was, and it was annoying because I'd spend hours trying to get this thing set up and then forget what we were working

5:18on. So, it does take time to about it, and so be patient. So, it does open up a lot of possibilities with this new update 3.7. So, imagine like an agent that remembers not just like what I told you, but how you said it, when you said it, what your mood was, what projects you were working on at the time. Context engines make that possible, and I'm excited to see what comes out of this newest release. So, for my agents, the game is changing. Right now, Atlas is keeping track of all the projects, the deadlines, the goals, all that stuff. But, with a custom context engine, I could have it understand the relationships between different projects, how my YouTube content connects to Shipping School, how both of those connect to my SaaS products. The

6:05agent will have a much richer understanding of how everything fits together. So, I have multiple businesses, and I was hesitant to running all of that inside mission control. But, now that's actually more possible than ever because I'm trying to do different things. You know, I run a reselling business that resells clothes on a live selling platform. We run a a couple SaaS projects. You know, we do the Shipping School. We have the newsletter. So, I'm trying to kind of like keep these separate and not, you know, overload the mission control. But, with all these new updates now, I could plug all this information in and is context aware of the goals and the deadlines per each business. So, pretty crazy, and that's because everything is

6:50like plug and play. So, the community can build them. You don't have to wait for Open Claw to add exact memory system if you want. There is builders. If you go into the Discord on Open Claw, it is wild in there. And there's a reason why this repo has 275,000 stars and over 100 developers working on it. These updates are coming out fast, and it seems like it's been harder and harder to keep up and keep track with what's going on. So, basically when someone has a problem and it gets pinged enough to like Peter, you know, it's going to get fixed. So, all the things that we're optimizing for with people using Open Claw, it's getting shipped in faster than ever. So, the community is building. You know, if someone wants to

7:35build a context engine that fits your use case, you can just plug it in. So, here's the feature that solved one of my biggest painpoints is ACP bindings that survive restarts. So, ACP stands for agent communication protocol. It's how agents talk to external services like Discord, Telegram, Slack, whatever you're trying to do with communication. I use Telegram. So, before this update, if I restarted my Mac or updated my OpenClaw, all those connections would break. Super frustrating. My Discord agent would lose its thread. My Telegram bot would stop responding. I'd have to manually reconnect everything. That was honestly one of the most frustrating parts of running agents in production. You know how it is. You're running everything smooth for days or weeks.

8:20Then you need to restart for some reason and suddenly you broke half the automation. And to this day, things still break, but not as often as they used to. So, with ACP bindings that survive restarts, those connections persist. My Discord agent now remembers which channels it was monitoring and they have a new Discord agent that connects with Discord. It scrapes a lot of data. It will message you much more efficient. It will ping you at an actual right time when you set the time to be pinged. So, all of these updates that come out, this last one I think is the most significant. So, my Telegram bot knows which conversations it was having. Everything just picks up where it left off rather than having to work on fixing all the things that break

9:06overnight. So, I tested this last night right before I went to bed and things seemed to be working a lot more smoother. Though, I only had about an hour with it before I passed out. So, I had conversations going with the agents through Telegram. Then I actually force restarted the Mac Mini for the first time in like 3 days. I came back and it picked up exactly where the conversations were. So, the agents didn't miss a beat. This is what I mean when I talk about production ready AI. It's just not about having smart agents anymore. It's about having reliable agents that work when you're not babysitting them. And I believe that's like one of the biggest problems with open claw when people are setting it up. They find that they're spending more of their time trying to fix the errors when they're like, I could have

9:51just done this in cloud code and be done by now. I do understand that, but once you have it built out and it's working 90% to 95% of the way there, you will find yourself being able to work on other features that you want to ship and other things that now that that other task has been, you know, somewhat automated. So, I'm going to tell you a little bit more about my actual setup and how I'm changing things and how it kind of affects the daily workflow. So, the Mac mini it runs everything 24/7. Sitting right behind the desk here, it's basically the brain of the entire content operation. Every morning at 6:00 a.m. a cron job kicks off. The Rizza agent wakes up and starts scraping trends from Twitter, Reddit, YouTube, everywhere all through the API. And we use Brave Search API,

10:36which is super helpful for searching the web. So, all this happens every morning and analyze what's happening in the AI space, like what people are talking about, what questions they're asking, and then it writes scripts based on those trends as well as like the blog and the newsletter. But, we have a special hard rule that humanizes the content, purposely makes mistakes in the copy, and no AI slop allowed. This is where people who use AI and the people who, you know, are reading AI content, this is the difference. You know, more and more AI content's being produced and it's going to it it if not already, it is outcompeting organic content. You won't see that on this channel. I just want to be real with you

11:21and have an actual conversation. Like I am real. I am a real person. I'm standing here talking to you, but I've seen so many of these AI channels that are, you know, and I I ain't got to I won't say anything. It's just good for them, but I think being authentic and real with you is going to be like the new value and and that's what I want to bring to this channel is having like an actual conversation with you while talking through my AI journey and AI updates. So, you got to be really good with the copy and in the the content you are producing with AI, it's not automatically approved and posted by me. It's getting me 90% of the way there. I'm proofreading everything. I'm adding in some extra thoughts or concepts or key points that I'm learning in addition

12:08to what's being written. Though, I do think, you know, by the end of the year we're much better. It's still you got to you know, you got to finesse it a little bit to make it how you want to. So, and then like by the time I wake up at 7:30, I have all the updates of what's going on. You know, I've got three to five video scripts ready in the Google Drive, all research, all written in my voice, all ready to go. The agent knows my style because it's been trained on hundreds of YouTube videos from other YouTube channels on just my talking style. You know, I say so a lot and I say be blessed at the end of each video and just comment things that I say. So, Atlas, the chief of staff, which is ran by Opus 46, is constantly monitoring my calendar, my task list, my project deadlines. If something big is coming

12:54up, it reminds me. So, I have a meeting with two huge suppliers next week for our resell business. And it already researched those people specifically and it talked about where they started, how they got going, numbers that they did this year, and now I have an outline right here that I study each night because these people are very busy people and I read this every night, so it helps me prepare for the Zoom call next week with two of these people. And it did that on its own. So, it's pretty insane. That's just an example. So, Atlas is kicking butt. It does a lot of the coordination, takes action. Inspected Deck handles all of the Twitter content. It takes like my YouTube videos, breaks them down into key insights, creates

13:39Twitter threads. It knows my posting schedule, my audience, what kind of content performs well. I barely think about Twitter anymore because Inspected Deck handles it all. And if you're wondering, why do they sound like Wu-Tang characters? Well, I love the Wu-Tang and I just rolled with it from day one. And we got the GZA, we got Inspected Deck, we got Ghostface. We're adding more people to the team. So, that's why you're probably wondering. So, with this new Open Claw update, all of this gets more reliable and more powerful. The context engine means my agents can have much richer memory about what I'm actually working on. The restart survival means that I don't lose connections when I need to update my system. The new model support means everything runs faster and costs less. So, if you're rocking with 5.4 or you're

14:24rocking with 3.1 Gemini Flash, you're just going to get these upgrades every time Open Claw runs an update. If there's a new model released, they're going to be integrating it. So, let's talk about some of the technical improvements, too, because there are some really solid stuff that are kind of getting swept under the rug. So, the slim Docker multi-stage builds are a big deal if you're deploying via VPS. So, Docker images for AI applications tend to be massive. We're talking gigabytes, sometimes. That makes deployment slow and expensive. With multi-stage builds, Open Claw can create much smaller production images. It builds everything in one stage, then it copies only what's needed to run the application into the final image. So, I don't run Docker, but people who do deploy to VPS, this can be

15:10a great option to keep costs down. So, you might go from like a 4 GB image to a 500 MB image. So, that's faster to deploy, cheaper to store, and uses less bandwidth. So, I'm not Docker. I I used it in the past. It can sometimes be a headache to set up, but if you've been thinking about deploying some agents to the cloud, uh that could be helpful for you. So, with some of these smaller images, that becomes much more practical for people that don't want or don't have a backup computer, and they don't want to buy a Mac mini. So, here's another one. The secret ref system for gateway authentication is another solid improvement. Before this, you had to manage API tokens and secrets manually. Easy to mess up, easy to accidentally

15:55expose them in the logs or config files. Secret ref handles all of that securely. Your tokens are encrypted and stored separately from your configuration. When the agent needs to authenticate with an external service, it requires the token through secret ref. The token never appears in plain text in your logs or your config. Security is the biggest thing that people talk about when getting going with Open Claw, and you can still mess it up. But, from 30 days ago to where we are now, it is much more locked down and safer, but you do need to take precautions when getting this going. Like, I have nothing logged in in the computer behind me. It has its own email. It has its own setup. It is not anything to do with any of my personal information, still. And I recommend

16:40doing it like that. So, this might be just like a small thing, but it's these kind of production-ready features that separate Open Claw from a toy AI platform or any other chatbot. When you're running a real business operations on AI agents, security is not optional. It's the first thing you need to be taking advantage of when setting up agents or your mission control security. So, they also added HEIF image support for iPhone photos. This is one of those features that seem minor until you actually need it. So, if you're building AI agents that process images from iOS devices, they're often in HEIF format. Don't get me started there. but before this update you'd have to convert them manually or use like an external tool. Now it just works. Your agent can process iPhone photos directly. For

17:26content creators like me who are always taking screenshots and photos on their phone, this removes a lot of that friction. So the release notes mention a ton of bug fixes too. I won't go through them all, but I did notice a few things that are just more stable, fewer random crashes, better error handling, faster response times. So what does this mean if you're building with AI? And why do you even care? First, local AI is getting more powerful and more reliable. You don't need to rely on cloud services that go down or change their pricing overnight. You can run serious AI operations on your own hardware with enterprise level features. Second, the barrier to building complex AI applications is getting lower. These pluggable context engines mean you don't have to build everything from scratch. The community is going to create all kinds of memory systems. You could just

18:13plug them in. Third, this is what production-ready AI looks like. It's not just about having smart models, it's about reliability, security, performance, and all that boring stuff that no one talks about. That's what really matters when you're running real businesses on AI. If you're just getting started with AI, this might seem overwhelming. I know when I first got involved in this space about a year ago, it was crazy. You just don't know what to focus on. And now more than ever things are being released and shipped faster than ever. There is a lot of technical complexity here, but that's exactly why we built the community in school. We have people that are wanting to build around other builders and being held accountable for the ideas that they have but haven't shipped anything yet. So in Shipping School, I work with you directly to build and ship AI

18:59applications. We do three live calls a week, build sessions where we code together, bootcamps where I teach specific skills about what I'm doing in my setup, and Q&A sessions where we can ask anything and chat back and forth about your goals and the business that you're trying to build. You also get a Next.js starter kit that's ready to go with Open Claude integration. This is how I built three of my apps within days. Easy Flip did take 45 days, but that's the first one that I tried to get going. We're at $300 MRR with that one. And the two other ones were built much quicker using this starter kit. So, you learn how to use Claude code, you learn how to use those AI coding tools, and all the stuff I used to build my own applications. Most importantly, you actually get to ship something. We don't

19:45just talk about building, we build together. So, I help more non-technical entrepreneurs get going from like their idea to an actual shipped product. If you got an idea for an AI app, but you don't know how to code, that's fine. This is exactly what Shipping School is for. The whole goal is to get you from zero to one fast as possible, not to make you an expert developer overnight. I don't claim to be some guru that has 50 years of software development experience. I'm just sharing my journey and what I'm learning so I can teach you to get the hit the ground running and get going fast. So, you get your first application live and start making money. We talk about marketing strategies, how to get customers, whether through content creation or building in public. So, what I really think about this Open

20:31Claude update, we're at a turning point with local AI. For the first time, you could run enterprise-level AI operations on consumer hardware. You don't need a Mac Studio to do this. The features we're getting now used to require massive cloud infrastructure and teams of engineers. My entire content business runs on a $600 Mac mini. I'm generating scripts, thumbnails, social posts, newsletters, everything. And it's getting more powerful every month. I'm not talking about how AI could transform your business. You're going to If you're not thinking about how AI can transform your business, you're going to get left behind. If But if you're paying hundreds of dollars a month for cloud AI services, you might be overpaying. Local AI isn't just cheaper in the long run, it's more private, more reliable, and

21:17more under your control. Updates like 3.7 prove that local AI is ready for serious business use. The hardest part is getting started. There's a learning curve, there's technical complexity, but once you get over that hump, the possibilities are incredible. That's what I do in Shipping School. That's why I built the community sharing the journey that helps you get over the initial hump and get going. Start before you're ready. Starting messy, but building with other people in a community is probably the best part. If you want to start building with AI, you could join me. I got a link down below. I'll help you get your idea off the ground. The Open Cloud update is a big step forward for local AI. Better models, better memory, better reliability. If you're serious about building with AI, this is the direction

22:02where things are heading. We'll see you guys in the next one.