Anthropic Reversed the OpenClaw Subscription Ban

0:00In this video, I'm going to talk about how you're going to save a ton of money right now when you're using OpenClaw because Anthropic just suddenly changed their mind on how you can use some of their models inside through the Claude CLI in your OpenClaw setup. So, we're literally talking saving thousands of dollars a month. So, as you know, a few weeks ago, they banned third-party harnesses like OpenClaw with the use of your Claude subscription. That is no longer the case. I have the source link down below. You should check it out. So, if you were in this space at all, you saw the headlines, right? Anthropic comes out April 4th and says, "No more using your Claude Pro or Max subscription with third-party tools." You know, if you want to keep running OpenClaw or OpenCode or any of those frameworks on your subscription, forget

0:45about it. You have to switch over to the Anthropic API or pay per token. I was on track to spending over $1,500 this month using Open or using Anthropic's token usage and using Opus 4.7 that runs the whole content pipeline. Now, I'm literally just paying my $200 subscription plan. So, basically, Hacker News today blew up saying now you can use the Claude CLI through OpenClaw. So, those blog posts rolled in and Hacker News blew up and and everybody was saying Anthropic killed the subscription play for OpenClaw. And I was right there in the mess with everyone else. You know, I had cron jobs running on OAuth tokens. My content pipeline leaned heavily on that budget and I watched the bill start to look pretty scary, but I was willing to pay it because when you

1:30switch over to pay per token, especially for the agentic stuff in your agentic workflows, that chews through a ton of contexts. The numbers add up real fast. I had GPT-4 running a lot of the cron jobs through my $100 month subscription and I just kept my main chat on Opus 4.7, but still $1,500 a month just to chat back and forth, that is rough. And you know, nobody's really talking about this second chapter of this evolution yet. This just dropped today. It's news to me. I just reconfigured my whole OpenClaw setup now to go through the Claude CLI and it's been amazing. It's like good old times. So, what happened is just to break it out. Anthropic staff members came back around and clarified the policy. So, no more gray area. The thing they were shutting down was one

2:16very specific pattern. And if you understand the difference, you can keep running OpenClaw on your existing subscription today, legally, officially sanctioned. 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

3:02it. No fluff. I built this community because watching YouTube's only gets you so far. We launched it just 3 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. So, I'm going to break it down because the technical piece is the whole game here. When they said no more subscriptions for third-party tools, what they actually meant was this. Some of the features and tools were pulling your OAuth token out of the Claude code install on your machine. Then they were turning around and calling the Anthropic API directly, pretending to be Claude code. That pattern makes Anthropic unhappy for just

3:47a couple reasons. The first one is it skips the prompt caching and the infrastructure tricks that Claude code actually does to be efficient. The second thing, it puts a ton of load on their servers without them getting the benefit of running these sessions themselves. Boris Cherny, who runs Claude code at Anthropic, was very direct about that on X. He said subscriptions were never designed for that kind of continuous automated demand these tools were basically generating. And you know what? I get it. It's a great value for agentic workflows. But there is a second pattern that looks similar on the surface and is actually completely different when you look at what is happening under the hood. The pattern is shelling out to the Claude

4:33CLI in the binary itself. So, instead of extracting your token and impersonating Claude code, OpenClaw literally invokes the Claude command on your machine with the dash P flag. Claude code runs the request, Claude code does the caching, Claude code handles the authentication, and OpenClaw is just the thing that kicks off the prompt and reads the result back. That is the pattern that is still sanctioned. Anthropic staff confirmed that to the OpenClaw team directly. OpenClaw treats that as official green light unless Anthropic publishes a new policy saying otherwise, which honestly I wouldn't be surprised. But why do they care about the difference? Because when you shell out

5:18to the Claude CLI, all of the infrastructure concerns Anthropic had about the first pattern basically go away. Cache hit rates stay high because Claude code is running the show. Rate limit accounting stays clean because it is the same binary they wrote. There is no pretending. You are just running the tool they ship on the machine where you installed it with the login you already have. And that means you can wire OpenClaw up to use it, which is exactly what I did this afternoon. So, how does it look in practice? If you want to keep it simple, get Claude code installed on your machine, open it up in a terminal, point it to your OpenClaw workspace folder, and say, "Switch me back to the OAuth token. I'm tired of paying per token usage." And it will take you through the steps to connect back

6:04through your subscription using the Claude CLI. So, in practice, you could, if you want to dive deep into it, I'm just saying that for easy-to-understand people, if you just want to get it going, have Claude code do it. But if you want to do it yourself, you can open your OpenClaw JSON file, and then you just find the section where the provider block is, and then you set the provider to Claude-CLI, and then you set the mode to OAuth, and that's it. But I don't think I I think most people want to just go the easy way and just have Claude code do it. That's what I did. You could do it. But if you want to get into your JSON file, that's what you could do. It's just a one-line change. But the next time your session runs now, you'll want to start a new session, and then OpenClaw will fire off and the Claude command on your Mac will be fired off, and instead of basically calling the Anthropic API or the pay per

6:50token, the login you already have from Claude code and the subscription you already are paying for, that's what's running the tokens, not the API key and not per token. So, I'm running this right now in my main session, Opus 4.7, and I love it. I've missed it so much. Now I can just blab all day long and it's amazing. And I feel like that's what agentic workflows are supposed to be like. You're supposed to be able to build in this way. You know, $1,500 a month, but I was now using my OpenClaw less. I was working more in Claude code. And Claude code pointed at my OpenClaw workspace folder and open and Claude code pointed at my mission control folder. Now to have this back, I'm not spending $1,500 a month and I'm just going through my subscription and I'm

7:35able to strategize and build out plans and workflows and outreaches back how I used to, back in the olden day. And the cool part is is I have my voice profile set up. My whole setup is going through my Max subscription and I love it. Now, here's the part that nobody wants to talk about because every hero story needs a catch. There is a little catch. The catch is that your subscription has a usage cap. And this usage cap is pretty messed up. As you know, you could hit your 5-hour usage in 10 minutes, and that's a problem. So, pro plan has one, the max plan has one, and it's still finite. Claude code tracks every call you make, and when you hit that cap, you get a message telling you to add more. And I learned this the hard way a few days ago. What happened is I had a cron job, I call the Rizza, does the outlines

8:20and script videos every morning at 6:00 a.m. It's pretty heavy usage, long context, big model. But the Rizza fired off, started pulling research, started writing, and you know, part way through it, it slammed into the 5-hour usage cap, dead in the water. Got a message saying to go to claude.ai to up the usage. You know, that was a wake-up call. The subscription was not actually unlimited, it's just cheaper until I pushed it too hard. So, how I fixed it, and you know, this is part of where you really need to pay attention here if you're running OpenClaw just at any kind of scale. You don't want to route everything through your Claude subscription. You save the quota for the stuff that actually needs Opus level quality. So, for example, my main session where I chat back and forth with my main agent, If you're struggling to keep up with content, well, I'm about to

9:05save 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, thumbnails, X posts, [music] blogs, outreaches, clips, newsletters, all of it. So, I went from 1,000 subscribers to 4,000 subscribers on YouTube in 7 days using this exact system. Every single morning I wake up and the content's already done. I spend maybe 15, 20 minutes reviewing and approving them and I move on with my day. [music] It works for any niche, fitness, finance, real estate, marketing, whatever you are building, and it is 100% completely customizable to your use case. [music] So, you get

9:50the 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 so it doesn't sound like AI slop. $97 one time, it's not a subscription. [music] I'll put the link down below and you'll thank me later. The one where I'm thinking and planning and making decisions, that stays on the Claude CLI subscription path. Every content agent, every cron job, every automated job that runs in the background, those are all moved over to GPT-4. And the output's been really good. And I'm using my $100 Open AI plan for that. It's way cheaper per token, different money pool, and doesn't touch my Claude cap at all. The rule I put my tools file is the one

10:38sentence, right? Only the main session touches the Claude OAuth path. All crons and coding tasks go to the Open AI API through the subscription, through my Codex subscription. That single rule took my burn rate from crazy to manageable in about 10 minutes of just editing a few cron configs. You can have Claude Code do that as well for you. Here's the thing, if you do not have a Claude Pro or Max plan, this whole trick still works, you just do it with a different provider. Because the whole point is that you're running a CLI tool that somebody else wrote with their subscription off through their binary, any provider who offers that pattern works. So, for example, GLM 5.1 has a coding plan, they call it the GLM 5 coding plan. It starts around $10 a month and for the light tier it tops out

11:23at about 80 and then for the max tier about 100. You get access to GLM 5.1, GLM 5 Turbo, and few others. Their plan is specifically designed for this use case, run a coding CLI locally and then it's billed to the subscription. Even Quine has the same coding plan, too. It's the same idea, it's around $28 a month for their main tier, six plus models you can switch in between, and it includes Quine 3.5, Deep Seek, and GLM Flash, MiniMax, Kimmy, and a couple others. So, good option if you want one subscription that covers a whole slew of models, right? MiniMax does have a starter tier at $10 a month for 100 prompts per 5 hours. It's a solid entry point if you are just testing the workflow and Open AI has their own coding plan that works a lot like Claude

12:09Code does. It's pretty good, Codex 5.3. It's a different off, different usage terms, but same core pattern. You install the CLI, you log in, and then you run it through your Open Claude config. Pick whichever one fits your budget, but honestly, you need to get Opus 4.7 for the best output. I don't think a lot of us want to waste time with anything else. The pattern is still the same. Just set your provider for the CLI, set the mode to the OAuth or whatever that provider calls its login flow, and let the subscription handle the billing. The big picture is that the story that got told really loudly in early April, that first week, that Anthropic killed Open Claude subscription play, that story was half right. We just didn't get a lot of clarity. What actually happened is they just shut down the OAuth token

12:54extraction pattern, which was legitimately bad behavior on the part of those tools, but the Claude CLI shell pattern is still officially fine. Open Claude supports it out of the box. My main session is running on it right now and if you don't have a Claude subscription, what are you waiting for? Get one. But there is also four or five other providers you can try out, but still, come on. One word of caution before I wrap this up. All of this is based on where the policy stands today. So, just few hours ago Anthropic released this statement and, you know, Anthropic could also revisit this down the road, I wouldn't be surprised if they do, but they have been iterating on how third-party tools interact with their infrastructure for months and they probably will keep iterating for a while. So, if you are using this pattern in production, watch for the policy

13:40page, I always have mine open, and watch Boris Journey on X just to see for any updates. Watch the Open Claude change log, when the rules change, you will want to know the same day, not months down the road. But for right now, today, if you want to run Open Claude not pay per token on the Anthropic API, you absolutely can. The CLI shell pattern is how you do it. One line in your config, your existing Claude Code login handles the rest, and that's the direction I would go. If you want to learn to build workflows just like this inside of our community, we do nine boot camp calls a week, we do a local LLM call, we do Open Claude boot camp calls, we do a beginner call, a ship and roast call, and we have 215 members that have joined this community in just the last 30 days. I also release videos just like this

14:26keeping you up to date on all AI news. Please hit the subscribe button and hit the notification bell cuz I release about four videos a day. We'll see you in the next one and have a blessed day.