A self-hosted mission control dashboard for Claude Code enables running multiple AI agents in isolated sandboxes with live terminal monitoring, while maintaining compliance with Anthropic's terms of service by using interactive terminal hooks rather than programmatic API calls, allowing developers to manage multiple concurrent tasks, track usage metrics, and integrate with self-hosted Git repositories like Forgejo.
Deep Dive
Prerequisite Knowledge
- No data available.
Where to go next
- No data available.
Deep Dive
My Exact Claude Code Setup Still Allowed After BanAdded:
This is mission control for Claude Code.
I can spin up as many agents as I want.
Each one runs in its own sandbox, and I can watch every single one of them work live right here in the browser.
By the end of the video, I'll tell you how you can get this to and run it yourself.
The whole thing, by the way, runs on my own server, so I don't use GitHub.
I don't use cloud for the coordination or orchestration of Claude Code.
Nothing leaves the building apart from calls to Anthropic.
And after all the bans and billing chaos that happened earlier this year, this setup is actually still allowed.
Let's dive in.
So you'll see up in the top right, I can start a new task.
I can change my settings sign into my cloud account, look at my weekly and five hourly usage.
Let's actually do that.
Let's start a new task. I'll type in a prompt.
Let's say make a favicon for this app.
That's the really simple prompt.
I hit enter and look at this immediately.
The task is given a name create app favicon that's done by local AI.
It's now prompting and if I click terminal it pops open the terminal and shows Claude Code running for me right here.
Kind of like Termius, but with my own rules.
Now, first the question.
I know every single one of you is about to type in the comments.
Is this even allowed?
Now, I will say here, I'm not a lawyer, But I did read through Anthropic terms.
And look at this. This is actually running.
This is controlled by me. This is interactive.
I'm actually using hooks to talk back and forth between Claude Code.
So everything that's happening here in the terminal is being echoed right here in my mission control.
Now, back in April, if you were running OpenClaw or Cline or anything else, that was the policy.
Then basically if you were running it from a Claude subscription, it stopped working overnight.
The head of Claude Code put it really bluntly, Your subscription will no longer cover third party tools like OpenClaw now, one loop that Anthropic closing on the 15th of June this year is the ability to use programmatic things, such as the Agent SDK and even the Claude Dash P command, but they don't ban interactive terminal use.
and essentially what is happening here in my mission control is I'm doing interactive use with a terminal and then getting the hooks back and even the response here.
So everything that Claude Code responded to me is put right here.
now you see that it's done the task I've asked for.
So I can actually go ahead and say, this looks really good.
So I'll tell it to commit and merge.
Let's do that right now.
Now notice I've actually put in icons here that allow me to open pull requests and even mark the task is done.
I'm just going to say commit and merge and send it off to Claude Code.
notice here it resumes the terminal.
It loads it back up.
It puts in the commission merge that I've just written in my prompt box.
And this will go ahead and get Claude Code to actually merge this into my repo.
It's really, really cool.
And you might say, well, Mike, why did you build something like this when you can just use something like Termius or any other app allows you to spawn multiple terminal windows and run a bunch of Claude Codes.
The reason is I've realized I'm the bottleneck.
I literally can only work as fast as my brain can think.
So I basically needed to build a mission control like this so I can keep an eye on everything I'm doing, include code, and even run multiple instances at the same time.
Now you might be asking, what am I using?
Forgejo because I'm certainly not using GitHub.
now. As much as I love GitHub, it recently had a security vulnerability, and even private repos could possibly have been read by hackers.
So you'll see, just as I said, that is background waiting for everything to deploy.
I get a new status light here and over here I'm actually using something called Forgejo, which is a local version of git, and it's hosting my entire repo for what I've called tank.
Now, some of you may be asking, why have I called this overpowered dashboard for Claude Code, interactive use tank.
Well, if you know, you know.
Okay.
Looking over here, it looks like my tank silhouette icon has been added and deployed.
So let's check. let's do a hard refreshing.
Oh look at that.
We've actually got a tank icon now up there that actually looks really neat.
That was designed by Claude Code.
And over here in the git repo. Yes.
We've got a beautiful tank icon.
I guess that kind of fits the theme of the name of my dashboard.
All right.
So you can see it's really cool down here.
I can see all the tasks that I've been working on.
This keeps me on point.
But most importantly, what you might notice is I've also got a to do list of things I need to do.
So things like the AI assisted to do refinement that's done the smoke test to make sure it works after each deploy that's done.
And where does this link?
Well, this links inside my repo.
If I actually go to the code here, you'll see everything.
I have a full read me on how this works, but I have a to do.md file, so if I click into that it's a simple markdown file.
If we go into the raw of all the to do things that I need to do with my app, This is mirrored inside my dashboard, meaning I don't have to have a million SAS apps to keep an eye on what I'm working on.
I can do it all in one dashboard.
That is kind of meta, because I'm building it using the thing I built to build it, if that makes sense.
All right, so let's actually use my dashboard to build a real feature for tank and see how it works in practice.
Now you'll notice because I'm using local Forgejo as my git repo, my source of truth.
I've got all of the code files here, and sometimes I like to have them as a flat text file that I can feed to AI, such as Google's Gemini, and ask questions before I go back and code more.
Now obviously I've got the ability locally here to download as a zip or a tar or whatever like that.
But if I want to get the flat file, that's a feature I need to bake into tank.
So I'm going to go up to new Task just here and create something new.
And I'll start by saying add a feature that flattens any repo in my Forgejo into a single text file with a button in the UI.
That's pretty simple.
Create and run and it will go straight to work.
Notice my local AI says exactly what it is.
In summary, add repo, flattening feature and Claude Code spins into action.
I can see how long it's been running and it indicates over here that it is running with a yellow flashing dot.
All of my completed tasks are all blue here, and if I click into the terminal, I can instantly see what Claude Code is doing.
It's running on the latest Claude Opus 4.8, So you can see I'm using this app to drive my genuine, real interactive Claude Code usage, and that puts me in the white area of the Anthropic ToS.
If I hijack the token, try and impersonate, use a third party tool, or farm this out to a team, then I'm in the black area of Anthropic ToS.
And you can see right now Claude Code is working away.
I can actually watch it.
I can even interact with it here inside this pop up terminal forward slash btw if I want to ask it by the way question I can do that.
Or I can just exit out and wait for it to complete and watch all the hooks in my overpowered dashboard.
now here's the coolest thing about what I've built.
I can obviously work on different projects right here in my projects tab.
I can see all chats here, but I can actually have a chat with Claude Code.
And you can see I've been researching taxi companies and household admin so I can chat with Claude Code like a general LLM, so I can say something like find the best Mexican food in Limassol, Cyprus.
I think I'm typing that badly.
Search online with sub agents and tell me, okay, very simple prompt there, create and run.
We can leave this running in the background.
Find best Mexican food in Lima.
So it's now started up that interactive terminal.
It started the prompt and Claude Code is going to go out and find me the best Mexican food while I'm still working on this repo flattening feature.
And if I go to the all view here, I can see everything running at the same time.
So I can see my repo flattening feature is running at the same time as I'm looking for the best Mexican food.
This allows me to multitask simultaneously on different projects, different ideas, and even chats on the food I'm going to eat for lunch.
Okay, we'll keep an eye on how this is going and you can see already it's finding out two agents.
Yep, it's finding restaurants and food from TripAdvisor.
So you can see the agents have spun up here from the hooks that are coming back to me.
It's searching a ton of searches online.
It's looking at a ton of different websites.
This is way more comprehensive than a generic chat with Claude or ChatGPT.
This is Claude Code going hard to solve my problem.
and as you can see up here, I've actually created this dashboard to update my five hourly usage and weekly usage in real time.
10% is what I've used so far in this session, and 3% of my weekly allowance.
I've got to say Anthropic usage limits are getting more and more generous by the week.
Okay, best Mexican food is complete.
Let's go back to that. And wow.
Okay, not only does it take the response from Claude Code, but it formats it with nice markdown headers and bold here.
So these are the places that I would go to get a good Mexican feed.
All summarized here for me and saved in my chat history.
So I can always go back and find it and look at what I need.
This is great.
I never lose a Claude Code chat again because it's recorded inside my dashboard.
I'm using a very simple SQLite database to make sure all the data is right there.
Now the best thing about this, because my Claude Code sessions will resume, I'll just say, do any of these deliver and we'll hit enter on that.
This will then resume my Claude Code session in the terminal.
There you go.
You can see its previous responses here, and it's puzzling over that fact.
It's checking the direct availability across world foodie and bolt food, which is great.
And for the observant people here in my audience, you may notice this green bar down the bottom.
Yes, I'm running every Claude Code session in a tmux window.
This means that I can have multiple Claude Codes running on my local machine, and there's no need to spin up a VPS to have multiple persistent Claude Code sessions, because my app does it all for me.
And look at that.
Best Mexican food has come back and wow, I can see a bunch of them actually do deliver, which is really cool.
This is the only one that I need to have a romantic dining experience at.
So I think my agent has done a really thorough search for me.
Not only can I use Claude Code to work on my coding projects, but I can also get it to go deep on something I need to research.
Now this is writing code for me.
This is fantastic. One on one is working away.
Look at this.
I've actually got a feature I was working on yesterday, which was the ability to drag and drop images into the chat box down here, which is something I could never do with Termius.
It's running on a VPS.
So as you can see, yesterday I jammed back and forth with exactly how I was going to make this happen.
Claude Code responded to me.
Now if that was running in a tmux window that closed, I would have lost all that context and had to do it all again.
here, I can resume the chat, it's gone.
Everything that needs to happen so I can just say yes, do this and let Claude Code go to work and resume the session.
Opening an interactive terminal window with our previous conversations all there in memory.
It's now going to go ahead and build two features at once.
While I sit back and drink coffee.
I can click into either one.
And actually, just as I've done that, this repo flattening feature has been completed.
This one is still running and I can view it writing the code in the terminal.
Even working on a to do list.
This is really, really cool.
All right well it's doing that.
Let's look at the repo flattening feature here and see what it's come up with okay done I added a little flatten new Forgejo repo into a single text file feature end to end.
Now I can open the pull request.
And actually I've created a button that does just that.
So let's click that button.
It's going to ask AI to open a pull request on Forgejo.
We'll give it a moment to do that.
This is amazing.
Here it is in my locally hosted git repo.
Ready to create the merge.
You'll see the buttons that I've created here for my mission control include the ability to view the pull request and even merge it.
So I'll click merge So you'll see at the moment this is open.
But if I refresh it and there we go, it's done.
It's merge the I did it for me.
Meaning now if I go here actions are running locally.
Yes that's right I've actually got this setting up the job and installing the SSH key.
This is all happening locally right now.
It's redeploying my mission control And it will test it to and within a moment.
Look at this. Yes, the job is complete. back to tank.
And I can see I've got a new icon here. This allows me to.
Yes. Flatten the repo.
Now I need to set up my Forgejo tokens.
So let me do that.
Okay I now save those details.
So let's click in and look at this.
Now I can access all of my repos here.
This is really cool.
And I can flatten and download any one of them.
This is a feature I literally created in front of you.
And look at that. The download happens.
This is insane.
just pop that file on my desktop so I can hit spacebar to preview and look at this flattened repository.
This is absolutely amazing.
All auto generated.
No need for me to use any tools online to get a flat repo.
I can do it in just one click.
It's absolutely amazing.
Now let's go back to all and have a look at my drag and drop image.
It looks like this is completed all three task complete summary of what shipped.
So it's got the back end, the front end and the ability to deploy it.
It says it works so I'm just going to trust it.
Let's open a PR here and we'll get this into my local git repo.
I'm not even going to check the work.
I'm just going to merge it in by using my merge button.
okay. So let's try something with a screenshot.
I'm going to screenshot the top left of this.
Now let's just mark up the screenshot by putting a little red line in here and clicking done.
now we'll go ahead and create a new task and say each project should have status lights.
If tasks running and past favicon use can choose where red line is in screenshot.
Okay, this is where I test the new feature by dragging and dropping the screenshot into my prompt.
And look at that.
Well, it's attached a screenshot.
Finally, I can command tmux terminals and attached screenshots.
Let's run and see if this actually works.
No guarantees it will actually run.
Look at that.
It's actually uploaded my screenshot to a temporary location where I'm running my mission control.
Open this up and see if it can actually read the file.
It's trying to do it right now.
And look it says here I can see the dashboard.
The red line in the screenshot runs down the left edge of the project rows, proving that it's seen what I asked it to do.
okay. We can see it's wrapping up here.
Now the tmux will close cleanly.
Once it's finished.
That's done. It's confirmed that.
So now I can open a pull request I can get this merged in and see what it looks like.
Okay.
It's deploy this refresh and wow, look at that tank is actually flashing And now it's completed. So the icon disappeared.
I can set the icon up here. Wow.
Okay, I can search through a bunch of icons from this library, so.
Well, let's see if we can find a tank.
Some kind of tank I don't know, there's no tank, so I'll just go for this ship.
And look at that.
Now I actually have an icon associated with the project I'm working on.
So there you go. That's the tool.
A fleet of Claude Code agents on demand live terminals that you can watch and jump straight into, interactively pull requests on your own self-hosted Git, and a tool that literally builds itself and explains itself as it goes.
Now, one honest warning before you go and copy all of this.
Every agent here runs with permissions skipped because I like to live dangerously.
So remember, it doesn't stop and ask you to do things.
It just does.
Which is the way I like it. That's powerful.
And also a little bit dangerous as well.
Run it isolated, have a lock down to test it and never run it as root.
Make sure you do all of those things and you'll be fine.
Now I'm putting the full source in my premium community.
The link will be down below.
And if you use the coupon code tank, you'll get discounted membership for your first three months of Creator Magic Premium.
Now remember, this will be a snapshot as is with no support.
Bring your own Claude Max or plan whatever you're using, and if you want me to further develop features, maybe you find this useful.
Or maybe this is just a funny toy that I'm enjoying this week.
Let me know in the comments and I'll make a video with an update in the future.
This has saved me a great deal of time.
Panic and worry around my Claude Max subscription.
So if you liked this video, throw a like on it and I'll see you inside the premium community.
And YouTube is showing a video on your screen now you should watch next.
Thanks.
Related Videos
OpenHuman VS Hermes AI: Who Wins?
JulianGoldieSEO
285 views•2026-05-29
Long-Running Agents — Build an Agent That Never Forgets with Google ADK
suryakunju
142 views•2026-05-30
5 Mind Blowing Omni Uses Cases
PaulJLipsky
1K views•2026-06-02
This computer is made from real human brain cells. And you can buy it.
Talktmsmedia
3K views•2026-05-28
BREAKING: Microsoft’s New Image Generating Model Beat Out GPT 1.5 and Nano Banana 2
aimmediahouse
122 views•2026-06-03
I Made the Same Anime Fight Scene in Every AI Video Generator
NobleGooseAnime
295 views•2026-05-30
Nvidia Bets Big On AI PCs | New Chip To Power Windows Laptops | Technology | AI Updates | N18S
cnnnews18
3K views•2026-06-01
I Tested NEW Opus 4.8 on Four Projects (Updated LLM Leaderboard)
AICodingDaily
298 views•2026-05-29











