← Claude Code
Creator Magic
My Exact Claude Code Setup Still Allowed After Ban
⏱ 17 min video · 3 min read29 May 2026Worth watching
TL;DR
The creator demonstrates 'Tank', a self-built browser-based mission control dashboard for running multiple Claude Code agents simultaneously on a local server, using Forgejo for self-hosted git and tmux for persistent sessions. The setup is designed to stay within Anthropic's Terms of Service by using interactive terminal use rather than banned third-party automation tools.
Key points
1
Tank is a custom-built dashboard that lets you run multiple Claude Code agents in parallel, each in its own sandboxed tmux terminal, all visible and interactive from a single browser UI
2
The setup runs entirely on a local server using Forgejo (self-hosted git) instead of GitHub, with no cloud orchestration — only Anthropic API calls leave the machine
3
Anthropic banned third-party tools like OpenClaw and Cline from Claude subscriptions in April 2025, and is closing programmatic/SDK access on 15 June 2025, but interactive terminal use remains permitted
4
Tank includes features like pull request creation and merging, a mirrored to-do list from a markdown file, real-time usage tracking (5-hourly and weekly), screenshot/image drag-and-drop into prompts, and repo flattening to a single text file
5
The source code is available in the Creator Magic Premium community with the coupon code 'tank' for a discount; it ships as-is with no support and requires a Claude Max subscription
Actionable insights
→
Use interactive Claude Code terminal sessions (not third-party automation tools) to stay within Anthropic ToS, especially after the 15 June 2025 programmatic access closure
→
Run Claude Code inside tmux windows to maintain persistent sessions across multiple projects without needing a VPS — context is never lost when a window closes
→
Self-host your git with Forgejo instead of GitHub to avoid cloud dependencies and recent GitHub security vulnerabilities around private repo exposure
→
Never run Claude Code agents as root and always test in an isolated environment — Tank skips all permission prompts by default, so agents execute without confirmation
→
Use a SQLite database to log all Claude Code chats so you never lose session history, and build real-time usage tracking to monitor your Claude Max allowance
Notable quotes
“I realized I am the bottleneck. I literally can only work as fast as my brain can think.”
“Every agent here runs with permissions skipped because I like to live dangerously. It does not stop and ask you to do things. It just does. Which is the way I like it. That is powerful. And also a little bit dangerous as well.”
“This is me building it using the thing I built to build it, if that makes sense.”
Worth watching?
✅
Worth watching the full video?
Watch if you are actively using Claude Code and want a practical, compliant way to run multiple agents in parallel — the live demo is genuinely impressive, but the key setup details and compliance reasoning are all captured here.
Saved you some time? The creator still deserves a like.
Found this by searching? That's the hard way.
summree watches the AI channels you follow and sends you the summary the moment a video drops. Nothing gets past you. From £4/month.
Try it free