he core premise of this guide is to empower users with an always-on, robust, and remotely accessible AI agent system by hosting Claude Code on a Virtual Private Server (VPS). This approach directly addresses the critical shortcomings of running Claude Code on a local machine, where workflows cease if the computer goes to sleep or is closed, and the inherent limitations of Anthropic's native cloud offerings. While Claude.ai chat is excellent for one-off conversations and thinking, and managed agents provide isolated sessions for single tasks, neither supports the persistent file structure and continuous operation essential for developing a true 'agentic operating system' or complex, context-aware AI agents.
The presenter meticulously details the advantages of a VPS: it never sleeps, is accessible from anywhere (VS Code, phone), allows entire teams to leverage the same file structure, and ensures scheduled jobs and workflows run uninterrupted. Crucially, a VPS setup grants tool and provider agnosticism, meaning users can easily transition from Claude Code to other AI tools like Codecs without reconfiguring their underlying infrastructure. The technical foundation for this remote interaction is Secure Shell (SSH), which facilitates a safe, encrypted connection between a local client (laptop, phone) and the remote server. This involves generating a public/private key pair, a more secure alternative to passwords, where the private key remains on the client and the public key is shared with the server for authentication.
Setting up the VPS begins with a choice between a DIY method (like Hetsner for maximum control and cost efficiency) or a managed service (like Als.io for automated backups and monitoring). The demonstration focuses on Als.io, guiding users through creating an account, adding their SSH public key, and provisioning an Ubuntu server. Once the VPS is deployed, the next step involves connecting it to the developer's environment, specifically VS Code, using the Remote-SSH extension. This extension allows developers to interact with the remote server's file system and terminal as if it were a local machine, providing a seamless development experience. Essential security steps include locking down key and configuration files so only the user account can read them. After proving the SSH connection via terminal, the guide walks through installing Claude Code and Git on the VPS, then cloning a GitHub repository—like an 'Agentic Operating System'—to inject the necessary context and file structure for the AI agent.
Perhaps the most significant challenge tackled is enabling a truly functional mobile workflow. The presenter identifies five key requirements: security, mobile accessibility (sending tasks, getting results), 'dispatch and walk away' capability (tasks continue if connection is lost), always-on operation, and access to full background context. Native Claude features like 'remote control' and 'channels' fall short on persistence, as sessions often die after disconnection or inactivity. The definitive solution presented is to combine T-Max (a terminal multiplexer that keeps sessions alive in the background) with Claude Channels (specifically via Telegram). This powerful pairing ensures that even if a user sends a task from their phone and loses connection, the Claude Code session on the VPS, nested within T-Max, continues running autonomously, leveraging all the contextual files. The setup involves configuring a Telegram bot, installing 'bun' and 'unzip' on the server, starting a T-Max session, and then installing and configuring the Telegram plugin within Claude Code. This enables secure, persistent, and context-aware mobile interaction with the AI agent, transforming it into a truly autonomous and accessible tool. The guide concludes by showing how to set up multiple Telegram bots and T-Max windows for different agents, each accessing specific client or project contexts within the overall repository, thereby creating a flexible multi-agent architecture.