A foundation for your personal AI workspace

Choose your agents. Build your stack.

Give your personal chief of staff, coding tools, and automations a place to work—on one cloud machine you control.

Build your workspace Explore how it works
Runs on
AWS, DigitalOcean, Hetzner or Google Cloud
Access
Private, over your Tailscale network
Source
Terraform, open on GitHub
The AgentStack workspace machine One modular cloud machine drawn as an appliance. A base chassis on feet holds the persistent data drive and a keyed private-access gate for Tailscale, with a backplane spine at the back. Three removable trays stack on corner standoffs and dock into the spine: a workflow tray with n8n as a pinned sequencer drum and a small Postgres cartridge wired only to it; the coding workspace tray with a Codex card and a slot for an optional Claude Code card; and the chief-of-staff module, whose status display names the personal agent runtime and the external model provider reached through its uplink port. A lid closes the stack. Scrolling lifts the lid and each tray out of its dock in turn. 04CLOUD HOST 03 claude code codex 02 01
  1. Chief of staff OpenClaw or Hermes
  2. Coding workspace Codex, optional Claude Code
  3. Workflows n8n, its Postgres database beneath
  4. Cloud foundation one host, data drive, private access
  5. Model provider an external API, via the uplink port
AS—01 Your personal AI workspace

How it works

One machine, four layers, each with a job.

Start with what you want done, then follow it down the stack. As you scroll, the drawing opens to show where each part lives and how it connects.

  1. 01 Chief of staff

    A personal agent that keeps your work moving

    Pick OpenClaw or Hermes as the agent you talk to. It keeps context, plans next steps and, once you connect it, hands work to your coding tools and workflows.

    Its intelligence comes from the AI provider you choose: OpenAI, Anthropic, Google Gemini or Groq. The provider is reached over its API from outside the machine. It is not a second server you run.

    • Runtime: OpenClaw or Hermes
    • Models: an external provider API
    • Dashboard: private, on your tailnet

  2. 02 Coding workspace

    The place where work actually gets done

    A persistent Linux workspace with Codex included and Claude Code available alongside it. Your home directory lives on the data volume, so projects and tools survive restarts.

    You reach it over SSH from devices on your tailnet. Each coding tool signs in on its own; picking an AI provider for your agent does not sign them in.

    • Codex included
    • Claude Code optional, alongside Codex
    • Home directory on persistent storage

  3. 03 Workflows

    Repeatable work, made dependable

    n8n runs workflows and automations that follow the same steps every time. Your personal agent or coding agents can help you build them, and can use them once you connect them.

    PostgreSQL sits quietly underneath as n8n’s database. You don’t need to manage it separately.

    • n8n for deterministic automations
    • Postgres as n8n’s local database

  4. 04 Cloud foundation

    One machine, in your own account

    Everything runs on a single cloud machine you own: an AWS EC2 instance, DigitalOcean Droplet, Hetzner Cloud server or Google Compute Engine VM. It has a persistent data volume and private access through Tailscale.

    Terraform describes the machine. You review the plan and its cost before anything is created, and your cloud provider bills you directly.

    • One machine, one data volume
    • Private access boundary
    • You approve the plan first

AgentStack installs the pieces on one machine. You decide what your chief of staff is allowed to reach. Build your stack →

Build your stack

Assemble your workspace in four choices.

Each choice fills in the drawing. The coding workspace and n8n are always included. Nothing here signs in, deploys anything or asks for secrets.

Step 1 of 4 · Nothing chosen yet

    Setup prompt

    Your setup prompt appears here.

    Finish the four choices above. You’ll get plain text to give an AI coding agent on your own computer, which then guides the real setup with you.

    agentstack-setup-brief.txt 
    
          
    
          
        
    
      
    
      
    
    

    
    
    
      
    
        
    
          
    Deployment
    
          
    What setup does, step by step.
    
        
    
        
    A preview of the journey your local agent walks you through. It’s a simulation: no commands run, and the checks show what to verify on your own machine, not real results.
    
      
    
      
    
        
    
          Simulation · nothing is executed
          Example configuration
        
    
        
    
          
          
    
            
    
              
    Press Run demo to preview the setup, or pick a stage.
    
            
    
            
    
              
              
              
            
    
          
    
        
    
        
      
    
    
    

    
    
    
      
    
        
    Get started
    
        
    Bring an agent, an account and a few keys.
    
      
    
      
    
        
    
          
    Before you begin
    
          
      
            
    • An AI coding agent on your own computer to follow the setup prompt.
      • 
            
    • A cloud account on AWS, DigitalOcean, Hetzner or Google Cloud, with permission to create a machine.
      • 
            
    • A model provider credential from OpenAI, Anthropic, Google Gemini or Groq.
      • 
            
    • A Tailscale account and auth key for private access.
      • 
            
    • Terraform 1.9 or newer. Your agent can help install it.
      • 
          
    
        
    
        
    
          
    What you own and decide
    
          
      
            
    • The machine runs in your cloud account, and the provider bills you directly. You see the plan and expected cost before it’s created.
      • 
            
    • Credentials stay in your own secret manager and Terraform state, which you keep private.
      • 
            
    • Some steps need you after deploy: approving devices, signing in to coding tools and connecting your agent to workflows.
      • 
          
    
        
    
        
    
          
    Ready when you are.
    
          
    
            Build your workspace
            github.com/ferrerluis/agent-stack