example-agent-stack.dev

Building an agentic AI environment that holds up in real use

AI is changing the way people work and live, and it is happening fast. This site documents one concrete response to that reality: a practical environment built around local coding agents, OpenClaw on a VPS, private networking, helper scripts, GitHub, and a memory model that preserves useful context over time.

Read the journey See the architecture

Purpose

What this site is for

This site is for people who want to build an agentic environment of their own and need a concrete starting point. It documents the setup in chronological order, from the first workspace and memory decisions to the current multi-layer system that combines local agent workflows with remote persistence and a narrow public edge.

What it covers

  • the workspace and memory model
  • the local laptop execution layer
  • Claude and Codex as working agents
  • helper scripts, hooks, and ingestion
  • OpenClaw on a VPS
  • GitHub, publishing, and the public site boundary
  • Tailscale, firewall rules, and private service exposure

What matters here

  • practical workflows and real commands
  • clear architecture and explicit boundaries
  • decisions and rationale, not just outcomes
  • memory that stays useful over time
  • security as part of the design
  • generic placeholders for public examples

Reading path

Follow the system from the beginning

The journey pages are the backbone of the site. They show how the environment was built step by step and how each stage led into the next.

01

Workspace and memory foundation

Set up OpenClaw, let it seed the workspace files, define the startup sequence, and establish the durable-memory versus daily-note split.

02

Local foundation

Build the local execution layer around the laptop, repos, coding agents, shell tools, and helper scripts.

03

OpenClaw on the VPS

Move the persistent layer onto a private VPS, keep it private, and connect it back into the local environment through Tailscale and SSH.

04

Hooks, scripts, and ingestion

Add the post-commit hook, the Claude Stop hook, helper scripts, and parser improvements so useful context feeds the memory layer automatically.

05

GitHub and publishing

Connect the repos to GitHub, use it as the external version boundary, and publish the site through GitHub Pages.

06

Current state

See how the local machine, OpenClaw, hooks, GitHub, and the security model fit together today.

Core idea

One system, several boundaries

The environment is intentionally layered. The laptop handles active work. OpenClaw on the VPS handles persistence and orchestration. GitHub handles versioned code and public publishing. Tailscale and the firewall keep the private layer private. Hooks and helper scripts are what make the pieces work as one system rather than as disconnected tools.