Viktor Vasylkovskyi
Back to Blog

Running an AI Coding Bot on Raspberry Pi - Complete Guide

Viktor Vasylkovskyi•

After setting up my Raspberry Pi with complete observability, secrets management, and CI/CD pipelines, I realized I had built the perfect platform for something more ambitious: an autonomous AI coding assistant that could work while I sleep.

This guide documents my journey building a self-hosted AI coding bot that:

  • Runs entirely on a Raspberry Pi (no cloud costs)
  • Delegates work to specialized AI agents (OpenClaw + Codex)
  • Maintains its own knowledge base (Obsidian vault)
  • Generates its own work queue from business roadmaps
  • Works autonomously 24/7 without human intervention

OpenClaw System Architecture

What You'll Build

By the end of this series, you'll have an AI coding system that:

  1. Reads your product roadmap written in plain Markdown
  2. Generates GitHub issues from business requirements
  3. Picks up issues automatically and delegates to Codex for implementation
  4. Writes, tests, and commits code with full validation
  5. Opens pull requests and notifies you on Discord
  6. Stores findings in an Obsidian vault as institutional memory

The entire system runs on a single Raspberry Pi 4B, costs pennies per month in electricity, and operates completely under your control.

Why This Matters

Most AI coding assistants are:

  • Cloud-dependent - require internet, subscriptions, and trust in third parties
  • Interactive - need you to give them every instruction
  • Stateless - forget everything between sessions
  • Expensive - $20-50/month per developer

This system is:

  • Self-hosted - runs on your hardware, your network, your rules
  • Autonomous - generates its own work from your roadmap
  • Persistent - builds institutional knowledge in Obsidian
  • Cost-effective - $2-5/year in electricity after initial hardware cost

Table of Contents

This tutorial series is organized into six parts:

  1. Secure Setup - Security-first configuration: Docker isolation, fine-grained GitHub PATs, and Anthropic spending caps
  2. Installing OpenClaw - Node.js setup, OpenClaw installation, Discord integration, and resolving dependency issues
  3. Performance & Stability - Expanding swap, tuning memory, systemd service setup, and log routing to Grafana
  4. Codex Delegation - Setting up OpenClaw as coordinator, ACP configuration, tmux sessions, and the mailbox protocol
  5. Heartbeats & Notifications - Wake triggers, completion signals, HEARTBEAT.md registry, and parallel agent spawning
  6. Obsidian & Autonomous Loop - Obsidian vault as second brain, roadmap parsing, automatic issue creation, and the full autonomous loop

Prerequisites

Before starting, you should have:

  • Raspberry Pi 4B (4GB or 8GB RAM) with Raspberry Pi OS
  • Docker installed and configured
  • GitHub account with repository access
  • Anthropic API key for Claude access
  • Discord account for bot interaction
  • Basic command line familiarity (SSH, tmux, systemd)

If you don't have the basic Pi infrastructure set up yet, check out my Private On-Premises Infrastructure guide which covers the foundation:

  • Raspberry Pi OS setup
  • Docker configuration
  • Grafana + Prometheus + Loki observability stack
  • HashiCorp Vault for secrets management

What This Series Is Not

This is not a guide for enterprise production AI systems. We won't cover:

  • Multi-tenant isolation
  • High availability or load balancing
  • Compliance or audit logging
  • Enterprise identity management

This is a guide for:

  • Personal development infrastructure
  • Learning AI agent orchestration
  • Cost-effective experimentation
  • Understanding autonomous AI systems

The Philosophy

The core insight is separation of concerns:

  • OpenClaw = Coordinator and interface (Discord, HTTP, dashboard)
  • Codex = Specialist worker for coding tasks
  • Obsidian = Persistent memory and knowledge base
  • GitHub = Source of truth for code and tasks

Each component does one thing well, and they communicate through simple, inspectable interfaces (files, HTTP, Discord messages).

Cost Breakdown

One-time hardware:

  • Raspberry Pi 4B (8GB): ~$75
  • MicroSD card (64GB): ~$12
  • Power supply: ~$10
  • Case (optional): ~$8
  • Total: ~$105

Recurring costs:

  • Electricity (~5W): ~$2-5/year
  • Anthropic API: Pay only for tokens used
  • **Total: $10/year **

Compare this to cloud-hosted AI assistants at $240-600/year and you break even in 2-4 months.

Code Repository

All configuration files, scripts, and examples from this series are available in the GitHub repository:

https://github.com/IaC-Toolbox/iac-toolbox-raspberrypi

The OpenClaw-specific configurations are in the openclaw/ directory.

Let's Begin

Ready to build your own autonomous coding assistant? Start with Part 1: Secure Setup where we lay the security foundation before installing anything.

The series is designed to be followed in order - each part builds on the previous one. By the end, you'll have a fully autonomous AI coding system running on hardware you control.

Next: Part 1 - Secure Setup