# I Built a FREE OpenClaw (no Mac Mini or API Fees) Source: https://www.youtube.com/watch?v=8uP2IrP3IG8 Recap page: https://rapidrecap.app/video/8uP2IrP3IG8 Generated: 2026-02-22T16:04:12.728+00:00 --- ## Quick Overview The creator successfully built and demonstrated an autonomous AI agent, named The Pope Bot, running locally using Docker and configured with free LLMs like Ollama, achieving full scalability without needing expensive Mac Minis or API fees for basic functionality. **Key Points:** - The autonomous AI agent, The Pope Bot, is set up using Docker containers (Event Handler, Reverse Proxy/SSL, and Runner) for modularity and scalability. - The setup utilizes free LLMs, specifically mentioning Ollama running locally, avoiding ongoing API fees associated with services like OpenAI or Anthropic. - The agent is configured via GitHub Actions workflows stored in the repository's configuration files, allowing for self-evolution and auditable changes. - The workflow demonstrated includes fetching current weather data for Claremont, CA, and successfully updating a cron job schedule via a GitHub pull request. - The setup process involves scaffolding a Next.js project, running a setup wizard to configure secrets (GitHub PAT, API keys), and using ngrok to expose the local service for webhooks/Telegram access. - The agent's core functionality allows it to perform complex tasks, such as modifying its own configuration (e.g., changing the heartbeat schedule from every 10 minutes to every 30 minutes) and successfully executing those changes via GitHub commits. - The system is designed to be highly scalable, theoretically supporting hundreds of self-hosted runners interacting with various cloud services. ![Screenshot at 00:04: The initial screen transition highlights the core concept: building a free, autonomous AI agent that runs on any computer the user already owns, contrasting with expensive proprietary setups.](https://ss.rapidrecap.app/screens/8uP2IrP3IG8/00-00-04.jpg) **Context:** The video details a tutorial on setting up an autonomous AI agent, 'The Pope Bot,' that runs entirely locally using Docker containers, eliminating the need for costly hardware like a Mac Mini or recurring subscription fees for external LLM APIs. The agent is designed to be self-improving and transparent, tracking all actions via Git commits within its own repository. The demonstration focuses on the local setup process, configuration via environment variables and cron jobs, and showcasing the agent's ability to execute tasks and modify its own behavior. ## Detailed Analysis The creator presents a method for setting up a powerful, autonomous AI agent called The Pope Bot entirely locally using Docker, bypassing the need for expensive hardware (like a Mac Mini) or ongoing API fees. The architecture involves three main Docker containers: the Event Handler (EH), a Reverse Proxy/SSL container (RP SSL, likely Traefik for secure external access via HTTPS), and the Runner container responsible for executing jobs. The setup is initialized via an NPM script that scaffolds a Next.js project, configures secrets via a GitHub Personal Access Token (PAT), and sets up ngrok to expose the local development environment for webhook communication, such as with Telegram. The agent leverages GitHub for version control, meaning every action it takes is committed, ensuring complete auditability and self-evolution capability. Configuration files, like 'CRONS.json' in the config directory, control scheduled tasks, which the agent can modify and commit back to the repository itself. The demonstration confirms the agent's capability by instructing it to update a cron job schedule (changing the heartbeat from 10 minutes to 30 minutes) and observing the successful update reflected in the GitHub Actions logs and the system's notifications. The system is highly scalable, as the Runner container can theoretically handle hundreds of concurrent jobs interacting with external services (represented by cloud icons). For local development, the agent uses a specific internal Docker URL (e.g., http://host.docker.internal:13434/v2/) for API calls to local LLMs like Ollama, with free tiers available for services like Brave Search API. ### Agent Setup and Architecture - Agent runs in Docker containers (EH, RP SSL, Runner) for isolation and scalability - Local setup uses NPM scaffolding, GitHub repo creation, secret configuration (PATs, API keys), and ngrok for local tunneling. ### Agent Capabilities - Agent is self-improving, modifying its own code via Git commits; supports web chat interface and Telegram integration; can perform complex tasks like fetching weather data and updating its own configuration files. ### Configuration & Control - Job schedules (CRONs) and agent behavior are configured via files in the 'config' directory (e.g., CRONS.json, HEARTBEAT.md); all changes are tracked via GitHub commits. ### LLM Flexibility - Supports free local LLMs like Ollama, or compatible APIs like OpenAI, with instructions provided for setting local API base URLs if running inside Docker. ### GitHub Integration - Utilizes GitHub Actions for running jobs, allowing for complete visibility into job execution history ('Swarm' view) and code changes via pull requests. ![Screenshot at 00:01: Title screen introducing the concept of an Autonomous AI Agent.](https://ss.rapidrecap.app/screens/8uP2IrP3IG8/00-00-01.jpg) ![Screenshot at 00:04: The screen shows the agent running locally, promising a free setup without Mac Mini or API fees, highlighting the core value proposition.](https://ss.rapidrecap.app/screens/8uP2IrP3IG8/00-00-04.jpg) ![Screenshot at 00:19: A whiteboard diagram illustrating the Dockerized architecture: Local machine connecting to Docker containers \(EH, RP SSL, Runner\) which then interact with external cloud services.](https://ss.rapidrecap.app/screens/8uP2IrP3IG8/00-00-19.jpg) ![Screenshot at 01:13: The upgrade notification showing the transition from version v1.2.67 to v1.2.68, indicating ongoing development and maintenance.](https://ss.rapidrecap.app/screens/8uP2IrP3IG8/00-01-13.jpg) ![Screenshot at 02:28: The chat interface confirming job creation success with a Job ID and Branch ID, demonstrating the agent's immediate feedback mechanism.](https://ss.rapidrecap.app/screens/8uP2IrP3IG8/00-02-28.jpg)