ThePepeBot

A free, open-source autonomous AI agent by Stephen G. Pope designed as an OpenClaw alternative. Runs 24/7 on any computer using free local LLMs via Ollama. No API fees, no Mac Mini required. All changes tracked via git with optional human approval.

Architecture

Three Docker Containers

  1. Event Handler — the core bot; starts jobs, receives notifications, processes messages
  2. Reverse Proxy — handles SSL certificates for remote deployment (Ngrok for local, auto-generated for Digital Ocean)
  3. Runner — executes jobs via GitHub Actions workflows; scalable to hundreds of parallel agents

GitHub as Orchestration Layer

All jobs run as GitHub Actions workflows. Benefits:

  • Change approval: Every agent modification creates a PR — review and approve, or set to auto-merge
  • Full transparency: Every action, thought process, and code change tracked in git
  • Scalability: Run locally (faster) or on GitHub’s infrastructure (2,000 free hours)
  • Configuration via files: Heartbeat, cron jobs, secrets, and settings all version-controlled

Key Features

  • Heartbeat: Scheduled tasks on configurable intervals (like loop but self-hosted)
  • Swarm: Dashboard showing all running jobs with status, history, and pagination
  • Web chat interface: Streaming chat UI similar to ChatGPT/Claude
  • API access: REST API for external integrations (Postman, webhooks)
  • Auto-upgrade: One-click system updates that rebuild under you
  • Telegram integration: Optional messaging gateway
  • Brave Search: Free web search capability
  • Attachments: PDF and image analysis in chat

LLM Configuration

Designed for free local models via Ollama, but supports any OpenAI-compatible API. The internal Docker URL for Ollama is provided for containerized setups.

Compared to OpenClaw

OpenClawThePepeBot
CostAPI fees + Mac Mini recommendedFree (Ollama + your hardware)
LLMCloud APIs requiredLocal models via Ollama (or any API)
Change trackingLimitedFull git history with PR approval
Job executionInternalGitHub Actions (scalable, free tier)
SecurityKnown vulnerabilitiesChanges sandboxed in Docker, approved via git
Multi-platform messagingTelegram, Discord, Slack, WhatsApp, SignalWeb chat + Telegram (more planned)
Self-improvementNoAgent can review its own logs and improve (via heartbeat)

Scalability Design

The three Docker containers can run together (simple) or on separate servers (scalable). Multiple cloud servers as runners enable hundreds of parallel agents. Configuration controls auto-merge scope — e.g., auto-merge log-only changes, require review for code changes.

Roadmap

  • More chat platforms (Slack, Discord)
  • Skill self-creation examples
  • More powerful local GPU setups for running frontier-class models
  • Community sharing of agent configurations

See Also