156 Files, 5 Agents, Zero Babysitting: Inside My Open-Source AI Stack

#opensource #ai #claudeai #automation

I open-sourced the full automation stack that runs my AI agency — 156 files across agents, scripts, a Flask dashboard, and launchd schedulers.

→ GitHub: https://github.com/Wh0FF24/whoff-automation

What's in the repo

whoff-automation/
├── atlas-ops/          # Flask dashboard (localhost:4000/atlas)
├── atlas_lib/          # Shared utilities across all agents
├── scripts/            # 111+ production automation scripts
├── atlas-reel-composer/# Remotion video pipeline
├── atlas-voice/        # Voxtral TTS voice interface
└── content/            # Spec → draft → production pipeline

The 5 agents

Agent	Model	Domain
Atlas	Opus 4.6	CEO — orchestrates, never executes
Prometheus	Opus 4.6	Strategy, content, foresight
Hermes	Sonnet	Trading, market data
Athena	Sonnet	Revenue, Stripe, Product Hunt
Apollo	Sonnet	Research, competitor intel

Each agent runs in its own tmux pane, spawned by launchd on a schedule.

The PAX coordination layer

Agents write structured handoff files — no message queues, no APIs between agents:

[ATL->PRO] 2026-04-15 23:00
PRIORITY: HIGH
TASK: Publish 3 dev.to articles targeting GitHub traffic
CONTEXT: Repo live at https://github.com/Wh0FF24/whoff-automation
DONE_WHEN: 3 articles published, session log written to vault

Star the repo: https://github.com/Wh0FF24/whoff-automation

Atlas is the AI that runs @atlas_whoff on IG.

Tools I use: