Full beginner tutorial (zero experience needed) → docs/SETUP-FROM-SCRATCH.md Every command, copy-paste, from buying a server to texting your own AI agent that runs it for you.
The 10 steps at a glance:
- Buy a server (Hetzner) + get root access
- Connect to it over SSH
- Install Claude Code (Node + one npm command)
- Log in to your account
- Run it inside
tmux(survives disconnects)- Install the official Telegram plugin
- Create your Telegram bot with BotFather
- Control the server from your phone ✅
- Add MCP tools + skills (the agent's hands & playbooks)
- Add Cloudflare DNS + the API tokens that make the agent your "brain"
Already comfortable with Linux/Docker? Skip to Quick Start below.
Complete AI operations infrastructure for agencies. One-command deployment of:
- Paperclip — AI agent platform (agents work autonomously on tasks)
- n8n — Workflow automation (email, CRM, webhooks)
- Supabase — Self-hosted database + auth
- OpenClaw — WhatsApp & Telegram bot gateway
- ruflo — AI agent orchestration (swarms, memory, tools)
- Coolify — Deployment & management UI (optional)
External SaaS integrations (no self-hosting needed):
- GoHighLevel (GHL) — CRM central + pipeline comercial
- Gmail / Google Workspace — Mailing transaccional y comercial
- WhatsApp vía GHL AppLevel — Comunicación bidireccional directa con leads/clientes
- WhatsApp Meta API Oficial — Comunicaciones masivas (campañas, broadcasts)
Once the stack is up, this is how we actually run it day-to-day — a single Claude Code instance in a tmux session on the VPS, controlled from Telegram by the operator and approved teammates. Knowledge bases, MCP servers, deploys, scrapes, and customer comms all happen from your phone.
→ docs/CLAUDE-CODE-TELEGRAM-WORKFLOW.md — full setup, the exact start command, tmux + systemd, pairing flow for teammates, memory layers, MCP catalog, troubleshooting.
🚀 Brand new? Start here → docs/SETUP-FROM-SCRATCH.md — the complete zero-to-running tutorial: buy a Hetzner server, SSH in, install Claude Code, log in, wire the official Telegram plugin + a BotFather bot, control the box from your phone, then add MCP tools + skills and the Cloudflare/API tokens that make the agent your orchestration brain.
Once you have more than one client project, you split Claude across UNIX users (one per project) and let a mother Claude read each child's screen, message them via
tmux send-keys, and aggregate everything into a daily journal on a single web panel. Call transcripts flow into the same panel.→ docs/MULTI-CLAUDE-MOTHER-AND-CHILDREN.md — the full architecture: per-user isolation, cross-tmux orchestration, Supabase journal schema, on-demand refresh button, Fathom webhook + cron pull, panel cron list, how to add a new project.
→ docs/FLEET-ORCHESTRATION-AND-MAINTENANCE.md — keeping the fleet alive and self-updating: the orchestrator user-account pattern (bot↔bot is impossible), allowlist hot-reload + pairing, and a 3-layer cron loop (hourly liveness/revive, 8-hourly end-to-end round-trip, weekly resume-restart for auto-updates). Plus production gotchas (bun PATH, tmux send-keys, pm2 venv).
- Docker + Docker Compose
- A domain (or use sslip.io for testing)
- API keys (see
.env.example)
git clone https://github.com/DoubleN96/stratoma-ai-stack
cd stratoma-ai-stack
cp .env.example .env
# Edit .env with your values
./scripts/setup.sh
docker-compose up -d# Create your first company + agents
./scripts/create-company.sh "My Client" "client-develop"
# Health check
./scripts/health-check.shThis entire stack runs in production on a Hetzner CPX42 — best price/performance ratio for AI workloads.
| Spec | Value |
|---|---|
| vCPU | 8 cores |
| RAM | 16 GB |
| Disk | 320 GB SSD |
| Bandwidth | 20 TB/month |
| Price | €19.49/mo (€0.031/h) |
| Location | EU (Nuremberg / Helsinki / Falkenstein) |
→ Get €20 free credit on Hetzner
Why Hetzner:
- Cheapest EU cloud with enterprise-grade SSD
- Snapshots + backups built-in via Coolify
- Low latency from Spain/EU (vs AWS/GCP)
- CPX42 handles: Paperclip + n8n + Supabase + OpenClaw + all containers simultaneously
Minimum recommended: CPX31 (4 vCPU / 8 GB / 160 GB — ~€11/mo) for lighter setups.
┌──────────────────────────────────────────────────────────────────┐
│ Stratoma AI Stack │
│ │
│ ┌──────────┐ ┌──────────┐ ┌────────────────┐ │
│ │Paperclip │ │ n8n │ │ Supabase │ │
│ │ :3100 │ │ :5678 │ │ :8000 │ │
│ │ Agents │ │Workflows │ │ DB + Auth │ │
│ │ Skills │ │ Webhooks │ │ Storage │ │
│ └────┬─────┘ └────┬─────┘ └───────┬────────┘ │
│ │ │ │ │
│ ┌────▼──────────────▼────────────────▼────────┐ │
│ │ OpenClaw :18789 │ │
│ │ Telegram Gateway (bots personales/equipo) │ │
│ └──────────────────────────────────────────────┘ │
│ │
│ MCPs disponibles para todos los agentes: │
│ ruflo | n8n-mcp | github | coolify | supabase │
└──────────────────────────────────────────────────────────────────┘
↕ Integraciones externas (SaaS)
┌──────────────────────────────────────────────────────────────────┐
│ Comunicaciones & CRM │
│ │
│ ┌─────────────────────┐ ┌──────────────────────────────────┐ │
│ │ GoHighLevel (GHL) │ │ Gmail / Google Workspace │ │
│ │ │ │ │ │
│ │ · CRM & Pipeline │ │ · Email comercial (mailing) │ │
│ │ · Contactos │ │ · Email transaccional │ │
│ │ · Automatizaciones │ │ · Calendarios y Drive │ │
│ │ · Seguimiento │ │ · Integración via GWS CLI │ │
│ └──────────┬──────────┘ └──────────────────────────────────┘ │
│ │ │
│ ┌────────┴──────────────────────────┐ │
│ │ WhatsApp (dos canales) │ │
│ │ │ │
│ │ ┌────────────────────────────┐ │ │
│ │ │ AppLevel vía GHL │ │ │
│ │ │ · Comunicación DIRECTA │ │ │
│ │ │ · Bidireccional 1:1 │ │ │
│ │ │ · Respuestas IA en tiempo │ │ │
│ │ │ real (n8n + OpenClaw) │ │ │
│ │ └────────────────────────────┘ │ │
│ │ │ │
│ │ ┌────────────────────────────┐ │ │
│ │ │ Meta API Oficial │ │ │
│ │ │ · Comunicaciones MASIVAS │ │ │
│ │ │ · Campañas & broadcasts │ │ │
│ │ │ · Templates aprobados │ │ │
│ │ │ · Alta deliverability │ │ │
│ │ └────────────────────────────┘ │ │
│ └───────────────────────────────────┘ │
└──────────────────────────────────────────────────────────────────┘
| Canal | Plataforma | Uso | Dirección |
|---|---|---|---|
| Email comercial | Gmail / GWS | Comunicación con leads y clientes | Bidireccional |
| Email masivo | Gmail + n8n | Campañas, newsletters | Saliente |
| WhatsApp directo | GHL AppLevel | Conversaciones 1:1 con leads | Bidireccional |
| WhatsApp masivo | Meta API Oficial | Broadcasts, campañas, notificaciones | Saliente |
| Telegram | OpenClaw | Notificaciones internas del equipo | Bidireccional |
AppLevel vía GHL — para conversaciones directas:
- Conectado al CRM (GHL): historial de conversación por contacto
- Los agentes de Paperclip pueden leer y responder vía n8n
- Ideal para: seguimiento de leads, responder dudas, cerrar ventas
Meta API Oficial — para comunicaciones masivas:
- Templates aprobados por Meta (alta tasa de entrega)
- Broadcasts a listas segmentadas desde GHL
- Ideal para: campañas de nurturing, recordatorios, lanzamientos
| Servicio | Puerto | URL |
|---|---|---|
| Paperclip | 3100 | https://paperclip.yourdomain.com |
| n8n | 5678 | https://n8n.yourdomain.com |
| Supabase Kong | 8000 | https://db.yourdomain.com |
| OpenClaw | 18789 | Interno |
| Supabase Studio | 3000 | https://studio.yourdomain.com |
./scripts/create-company.sh "Client Name" "client-slug"This will:
- Create the company in Paperclip
- Create default agents (CEO, Engineer, Sales, Marketing)
- Install standard skills (n8n, ruflo, GWS, etc.)
- Create the client project directory with MCP config
Every agent gets access to these MCPs by default (via /paperclip/stratoma-default/.mcp.json):
- ruflo — 60+ AI orchestration tools
- n8n-mcp — Create/edit n8n workflows
- github — Repository management
- coolify — Deploy applications
- supabase — Database operations
Ver n8n/workflows/ — 16 workflows listos para importar:
- Monitor commercial email + AI categorization + Telegram approval
- CRM → AI agent bridge in real time
- AI replies on WhatsApp (via CRM provider)
- Telegram correction bot
- Inbound listings scraping
- Bookings + marketplace email automation
- Automatic check-in
- Call transcript provider → CRM sync
- Error workflow for the technical team
- Template error handler (reusable for any client)
Issues and PRs are welcome! Some ideas for improvements:
- ARM64 / Apple Silicon support for local dev
- Kubernetes / Helm chart version
- One-click deploy button for Coolify / Railway / Render
- More workflow templates (real estate, e-commerce, SaaS)
- OpenClaw config wizard (interactive setup)
- Monitoring stack (Grafana + Prometheus)
- Backup automation scripts
Open an issue to share your setup, report bugs, or suggest features.
MIT
The stack ships with a reproducible set of Paperclip skills and a default agent roster.
paperclip/skills/catalog.yaml— external skills pulled from GitHub + skills.sh. Covers Google Workspace, n8n, marketing/SEO, Next.js, Supabase, Anthropic doc/pdf tooling, and Paperclip meta-skills.paperclip/agents/roster.yaml— 9 agents (CEO, Engineer, Marketing/SEO, Sales Manager, Sales Rep, Lead Qualifier, Follow-up, CRM Updater, Admin). Each agent is pre-wired to a sensible subset of skills for its role.
cp .env.example .env
# Fill in PAPERCLIP_ADMIN_API_KEY + other vars
./scripts/setup.sh # docker-compose up
./scripts/create-company.sh "Client Name" "client-slug"
# Creates company → installs skills from catalog.yaml → seeds 9 agents from roster.yamlRequires yq, jq, curl on the host.
- Add new skills: append to
paperclip/skills/catalog.yamland re-run./scripts/install-skills.sh <company_id>. - Add/remove agents: edit
paperclip/agents/roster.yamland re-run the relevant section ofcreate-company.sh. - Per-company AGENTS.md instructions (credential blocks, runbooks, tone) are uploaded separately via
PUT /api/agents/{id}/instructions-bundle/file.