DEV Community

AgentChip
AgentChip

Posted on

I Automated the Entire AI Agent Hosting Business — Zero Human Intervention

#ai #saas #automation #devops

The Setup

I run a fully automated AI agent hosting service. Customers pay via PayPal, get a Docker container with Hermes Agent pre-installed, receive their API key via email, and start using it — all without any human involvement.

Here is the entire pipeline:

Customer pays (PayPal)
  → Webhook fires to our server
  → Python script creates Docker container
  → New-API generates dedicated API token
  → Gmail SMTP sends credentials to customer
  → Customer opens WebUI and starts chatting
Enter fullscreen mode Exit fullscreen mode

Time from payment to working agent: ~90 seconds.

The Stack

  • Hermes Agent (Nous Research) — open-source AI agent framework
  • DeepSeek API — LLM backend (Flash model, ~90% cache hit rate)
  • New-API — token management and rate limiting
  • Docker — customer isolation (one container per customer)
  • PayPal — subscription billing with webhooks
  • Caddy — reverse proxy with automatic HTTPS

The Economics

DeepSeek Flash model costs ~$0.008 per million tokens with 90% cache hit rate. At $23.99/month for 100M tokens, the margin is ~89%.

Key insight: Use the Flash model, not the Pro model. Flash is 10x cheaper and good enough for most use cases. We disabled Pro access entirely to prevent losses.

What Is Automated

Everything:

  • ✅ Payment processing (PayPal subscriptions)
  • ✅ Container provisioning (Docker)
  • ✅ API key generation (New-API)
  • ✅ Email delivery (Gmail SMTP)
  • ✅ Customer support (AI agent, no humans)
  • ✅ Usage tracking (customer portal)

Lessons Learned

  1. Docker isolation > virtual environments — each customer gets their own container with resource limits (0.5 CPU, 256MB RAM)
  2. Webhook reliability matters — PayPal webhooks occasionally fail; implement retry logic
  3. Email deliverability — Gmail SMTP works but lands in Promotions tab; consider dedicated email service for production
  4. Cache hit rate is everything — DeepSeek caching saves 10x on costs; design prompts to maximize cache hits

Try It

If you want a managed AI agent without the setup hassle: AgentChip

$23.99/month, 100M tokens included. Deploy in 10 minutes.

Top comments (0)