Hermes + n8n Automation Stack

One-liner: A self-hosted market-intelligence agent and automation layer running on a personal VPS — proof that Alex builds infrastructure, not just apps.

Origin (the Build Loop)

Because I wanted a steady read on markets and opportunities without babysitting a dozen sources, I kept hitting the fact that the useful signal was scattered across the web and nobody was going to aggregate it for me, so I built Hermes — a market-intelligence pipeline on my own VPS that searches, summarizes, and delivers, that posts digestible summaries straight to where I’ll actually read them. It taught me to run and maintain always-on infrastructure — the unglamorous part of “it just works.”

The problem

Most “AI agent” projects live on someone else’s platform and die when the free tier ends. I wanted something I actually own and control: an agent that runs on my own box, pulls current information through real search tooling, and pushes summaries to me on a schedule — plus a broader automation layer for the other recurring jobs in my life.

What it does

• Hermes — a market-intelligence pipeline that runs on a schedule, searches the web via MCP-connected search providers, summarizes findings, and delivers them to a channel I read daily.
• n8n automation layer — the broader workflow platform on the same VPS running additional jobs, including a job-market scraper and newsletter automation.
• Human-in-the-loop review — summaries flow to a connected surface (e.g. a Google Sheet) so a non-technical partner can review them without touching the pipeline.

How it’s built

• Stack: n8n (self-hosted workflow orchestration) on a Servarica VPS; MCP servers for web search (Brave Search / Tavily); delivery to Discord (migrated off Telegram); Google Sheets for human review.
• Notable engineering:
  • Self-hosted and maintained — provisioning, uptime, and migration between delivery channels (Telegram → Discord) are all owned end to end.
  • MCP-server integration for tool-use, the same pattern used across Alex’s other AI work.
  • Multi-channel delivery (Slack + Discord for the newsletter automation) and a spreadsheet review loop for non-technical stakeholders.
• Architecture: scheduled n8n trigger → MCP search tools → summarize → deliver to Discord/Sheet. Additional independent workflows (job-scraper, newsletter) run on the same instance.

Proof points

• Always-on, self-hosted on owned infrastructure — not a hosted-platform toy.
• Multiple production workflows on one VPS (Hermes + job scraper + newsletter).
• Real migration executed in production (Telegram → Discord) without losing the pipeline.
• Human-review loop makes the output usable by a non-engineer.

What to show

• Demo: None — it’s backend infrastructure on a private VPS. Case-study-only.
• Visuals needed: a screenshot of the n8n workflow canvas (nodes visible); an example Discord summary post; a simple diagram (schedule → MCP search → summarize → deliver). Redact any keys/endpoints.

Cross-links

• Shares the MCP + AI + scheduled delivery DNA with POLYBOT (n8n workflow, Discord) and the job-market scraper that also surfaces in JOB-HUNTER.
• Part of the off-repo infrastructure story with Home Infrastructure and Security Research.

Case-study angle

The project that proves “builder” means infrastructure too: Alex stood up and maintains an always-on, self-hosted AI agent on his own VPS — provisioning, tool integration, channel migration, and a human-review loop included.