Hermes Agent + DeepSeek V4 Pro: n8n bot tutorial

This tutorial uses Hermes Agent with DeepSeek V4 Pro to create an n8n customer support workflow through chat. The presenter deploys Hermes through a Railway template, adds a DeepSeek API key, connects Telegram, and supplies the URL and access token for n8n's instance-level MCP server. The demonstrated setup uses hosted services rather than showing how to run models locally.

After installing the Supabase agent skill, the presenter asks Hermes to create orders, FAQ, products, and ticket history tables, then populate them with sample data. The requested support workflow routes Telegram and Slack messages to an AI agent with memory, SQL tools, FAQ vector search, and Perplexity web search as a fallback.

The generated workflow needs a correction: the presenter identifies an embedding-node problem and asks Hermes to fix it. Hermes reports a version issue, and the presenter shows the corrected OpenAI embedding node before assigning credentials and publishing. The final Telegram test asks about warranties and bulk pricing; the presenter reports correct answers from Supabase vector search. The transcript does not show a comparable Slack test.

The speaker says Hermes reuses logs of earlier workflow tool calls to improve later builds. Claims about DeepSeek's 1.05 million token context, Opus 4.6 parity, and cache savings are the speaker's claims, not independently verified results here.