Skip to main content

Hermes

Hermes is a self-hosted AI agent layer for work that needs memory, tools and a separate runtime. We use it when one agent must stay isolated from another agent, from the CRM, and from the rest of the system.

What Hermes is for​

  • one agent = one role
  • separate memory, skills and sessions
  • separate bot or token per agent
  • no mixing of tasks and data
  • a repeatable way to deploy autonomous AI workers

How installation works​

  1. Pick the agent's job.
  2. Create a separate Hermes container or profile.
  3. Connect its Telegram bot or another channel.
  4. Mount the vault or project folder it should read.
  5. Set access limits and test the first task.

How to use it correctly​

  • give one clear task at a time
  • include the context and the expected result
  • do not give write access unless the workflow needs it
  • store reusable prompts and procedures as skills
  • keep the agent narrow; broaden it later

When it makes sense​

  • support or ops assistant
  • content or workflow assistant
  • internal automation worker
  • a separate bot for a separate project
Project name

In our internal notes this deployment is called Hermes. The implementation note lives in Obsidian as Deployment Hermes.