Why self-host Hermes Agent on a VPS
An AI agent that learns from you raises an obvious question: where does that memory live? With a proprietary cloud agent, your history, your preferences and the "model of who you are" that the agent builds are stored on the vendor's side. Hermes Agent takes the opposite stance: open source (MIT) and self-hostable, it keeps all its memory and skills within your own infrastructure (the ~/.hermes volume). On a VPS, the agent runs around the clock, keeps learning, and you remain the sole owner of what it has learned. You plug in the LLM provider of your choice and avoid the dollar subscription of managed assistants.
The concrete benefits of a self-hosted Hermes Agent
- Sovereign memory: all of the agent's learning stays on your VPS, never with a third party.
- Self-improvement: Hermes creates and refines skills from experience.
- Lightweight: the project is designed to run on small configurations, down to an entry-level VPS.
- LLM provider of your choice: remote API or local model, you decide.
- MIT open source: auditable code, no dependency on a closed platform.
Build from source and LLM configuration
Unlike applications shipped as a ready-to-pull image, Hermes Agent is built from its official repository: you clone the project, build the Docker image, then start the stack with Docker Compose. Plan for 4 GB of RAM so the build goes smoothly. Configuration goes through a .env file (copied from .env.example) where you enter the LLM provider and the model key. The stack exposes two services: the gateway, which runs the agent, and a local dashboard to drive it. The dashboard listens locally on the VPS — you make it accessible through a reverse proxy.
Deploy Hermes Agent with Docker
Prepare the VPS
On an Ubuntu 22.04 VPS with 4 GB of RAM, install Git, Docker and Docker Compose. This amount of memory covers the image build and the running of the agent and its dashboard.
Clone and build
Clone the nousresearch/hermes-agent repository, copy .env.example to .env, then run docker compose up -d --build. The first build takes a few minutes.
Configure the LLM provider
In .env, enter the provider and the model key you want to use (remote API or local endpoint). Restart the stack to apply the configuration.
Expose the dashboard
The Hermes dashboard listens locally (127.0.0.1): put an Nginx or Caddy reverse proxy with SSL in front of it to access it from your domain over HTTPS.
Back up the memory
All of the agent's memory and skills live in the ~/.hermes volume. Back up this directory regularly so you never lose the accumulated learning.
The value of Hermes Agent grows over time: the more it exchanges with you, the richer its memory and skills become. Treat the ~/.hermes volume as precious data — include it in your automated backups just like a database, and mount it on a persistent volume so it survives redeployments.
The official documentation
For advanced configuration and tool-specific options, refer to the official Hermes Agent documentation. This guide covers going live on a VPS; the vendor's docs remain the reference for fine-tuning, major updates and specific use cases.
