hermes.md

Hermes Guide

Hermes is Nous Research's self-improving Python agent with persistent memory, and Maritime runs its official image as a managed agent with its own dashboard.

Screen recording: Recorded live on Maritime: deploy Hermes from a template, chat with it, and open its native dashboard, from the first click to the first reply.

Overview

Hermes is a self-improving Python agent from Nous Research with persistent memory and a multi-platform messaging gateway (email, Telegram, and more). You talk to it from the Maritime CLI or the dashboard Chat tab; inbound messages on connected channels pass through a pairing wall first. Approve a contact's code, or allow them outright, under Gateway pairings on the agent's Channels tab.

Docker Image

Image: nousresearch/hermes-agent:v2026.5.16

Maritime pins the Hermes version deliberately (the latest tag drifted) and upgrades after testing each release. Gateway on port 8642, dashboard on port 9119, HOME at /opt/data.

How It Works

Maritime configures Hermes's openai provider and points its base URL at the Maritime LLM proxy, so a fresh agent works with no key at all and OpenAI-style env vars work transparently.

# The container runs with:
HERMES_HOME=/opt/data
HERMES_DASHBOARD=1
HERMES_DASHBOARD_HOST=0.0.0.0
HERMES_DASHBOARD_TUI=1
API_SERVER_ENABLED=true
API_SERVER_HOST=0.0.0.0
API_SERVER_KEY=<auto-generated>
GATEWAY_ALLOW_ALL_USERS=true
HERMES_INFERENCE_PROVIDER=openai
HERMES_TUI_PROVIDER=openai
HERMES_INFERENCE_MODEL=gpt-5.4
OPENAI_BASE_URL=https://api.maritime.sh/api/llm/v1

Environment Variables

No key is required: Maritime injects a per-agent proxy token, whose model calls are paid from your prepaid AI credits. Set your own OPENAI_API_KEY as a secret env var to pay your provider directly instead.

Sizing

Hermes loads memory + dashboard at boot; the platform sizes the agent for that automatically. Hand-tune a single agent with --ram if you need to.

Deploy

In the dashboard, open New agent and pick the Hermes Agent template, or:

maritime create my-agent --template hermes

Talk to it

Hermes has no HTTP chat endpoint of its own, so Maritime invokes it locally (hermes chat -q). Ask a question from the terminal and the agent answers, running on a serverless Firecracker microVM:

Terminal recording: Shows a Hermes agent active on a serverless Firecracker microVM, then asks it a question. The reply names Linux 4.14.174, the microVM guest kernel it runs on. Recorded against production.

Or open the agent in the dashboard and use the Chat tab. Same agent, same answer:

Screen recording: Opens a Hermes agent's Chat tab, sends a question, and the agent answers in the browser. Recorded against production.

The native Hermes dashboard

Hermes ships its own full dashboard (sessions, files, models, logs, cron, skills, plugins, MCP, channels) with the real hermes chat TUI embedded in the browser. Click Hermes Dashboard on the agent page and Maritime proxies it with auth handled for you (no login screen):

Screen recording: The native Hermes dashboard served through Maritime's UI proxy: opening the Chat view and typing a question straight into the live TUI. The agent answers with its toolbox. Recorded against production.

Upstream links & resources