Hermes Agent

Hermes Agent is an autonomous AI agent built by Nous Research. It features persistent memory, 40+ built-in tools, cron scheduling, and multi-platform messaging support.

Quick Start

1. Deploy this template — pick a dashboard domain (e.g. my-hermes). Optionally supply your LLM provider now by setting LLM_API_KEY, LLM_BASE_URL, and LLM_MODEL (any OpenAI-compatible provider); leave them blank to configure the model later.
2. Open the dashboard — go to your domain (auto-bound to the dashboard). Add a custom domain later in the Networking tab if you like.
3. Log in — username admin, password = your service PASSWORD (shown in the service instructions). Start chatting in the browser — no Telegram needed.

Prefer another provider or Telegram? Set your OpenAI-compatible provider variables (LLM_API_KEY / LLM_BASE_URL / LLM_MODEL) and add any messaging environment variables after deploy — see Models & Providers and Messaging Platforms below.

Ports & Domains

This service exposes two HTTP ports:

Port	Id	Purpose	Domain
9119	web	Web dashboard (on by default, login-protected)	The service's main domain — open it and log in
5000	api	OpenAI-compatible API server	Bind a domain only if you need external API access

Messaging platforms (Telegram, Discord, Slack, etc.) use outbound connections — they do not require any public URL or open port.

Using the Web Dashboard

The dashboard is enabled by default and protected by login. Open your service's domain (the web port) and log in:

• Username: admin
• Password: your service PASSWORD value (shown in the service instructions)

To disable it, set HERMES_DASHBOARD=false in the Variables tab. To change the credentials, set HERMES_DASHBOARD_BASIC_AUTH_USERNAME / HERMES_DASHBOARD_BASIC_AUTH_PASSWORD.

Security note: Basic auth login is best for a backend behind a trusted network; for hardened public exposure use OAuth (HERMES_DASHBOARD_OAUTH_CLIENT_ID) or self-hosted OIDC (HERMES_DASHBOARD_OIDC_*) — see the environment variables reference.

Using the OpenAI-compatible API

Bind a domain to the api port (5000) in the Networking tab, then send requests with your auto-generated API key (shown in the service instructions):

curl https://YOUR_API_DOMAIN/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_SERVER_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "default", "messages": [{"role": "user", "content": "Hello!"}]}'

Models & Providers

This template runs against any OpenAI-compatible provider you bring. Set these in the Variables tab and restart:

• API key — set LLM_API_KEY to your provider's key.
• Endpoint — set LLM_BASE_URL to the provider's OpenAI-compatible base URL, e.g. https://api.openai.com/v1. If you set a key but leave the base URL empty, the agent will error until you provide one — this is intentional so misconfiguration is visible.
• Model — set LLM_MODEL to the model id, e.g. gpt-4o-mini.
• Another provider (Minimax, Kimi, GLM/Zhipu, Gemini, NVIDIA, xAI, …) — set LLM_API_KEY / LLM_BASE_URL / LLM_MODEL for that provider, or add that provider's own API key env var (e.g. KIMI_API_KEY, MINIMAX_API_KEY) and pick the model with hermes model in the Exec tab. See the Configuration guide and Environment Variables reference.

When LLM_API_KEY is set, LLM_API_KEY / LLM_BASE_URL / LLM_MODEL are applied to the model block of /opt/data/config.yaml on every start (provider custom). Leave LLM_API_KEY unset to manage config.yaml by hand instead.

Messaging Platforms

Messaging is optional — by default you interact through the web dashboard (no platform needed). To connect Telegram, add TELEGRAM_BOT_TOKEN (from @BotFather) and TELEGRAM_ALLOWED_USERS (your Telegram user ID from @userinfobot) in the Variables tab, then restart. Hermes supports 14+ platforms (Discord, Slack, WhatsApp, Signal, Matrix, Email, WeChat, WeCom, Feishu, DingTalk, QQ Bot, Mattermost, BlueBubbles/iMessage, SMS, Home Assistant, generic webhooks) — each is enabled by setting its environment variables after deployment.

See the official Messaging Gateway guide for per-platform setup.

Configuration

• LLM providers, tools, and all other settings — refer to the official Configuration guide and Environment Variables reference.
• Edit config files via the Zeabur Files tab at /opt/data/:
  • config.yaml — model selection, terminal backend, platform settings
  • SOUL.md — agent personality and system prompt
  • .env — all environment variables
• After editing, restart the service to apply changes.

CLI Access

You can use the Hermes CLI directly via the Zeabur Exec tab (or zeabur service exec). The hermes command is available out of the box:

Command	Description
hermes	Interactive chat (REPL)
hermes --tui	Modern terminal UI
hermes -z "prompt"	One-shot query (for scripting)
hermes status	Show status of all components
hermes model	Switch model / provider
hermes config	View or edit configuration
hermes sessions list	List past conversations
hermes -c	Resume last conversation
hermes cron	Manage scheduled jobs
hermes skills	Browse and install skills

Run hermes --help for the full command list.

Notes

• The Docker image is ~0.9–1.0 GB compressed (download); first pull may take a few minutes. First deployment usually pulls within a minute; later deploys reuse the cached image.
• First boot may take 30–60 seconds as the agent initializes config files and syncs skills.
• Using browser tools may cause high memory usage — if you encounter OOM (out of memory) errors, consider upgrading to a larger instance (recommended 4 vCPU / 8 GB RAM).

License

Hermes Agent is licensed under the MIT License by Nous Research.