Skip to main content
Hermes Agent is an AI agent that learns from use: it keeps memories, sessions and skills, runs scheduled tasks and answers you in a chat or in the messaging apps you connect. This guide takes you from the provider key to talking with your agent. The Hermes Agent template packages Hermes Agent v0.22.5 with a web panel, an optional browser chat and channels, behind a single login.
Vertra Cloud does not host a model. You bring the key of a provider, and the model usage is billed to you by that provider. This template is a Vertra Cloud packaging and is not affiliated with the Hermes Agent project (MIT license).

​
Requirements and plan

​
1. Get the provider key

Hermes works with OpenRouter (default), Anthropic, OpenAI, Google Gemini or any OpenAI-compatible endpoint. Copy the key now: most providers show it only once. It is stored only in the app’s encrypted variables.

​
2. Create the agent

1

Open the template

In the dashboard open Templates, pick Hermes Agent and click Use template.
2

Choose provider and model

Pick the model provider and paste the API key. With a compatible endpoint, also enter the base URL and the model, which is required. For the others, leave the model empty to use the provider’s default, or type the model you prefer.
3

Browser chat (optional)

Turn on Browser chat (Hermes WebUI) if you want a chat page with sessions, files and settings. It costs about 150 MB of memory more. Without it the agent still works in the panel and in your channels.
4

Connect channels (optional)

Each channel asks for credentials and the list of people allowed to talk to the agent, which is required. See step 4; you can also do it later in the panel.
5

Create

Name the app and click Create. The outputs show the panel address, the user admin and a generated password, shown once. Copy it.

​
3. Sign in to the panel and chat

Open <address>/admin/. The browser asks for a user and a password: use admin and the password from the outputs. This single login protects the panel and the chat.
  • Chat in the panel: /admin/chat is the Hermes terminal chat. Say hello there first to check that the key works.
  • Chat in the browser: if you turned it on, it lives at <address>/ with the same login. Without it, / takes you to the panel.
  • Lost the password? Change the HERMES_ADMIN_PASSWORD variable in the app and restart.
The first boot takes 15 to 30 seconds while Hermes syncs its skills. Later starts take about 15 seconds. Sharing usage metrics with the Hermes project is off by default: the template answers the panel’s “Help improve Hermes?” question with “No thanks” at the first start, so the notice does not appear. If you turn it on later in the panel, your choice is kept.

​
4. Connect a channel

Each channel needs two things: the bot credential and who may talk to the agent. Hermes ignores everyone who is not on that list, so the agent never answers strangers. You can fill these in the wizard or later in Channels (/admin/channels).
  1. In Telegram, talk to @BotFather, send /newbot and follow the questions. Copy the bot token.
  2. Find your numeric id by talking to @userinfobot.
  3. Enter the token and your id as the allowed user.
  4. Open a chat with your bot and send a message.
Signal, Matrix, Mattermost and other channels are not in the wizard. Configure them in /admin/channels or in the app variables. Signal needs an external signal-cli daemon and does not run in this container.

​
5. Change the model later

Provider and model are applied on every start from the app variables, so the variables are the source of truth. To change them, edit these variables in the app and restart:
Changing provider or model inside config.yaml does not last: the file is overwritten from the variables on the next start. Change the variables.

​
Skills, memory and schedules

  • Skills. Hermes syncs its skills on the first boot; you manage them in the panel.
  • Memory. Memories, sessions, skills and configuration are kept in /app/.hermes on the app’s disk, so they survive restarts, updates and rollbacks.
  • Schedules. In Cron you create recurring tasks. The agent runs them and talks to you through the channel you choose.

​
The cost of the AI is yours

The platform charges the plan for the memory the app uses. The model calls are billed by your provider, and an agent that runs scheduled tasks or reads long files can use many tokens. Set a spending limit in your provider’s account, and watch the usage while you test.

​
Update and roll back

New versions arrive as new versions of the template. A snapshot is taken before the update, and a rollback brings the state back (/app/.hermes). See Update and roll back. Zero-downtime deploy is off: a second copy on the same disk exits with an error.

​
Limits to know

  • Caddy is the only gate, with no attempt limit. The generated password is long; for more protection add a rate-limit rule on your own domain.
  • The generated password and the provider key are in the app’s variables: do not share screenshots of them.
  • The email channel and each provider’s default models were not exercised with a real account or key, so test them before depending on them.

​
Common problems

That is the login. Use admin and the password from the outputs, or change HERMES_ADMIN_PASSWORD and restart.
Check the memory (1.5 GB minimum) and the app log. HERMES_ADMIN_PASSWORD is required means the password variable is missing.
Open the app log. An invalid token appears there (“token … was rejected”) and the app stays up with that channel parked. Do not share this log. Then check that your own id is on the allowed list.
Provider and model are reapplied from the variables on every start. Change the variables and restart.
Reload Channels > WhatsApp and scan right away. The first time takes longer because the bridge is downloaded.

Hermes Agent in the catalog

What the template creates and its variables.

Update and roll back

The safety snapshot and rollback.

Public page

The template in the public gallery.

Use a template

The gallery, the wizard and the outputs.
Last modified on October 28, 2026