Hermes Agent
Run Nous Research's Hermes Agent with its OpenAI-compatible API, using models through OpenRouter and keeping its memory on a volume.
Hermes Agent by Nous Research is an agent that learns skills and keeps memory as you use it. The Hermes Agent template runs the nousresearch/hermes-agent:stable image with its API server on, so your apps can call it like any OpenAI-compatible endpoint.
What the template creates
| Resource | What it is |
|---|---|
hermes |
The agent's gateway and API on port 8642, started with gateway run, with 512 MiB of memory and a 1 GB volume at /opt/data |
API_SERVER_KEY |
A key Dekka generates, which callers send as a bearer token |
Hermes keeps its memory, skills and settings in /opt/data, so they survive every deploy. It needs 512 MiB of memory, which the Starter plan allows.
What you provide
| Input | What it's for |
|---|---|
| OpenRouter API key | Required. Hermes reaches its models through OpenRouter. |
| Telegram bot token | Optional, to message Hermes from Telegram. |
| Discord bot token | Optional, to message Hermes from Discord. |
After it deploys
Reveal API_SERVER_KEY on the Environment tab by choosing hermes and pressing and holding the value. Your apps call the service's URL with that key as a bearer token, using the OpenAI-compatible endpoints described in the Hermes Agent docs, so most OpenAI client libraries work once you point them at it. If you added a chat token, message your bot to talk to the same agent.
Limits
The volume means the service runs a single replica. During the pilot, deploys run on a simulated engine, so Hermes doesn't actually start or answer requests.
Hermes Agent is open source under the MIT licence. OpenRouter's terms and charges apply to the model requests it makes.