Quickstart: Hermes Agent

# ~/.hermes/config.yaml
model:
  provider: custom
  base_url: https://uncensoredopencode.com/v1
  default: glm-5.3-flash-uncensored

# ~/.hermes/.env
OPENAI_API_KEY=sk-nc-...

Or run hermes model, choose the custom endpoint, and paste the same base URL, key and model id.

What to know for a long run

• An agent that runs unattended should have its own key with a cap on it, so a loop that goes wrong stops at a figure you chose. See Spend caps.
• Every model here holds about a million tokens, so a long session keeps its early tool results in view. The exact figure for each is on Models.
• Streams carry a keep-alive every 15 seconds while a model is thinking, so a slow step is not mistaken for a dead connection. A stream that sends nothing at all for 120 seconds is closed.
• Turn on the low balance email under Billing before leaving it overnight. A request that needs more credit than the account has is refused with a 402, and repeated refusals pause the account until credit is added.
• If the agent sets a small output limit, a model that reasons can spend it all thinking and return nothing. See Tool calling.