Hermes Agent is Nous Research’s terminal agent. Lyceum plugs in as a custom OpenAI-compatible endpoint.
Install
The installer bundles its own Python and Node and puts hermes in ~/.local/bin. Without --skip-setup it ends in an interactive wizard that asks you to pick a provider, which the next step replaces.
Export your key, then start Hermes:
The same settings can be written from the CLI instead of editing the file:
hermes config set rewrites config.yaml with resolved values only, dropping the comments that ship in the default file. Copy it first if you want to keep them.
Verify
Run a one-shot prompt without entering the TUI:
Switch models inside a session with /model, or set a new default with hermes config set model.default <model-id>.
Notes
- Key from the environment. Hermes expands
${LYCEUM_API_KEY} at launch. Put the export in ~/.zshrc or ~/.bashrc to persist it, or store the key in ~/.hermes/.env.
- Context limits. Hermes detects the context window from the provider. Our
/models endpoint does not report it, so set model.context_length by hand from active models if history compaction triggers at the wrong point.
- Tool calling. Hermes calls a tool on nearly every turn.
moonshotai/kimi-k2.7-code and z-ai/glm-5.2 both handle its loop well.
