Models
The catalog belongs to the active provider. Codex and Grok each read their own catalog from the subscription, so the list changes with the provider and with the plan behind it. On Codex the default is gpt-5.6-luna when the plan lists it; otherwise, and on Grok, the default is the first model the subscription lists.
Before you sign in, the catalog is empty. nod models lists every provider, marks the ones that are signed out or unreachable, and says why.
Browse and select
List models outside the shell:
nod models
nod models --json
Inside nod, /model alone opens the catalog; give it an ID or a search query to select directly:
/model
/model gpt-5.4
The selected model and reasoning effort are saved as user defaults in ~/.nod/settings.json. The model is stored per provider, so switching providers and switching back restores what you had.
Selection order
The first available source wins:
NOD_MODELfor the current process- the user default for the active provider in
~/.nod/settings.json - the provider default
Project .nod.json files cannot set model, effort, or fast mode. This prevents a repository from silently changing your model preferences.
nod status
Use nod status or /status to confirm the effective model.
Temporary overrides
Override one process without changing your saved default:
NOD_MODEL=gpt-5.4 nod
NOD_PROVIDER=grok does the same for the provider. ACP has an equivalent flag:
nod acp --model gpt-5.4
Reasoning effort and fast mode
Effort is saved as a user preference and accepts auto, none, minimal, low, medium, high, xhigh, and max; auto lets the provider decide. /fast toggles fast mode where the selected model supports it. The setting is saved as a user preference.
Controls depend on the model
Reasoning effort and fast mode appear only when the selected model supports them.