Skip to content

Model and configuration commands

These commands create player and provider profiles, list selectable configurations, and measure a player's image-token cost.

gptnt new player

Command syntax
gptnt new player NAME
Input Contract
NAME Required. Letters, digits, -, and _ only.
Output configs/player/<NAME>.yaml from the current player scaffold.

The command creates parent directories when needed and refuses to overwrite an existing profile. Its completion message recommends gptnt doctor.

gptnt new provider

Command syntax
gptnt new provider NAME

NAME has the same character constraint as a player name. The command writes configs/player/provider/<NAME>.yaml and refuses to overwrite an existing profile. The scaffold uses an OpenAI-compatible provider as an editable example.

gptnt list players

Command syntax
gptnt list players

The command takes no arguments. It prints sorted player profile names from configs/player/*.yaml. When provider profiles exist, it prints a separate sorted Providers group from configs/player/provider/*.yaml.

gptnt list suites

Command syntax
gptnt list suites

The command takes no arguments. It prints sorted suite names from configs/suites/*.yaml and omits underscore-prefixed templates.

gptnt measure-tokens-per-image

Command syntax
gptnt measure-tokens-per-image PLAYER CALIBRATION-IMAGE [--provider PROVIDER]
Input Contract
PLAYER Required player profile name under configs/player/.
CALIBRATION-IMAGE Required existing manual-page PNG.
--provider PROVIDER Optional provider profile override under configs/player/provider/.

The command composes the player and provider, loads the configured capabilities and agent, and resizes the PNG to the portrait orientation derived from image_dimensions. It sends the same prompt once without an image and once with the image. The difference between the reported input token counts becomes capabilities.tokens_per_image in the player YAML.

This command sends two model requests

Calibration can incur provider charges. The command preserves the agent's configured model settings and raises the output allowance to at least 256 tokens for calibration.

A zero or negative difference fails instead of updating the file. This can indicate that the provider does not include image tokens in its input-token report.