Command line
The commands as ignis-server --help prints them. Every field’s flag is in every config field; ignis-server help --fields --format json gives the same table to a tool.
ignis-server: the OpenAI-compatible HTTP entrypoint
USAGE: ignis-server [--config <path>] [--profile <name>] [FIELD FLAGS] ignis-server help [--fields [--format text|json]] ignis-server version ignis-server config generate [--format json|yaml] [--out <path>] [--force] [--dry-run] [FIELD FLAGS] ignis-server config print [--file <path>] [--format json|yaml] [FIELD FLAGS] ignis-server config patch [--file <path>] [--out <path>] [FIELD FLAGS] ignis-server model download [<id>...] [--all] [--out <dir>] [FIELD FLAGS] ignis-server model list [--format text|json] [FIELD FLAGS]
Every field has a flag (--<group>-<field>), an env var (IGNIS_<GROUP>_<FIELD>)and a config-file key (<group>.<field>). Precedence, highest first: flag, envvar, config file, profile, built-in default; a field's qwen38 / qwen38flashnextvalue wins within each. `help --fields` prints every field in full.
--profile defaults to `auto`: the built-in profile of the card's total memory(16 GB: rtx50-16gb, 32 GB: rtx5090, 96 GB: rtxpro6000); a card of no class is refused, naming --profile. A named profile(or `none`) always wins.