brain/cli/models.py

"""Lists the models the API actually offers today.

    python -m brain.cli.models

Model names are settings, never logic (CLAUDE.md), and Gemini renames and retires them often.
This is the command that tells you what to put in GEMINI_MODEL — the web app gets the same
list through the brain's `/models` route in P3, which is what makes a renamed model a settings
change rather than a redeploy.
"""

import asyncio
import os

from rich.console import Console

from .. import config
from ..llm.gemini import GeminiClient
from ..llm.keys import KeyRing


async def main_async() -> int:
    console = Console()
    settings = config.load_settings()
    ring = KeyRing.from_env(dict(os.environ))
    if not ring.configured_slots():
        console.print("[red]No GEMINI_KEY_1 in the environment.[/red] See backend/.env.example.")
        return 1

    async with GeminiClient(ring, settings) as client:
        models = await client.list_models()

    console.print(f"[bold]{len(models)} models[/bold]\n")
    for name in models:
        marker = ""
        if name == settings.chat_model:
            marker = "  [green]← GEMINI_MODEL[/green]"
        elif name in settings.fallback_models:
            marker = "  [dim]← fallback[/dim]"
        elif name == settings.embedding_model:
            marker = "  [cyan]← embeddings[/cyan]"
        console.print(f"  {name}{marker}")

    configured = set(settings.model_chain) | {settings.embedding_model}
    missing = sorted(configured - set(models))
    if missing:
        console.print(
            f"\n[yellow]Configured but not offered by the API:[/yellow] {', '.join(missing)}\n"
            "Set GEMINI_MODEL in backend/.env (or the fallbacks in brain/config.py) to ids from "
            "the list above."
        )
    return 0


def main() -> int:
    return asyncio.run(main_async())


if __name__ == "__main__":
    raise SystemExit(main())