nc://model-presets

Model Presets: named model, provider and reasoning-effort picks, landed with one /preset or one tool call. A Nodecode rewrite of will-lynas/hermes-model-presets by will-lynas.

A name you keep saying — the small fast one, the one that thinks hard, the one on a subscription rather than a key — is a preset. Write it once in the config, and moving a session onto it is one short line instead of a model id, a provider and a reasoning rung. The original plugin did this by rewriting the /model line at the harness gateway; this cell hands the pick to the gateway's own SELECT-MODEL and SELECT-EFFORT, so a preset switch pins the session, checks the rung against that model's ladder, makes the pair the default a new session starts on, and every attached shell's footer follows — the same thing a /models pick does.

Get it

Tell the agent "add the model-presets cell", or run nodecode add nc://model-presets.

Set it up

No key, no network, nothing required: presets are names you choose. Name them under model-presets in ~/.nodecode/config.jsonc (or in /settings, as the same object), one object a preset:

"model-presets": {
  "presets": {
    "fast": { "model": "claude-haiku-4.5", "provider": "anthropic" },
    "deep": { "model": "gpt-5.4", "provider": "openai", "effort": "high" }
  }
}

model is the id a /models pick would take and provider the lane that serves it — both required. effort is optional: one of off, minimal, low, medium, high, xhigh, max, and left out means the model's own reasoning. The original plugin's words are read too: aliases instead of presets, reasoning_effort instead of effort. A preset of the wrong shape refuses at start and says which member is wrong rather than being dropped silently.

config.example.jsonc beside this file is the whole section, commented.

Use it

/preset                     the presets, as a picker; a row switches this session
/preset list                names, models, providers and efforts, one a line
/preset status              what this session is pinned to now
/preset fast                switch this session to `fast`
/preset fast --effort high --provider openrouter    one pick's overrides

The agent can do the same from a turn, which is what makes this worth having in a conversation: "put this session on my deep preset" is one call of the model_presets tool (action: "switch", preset: "deep"), and "which presets do I have" is action: "list". Overrides take precedence over the preset's own values, exactly as the original's --provider and --reasoning flags did.

A switch applies from the next turn, and a session that has already begun keeps the model it began on — so if a different model is wanted in the middle of long work, say so before the next ask. A rung the model does not admit is refused with the rungs it does, the model having already moved; the answer says so in as many words. /preset status reads what is pinned without moving anything.

(help :model-presets) answers the manual while the cell runs. The picker, the lines and the refusals all come from the one table the section declares, so a preset that reads right is a preset that works.

Credits and licence

MIT, the original's licence, kept byte for byte in LICENSE; NOTICE says what was rewritten and what was left out. Tests: .github/test-cell.sh tools/nodecode-model-presets from the cells repository — the pick is stubbed, so they need no provider, no key and no network.