Arcane OS Docs Development only

Reference

Arcane.ollama.select()

Selects a managed size preference and reconciles the model.

  • Reference

This focused page is derived from the mechanically checked full member inventory.

Syntax

Arcane.ollama.select(preference)

Parameters

preference: ArcaneModelPreference

Return value

Promise<ArcaneModelSelectionResult>

Description

Selects a managed size preference and reconciles the model.

Overview

Arcane.ollama.select(preference) runs the managed model-selection workflow. It can download a verified base, create or replace a managed alias, perform native resource admission, and change persisted model state. Use it only from an admitted Settings or Shell choice with explicit user intent.

Parameters

preference is one of "auto", "3b", "8b", "12b", "20b", or "120b". It is a managed size preference, not an arbitrary Ollama model name. The wrapper maps an omitted or empty value to "auto".

Return value

It resolves to the selected model, alias, variant, preference, recommendation, base model, creation and alias-change effects, model-store and GPU evidence, final compatibility, automatic-candidate degradation/failures when relevant, rollback-snapshot state, and a completed ArcaneOperation.

Availability

This is an exclusive desktop Core mutation requiring ai.models.manage and Settings or Shell identity. Package-bound verified definitions and current native capacity govern which preference can complete. Android does not expose the workflow.

Errors and recovery

INVALID_ARCANE_MODEL_PREFERENCE requires an enum value. Registry, integrity, definition, native-admission, storage, alias, load, and rollback errors preserve stable codes and resolutions. If rollbackSnapshotRetained is true on success, surface the recovery detail rather than deleting the retained alias from renderer code.

Streaming, cancellation, and events

Selection emits standard operation lifecycle, log, and progress events. The wrapper uses a 50-minute timeout and exposes no signal; Core does not cooperatively cancel this managed workflow. A renderer timeout is not evidence that download or alias mutation stopped, so refresh selection() and status before retrying.

Example

The helper is deliberately not invoked; call it only after the user confirms the managed size choice and possible download.

async function selectManagedModelAfterConfirmation(preference) {
    const allowed = new Set(['auto', '3b', '8b', '12b', '20b', '120b']);
    if (!allowed.has(preference)) {
        throw new TypeError('Choose a documented Arcane model preference.');
    }
    return globalThis.Arcane.ollama.select(preference);
}

Reference group

AI and Ollama

Repository and reviewed source access