Arcane.ollama.embed()

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

Syntax

Arcane.ollama.embed(request)

Parameters

request: OllamaEmbedRequest

Return value

Promise<OllamaEmbedResponse>

Description

Creates embeddings after Core re-admits the exact requested model.

Overview

Arcane.ollama.embed(request) creates provider-native local embeddings after Core re-admits the exact model and requested context. Use it only when an application intentionally owns the Ollama embedding contract and the returned vectors' storage/privacy lifecycle.

Parameters

request is a closed plain object of at most 8 MiB. Required model uses the Ollama name pattern. Supported provider fields are input (text or a batch), truncate, dimensions, keep_alive, and options; when present, options.num_ctx is 1,024-262,144. Nested provider-native values are forwarded for Ollama to validate.

Return value

It resolves to Ollama's bounded provider-native embed envelope, including the provider's vectors and metadata. Arcane does not normalize vector count or dimension into a separate stable entity.

Availability

This is a desktop Core method requiring ai.inference. Verified models can use the managed ensure/repair path before admission. Unverified installed models also require ai.models.unverified.inference. Applications with an isolated model contract must use their application-owned API, and Android does not project direct embed.

Errors and recovery

Correct invalid object, unknown field, model, serialization, request-size, or context errors before retrying. Model-policy, installation, isolation, and resource-admission errors require a currently admitted embedding model. Provider failures remain local-Ollama errors and never trigger remote fallback.

Streaming, cancellation, and events

The public wrapper does not accept stream controls, a chunk callback, or an AbortSignal; it resolves one result and emits no method event. Its renderer timeout is ten minutes. Although the Core RPC boundary can cancel an embed request when a host control arrives, application code has no per-call signal in this API.

Example

async function embedNonSensitiveText(verifiedEmbeddingModel, text) {
    if (typeof text !== 'string' || text.length === 0) {
        throw new TypeError('Embedding input must be nonempty text.');
    }
    return globalThis.Arcane.ollama.embed({
        model: verifiedEmbeddingModel,
        input: text
    });
}

Reference group

AI and Ollama