CherryHQ/cherry-studio · error · Error

the local embedding model is not fully downloaded

Error message

the local embedding model is not fully downloaded

What it means

Thrown by `currentModelDir()` when `localEmbeddingDownloadService.completeCacheDir()` returns null — i.e. the embedding model files are not fully present on disk. The runtime resolves the cache directory up front (rather than letting transformers.js discover missing files per candidate) so a missing cache fails in the main process with a clear message instead of an opaque worker resolution error.

Source

Thrown at src/main/ai/provider/custom/localEmbedding/localEmbeddingRuntime.ts:15

import { application } from '@application'
import type { EmbeddingModelDir } from '@main/ai/inference/inferenceProtocol'
import { LOCAL_MODELS } from '@main/ai/inference/localModelCatalog'
import { localEmbeddingDownloadService } from '@main/services/localModel'

/**
 * The cached model's directory, for loading it straight off disk. Resolving it here — from
 * the download service's own on-disk probe — rather than handing the worker a list of mirror
 * revisions to try means a missing cache fails in the main process with a clear message,
 * instead of surfacing as a transformers.js resolution error per candidate.
 */
export function currentModelDir(): EmbeddingModelDir {
  const modelDir = localEmbeddingDownloadService.completeCacheDir()
  if (!modelDir) {
    throw new Error('the local embedding model is not fully downloaded')
  }
  return modelDir
}

/**
 * Embed texts on the inference worker (off the main thread). Pooling and
 * normalization run inside the worker; this is a thin main-process entry point.
 * Model files must already be downloaded; inference never fetches missing files.
 */
export async function embedTexts(texts: string[], signal?: AbortSignal): Promise<number[][]> {
  if (texts.length === 0) return []
  return application
    .get('EmbeddingInferenceService')
    .embed(texts, currentModelDir(), LOCAL_MODELS.embedding.dtype, signal)
}

View on GitHub (pinned to 726446b54c)

Solutions

1. Trigger and await the embedding model download before indexing (use `localEmbeddingDownloadService`).
2. Check disk space and re-download if the cache is partial.
3. Gate the embedding UI on download completion so requests cannot start early.

Example fix

// before
await embedTexts(texts) // throws if not downloaded
// after
if (!localEmbeddingDownloadService.completeCacheDir()) {
  await localEmbeddingDownloadService.ensureDownloaded()
}
await embedTexts(texts)

if (!localEmbeddingDownloadService.completeCacheDir()) {
  throw new Error('Embedding model not downloaded — trigger the download before indexing')
}

const isEmbeddingModelReady = (): boolean =>
  localEmbeddingDownloadService.completeCacheDir() != null

When it happens

Trigger: Calling `embedTexts` (or otherwise resolving the model dir) before the download finished, or after a partial/interrupted download left the cache incomplete.

Common situations: Knowledge-base indexing started before the model download completed; the download was interrupted (app quit, network drop); disk corruption or manual deletion of cache files; insufficient disk space aborted the download.

Related errors

• Rerank response results must contain numeric index and relev
• local-embedding provider only supports text embeddings, not

AI-assisted analysis of CherryHQ/cherry-studio@726446b54c (2026-08-12). Data as JSON: /api/errors/685d75e5e47292fc. Report an issue: GitHub.