janhq/jan · error · Error

Embedding dimension not available

Error message

Embedding dimension not available

What it means

Thrown by VectorDBExtension.ingestFileForProject() after re-embedding chunks when the resulting embedding vectors have length 0 (finalDimension <= 0). This means the embedding engine returned empty/malformed vectors even though chunks exist. It surfaces an upstream embed() failure that did not throw but produced no usable vector data.

Source

Thrown at extensions/vector-db-extension/src/index.ts:118

    const collectionDimension = dimension > 0 ? dimension : 384
    await this.createCollectionForProject(projectId, collectionDimension)

    // Now check for duplicates
    const existingFiles = await vecdb.listAttachments(this.collectionForProject(projectId)).catch(() => [])
    const duplicate = existingFiles.find((f: any) => f.name === file.name && f.path === file.path)
    if (duplicate) {
      throw new Error(`File '${file.name}' has already been attached to this project`)
    }

    if (!chunks.length) {
      const fi = await vecdb.createFile(this.collectionForProject(projectId), file)
      return fi
    }

    // Re-embed if we got dimension from createCollection
    const embeddings = await this.embedTexts(chunks)
    const finalDimension = embeddings[0]?.length || 0
    if (finalDimension <= 0) throw new Error('Embedding dimension not available')

    // Ensure collection has correct dimension
    if (finalDimension !== collectionDimension) {
      await this.deleteCollectionForProject(projectId)
      await this.createCollectionForProject(projectId, finalDimension)
    }

    const fi = await vecdb.createFile(this.collectionForProject(projectId), file)
    await vecdb.insertChunks(
      this.collectionForProject(projectId),
      fi.id,
      chunks.map((t, i) => ({ text: t, embedding: embeddings[i] }))
    )
    const infos = await vecdb.listAttachments(this.collectionForProject(projectId))
    const updated = infos.find((e) => e.id === fi.id)
    return updated || { ...fi, chunk_count: chunks.length }
  }

View on GitHub (pinned to fad3f12a14)

Solutions

1. Verify an embedding model is fully loaded and embed() returns vectors (test embedTexts on a sample).
2. Restart/reload the llamacpp extension and embedding model.
3. Reinstall or pick a different embedding model whose output dimension is non-zero.
4. Inspect llamacpp logs for the embed() call to see why vectors are empty.

Example fix

// before
await vecdbExt.ingestFileForProject(projectId, file, opts)

// after
const probe = await vecdbExt.embedTexts?.(['probe']) ?? await rag.embed(['probe'])
if (!probe?.[0]?.length) {
  throw new Error('Embedding model returned no vectors; reload the embedding model')
}
await vecdbExt.ingestFileForProject(projectId, file, opts)

const probe = await rag.embed(['dimension probe']).catch(() => [])
if (!probe?.[0]?.length) {
  // embedding model not ready; do not ingest
}

try {
  await vecdbExt.ingestFileForProject(projectId, file, opts)
} catch (e) {
  if (e instanceof Error && e.message === 'Embedding dimension not available') {
    await reloadEmbeddingModel()
    return vecdbExt.ingestFileForProject(projectId, file, opts)
  }
  throw e
}

When it happens

Trigger: Embedding model loaded but returning empty vectors; embed() resolves with an empty/short data array so embeddings[0]?.length is 0; embedding dimensionality could not be derived from the response shape.

Common situations: Embedding model partially loaded or corrupted; llamacpp embed() returned an error object treated as data; mismatch between requested text count and returned indices.

Related errors

• Vector DB extension does not support project-level ingestion
• File '${file.name}' has already been attached to this projec
• Failed to determine embedding context size: ${e instanceof E
• Failed to count embedding tokens: ${e instanceof Error ? e.m
• llamacpp extension not available

AI-assisted analysis of janhq/jan@fad3f12a14 (2026-08-12). Data as JSON: /api/errors/74fca0024b1793a8. Report an issue: GitHub.