FlowiseAI/Flowise · error · Error

Vectors must have the same length as the number of dimension

Error message

Vectors must have the same length as the number of dimensions (${this.numDimensions})

What it means

addVectors pins the collection dimension on the first batch (this.numDimensions = vectors[0].length) and rejects any subsequent vector whose element count differs. Embedding dimension must be invariant for the collection's lifetime; switching embedding models produces this error.

Source

Thrown at packages/components/nodes/vectorstores/Chroma/core.ts:143

     * Adds vectors to the Chroma database. The vectors are associated with
     * the provided documents.
     * @param vectors An array of vectors to be added to the database.
     * @param documents An array of `Document` instances associated with the vectors.
     * @param options Optional. An object containing an array of `ids` for the vectors.
     * @returns A promise that resolves with an array of document IDs when the vectors have been added to the database.
     */
    async addVectors(vectors: number[][], documents: Document[], options?: { ids?: string[] }) {
        if (vectors.length === 0) {
            return []
        }
        if (this.numDimensions === undefined) {
            this.numDimensions = vectors[0].length
        }
        if (vectors.length !== documents.length) {
            throw new Error(`Vectors and metadatas must have the same length`)
        }
        if (vectors[0].length !== this.numDimensions) {
            throw new Error(`Vectors must have the same length as the number of dimensions (${this.numDimensions})`)
        }

        const documentIds = options?.ids ?? Array.from({ length: vectors.length }, () => uuid.v1())
        const collection = await this.ensureCollection()

        const mappedMetadatas: Metadata[] = documents.map(({ metadata }) => {
            let locFrom
            let locTo

            if (metadata?.loc) {
                if (metadata.loc.lines?.from !== undefined) locFrom = metadata.loc.lines.from
                if (metadata.loc.lines?.to !== undefined) locTo = metadata.loc.lines.to
            }

            const newMetadata: Document['metadata'] = {
                ...metadata,
                ...(locFrom !== undefined && { locFrom }),
                ...(locTo !== undefined && { locTo })

View on GitHub (pinned to abe4a8601a)

Solutions

1. Use a single embedding model for the entire collection lifetime.
2. If you must change models, create a new collection and reindex — do not reuse the existing one.
3. Validate vector length equals the configured model dimension before each addVectors call.

Example fix

// before
// first batch: openai ada-002 -> 1536d
// second batch: sentence-transformers -> 384d -> throws
await store.addVectors(batch2Vecs, batch2Docs)

// after
const EXPECTED = 1536
if (batch2Vecs[0].length !== EXPECTED) throw new Error('dimension drift')
// or create a new collection with its own dim and reindex

const EXPECTED_DIM = 1536 // set per model
if (vectors.length && vectors[0].length !== EXPECTED_DIM) {
  throw new Error(`Embedding dimension ${vectors[0].length} != expected ${EXPECTED_DIM}`)
}
await store.addVectors(vectors, documents)

function allSameDimension(vectors: number[][], dim: number): boolean {
  return vectors.every(v => Array.isArray(v) && v.length === dim)
}

try {
  await store.addVectors(vectors, documents)
} catch (e) {
  if (/number of dimensions/i.test(String(e))) {
    throw new Error(`Dimension drift detected — use a new collection or the original model (${EXPECTED_DIM}d)`, { cause: e })
  }
  throw e
}

When it happens

Trigger: A second addVectors call uses a different embedding model, a mixed batch where some vectors come from a different model, or a corrupt/truncated embedding.

Common situations: Switching from text-embedding-ada-002 (1536d) to a smaller model (e.g. 384d) on the same collection, mixing providers, or a custom embedder returning padded/truncated arrays.

Related errors

• Vectors and metadatas must have the same length
• Model ID is required
• Input Type must be selected for Cohere models.
• Invalid JSON in the OpenAIEmbedding's BaseOptions:
• Invalid JSON in the ChatOpenAI's BaseOptions:

AI-assisted analysis of FlowiseAI/Flowise@abe4a8601a (2026-08-12). Data as JSON: /api/errors/395f90fb84eef655. Report an issue: GitHub.