mem0ai/mem0 · error · Error

`embeddingModelDims` must be a positive integer

Error message

`embeddingModelDims` must be a positive integer

What it means

embeddingModelDims becomes the vector dimension of the Oracle vector index and column; it must be a positive integer because VECTOR(N) dimensions are a fixed integer. The check uses Number.isInteger, so fractional values, zero, negatives, NaN, Infinity, and non-numbers all fail (defaults to 1536 when omitted).

Source

Thrown at mem0-ts/src/oss/src/vector_stores/oracledb.ts:344

  constructor(config: OracleDBConfig) {
    if (!config.connectionParams && !config.client) {
      throw new Error(
        "Must provide at least one of `connectionParams` and `client`",
      );
    }

    this.collectionName = quoteIdentifier(config.collectionName || "mem0");
    this.indexName = quoteIdentifier(
      config.indexName || `${config.collectionName || "mem0"}_VEC_IDX`,
    );

    this.embeddingModelDims = config.embeddingModelDims ?? 1536;
    if (
      !Number.isInteger(this.embeddingModelDims) ||
      this.embeddingModelDims <= 0
    ) {
      throw new Error("`embeddingModelDims` must be a positive integer");
    }

    const distanceMetric = (config.distanceMetric ??
      "COSINE") as string as DistanceMetric;
    this.distanceMetric = distanceMetric.toUpperCase() as DistanceMetric;
    if (!DISTANCE_METRICS.includes(this.distanceMetric)) {
      throw new Error(`Unsupported distance metric: ${config.distanceMetric}`);
    }

    const indexType = (config.indexType ?? "HNSW") as string;
    this.indexType = indexType.toUpperCase() as IndexType;
    if (this.indexType !== "HNSW" && this.indexType !== "IVF") {
      throw new Error(`Unsupported index type: ${config.indexType}`);
    }

    this.indexAccuracy = config.indexAccuracy;
    if (
      this.indexAccuracy !== undefined &&

View on GitHub (pinned to 001c235229)

Solutions

1. Set an integer matching your embedding model's output size: 1536 (OpenAI text-embedding-3-small), 1024 (Cohere embed-v3), 768 (many open models).
2. Parse env values as integers: parseInt(process.env.EMBED_DIMS!, 10) with a Number.isInteger guard.
3. Check for accidental string values or float arithmetic; wrap with Math.round/parseInt if the source may be fractional.
4. Omit the option entirely if 1536 is correct — the default applies.

Example fix

// before
new OracleDB({ connectionParams, embeddingModelDims: Number(process.env.DIMS) }); // '768' -> 768 ok, '768.5' or undefined -> NaN

// after
const dims = parseInt(process.env.DIMS ?? '1536', 10);
new OracleDB({ connectionParams, embeddingModelDims: dims });

function parseDims(raw: string | number | undefined, fallback = 1536): number {
  const n = typeof raw === 'string' ? parseInt(raw, 10) : raw;
  if (n === undefined) return fallback;
  if (!Number.isInteger(n) || n <= 0) throw new RangeError(`embeddingModelDims must be a positive integer, got ${String(raw)}`);
  return n;
}

const isPositiveInt = (v: unknown): v is number => typeof v === 'number' && Number.isInteger(v) && v > 0;

try { new OracleDB(cfg); } catch (e) { if (e instanceof Error && e.message.includes('embeddingModelDims')) { /* correct the dimension to the embedding model's output size */ } else throw e; }

When it happens

Trigger: new OracleDB({ connectionParams, embeddingModelDims: 1536.5 }); embeddingModelDims: 0 or -1; dims read from an env var without parsing (string '768'); dims computed as NaN from undefined arithmetic; passing dims as a string '1536'.

Common situations: Switching embedding providers (e.g. OpenAI 1536 → Cohere 1024) and hand-editing the number; reading dimensions from a config/env as a string; arithmetic like 3072/2 producing 1536 vs a typo producing 1536.0-style floats from JSON parsers that keep floats.

Related errors

• Must provide at least one of `connectionParams` and `client`
• Unsupported distance metric: ${config.distanceMetric}
• Unsupported index type: ${config.indexType}
• `indexAccuracy` must be an integer between 1 and 100
• Unknown embedder provider: ${providerId}

AI-assisted analysis of mem0ai/mem0@001c235229 (2026-08-15). Data as JSON: /api/errors/b3496dc5f262cb95. Report an issue: GitHub.