mastra-ai/mastra · error · MastraError

Vector contains invalid value (null, undefined, NaN, or Infi

Error message

Vector contains invalid value (null, undefined, NaN, or Infinity) at position [${i}][${j}]

What it means

validateVectorValues scans each vector's components and rejects null, undefined, NaN, and Infinity values. Embedding vectors must be finite floats; any non-finite component indicates a corrupted embedding or numeric bug.

Source

Thrown at packages/core/src/vector/validation.ts:115

    const vector = vectors[i];

    if (!vector) {
      throw new MastraError({
        id: createVectorErrorId(storeName, 'UPSERT', 'INVALID_VECTOR'),
        domain: ErrorDomain.MASTRA_VECTOR,
        category: ErrorCategory.USER,
        details: {
          message: `Vector at index ${i} is null or undefined`,
          vectorIndex: i,
        },
      });
    }

    for (let j = 0; j < vector.length; j++) {
      const value = vector[j];

      if (value === null || value === undefined || !Number.isFinite(value)) {
        throw new MastraError({
          id: createVectorErrorId(storeName, 'UPSERT', 'INVALID_VECTOR_VALUE'),
          domain: ErrorDomain.MASTRA_VECTOR,
          category: ErrorCategory.USER,
          details: {
            message: `Vector contains invalid value (null, undefined, NaN, or Infinity) at position [${i}][${j}]`,
            vectorIndex: i,
            componentIndex: j,
            value: String(value),
          },
        });
      }
    }
  }
}

/**
 * Validates all upsert inputs including vector values
 * Combines validateUpsertInput and validateVectorValues

View on GitHub (pinned to 75dd419e61)

Solutions

1. Sanitize with Number.isFinite check per component before upsert and drop/repair offending vectors
2. Fix the embedding source: check for zero-norm division and model output validity
3. Use finite check during normalization: divide only if norm > 0

Example fix

// before
const norm = Math.sqrt(vec.reduce((s, x) => s + x * x, 0));
const normalized = vec.map((x) => x / norm);
// after
const norm = Math.sqrt(vec.reduce((s, x) => s + x * x, 0));
const normalized = norm > 0 ? vec.map((x) => x / norm) : vec;
if (!normalized.every(Number.isFinite)) throw new Error('Non-finite embedding');

const bad = vectors.findIndex((v) => Array.isArray(v) && !v.every(Number.isFinite));
if (bad !== -1) throw new Error(`non-finite value in vector ${bad}`);

function isFiniteVector(v: unknown): v is number[] {
  return Array.isArray(v) && v.every((x) => typeof x === 'number' && Number.isFinite(x));
}

try {
  await store.upsert({ indexName, vectors });
} catch (e) {
  if (e instanceof MastraError && e.id.includes('INVALID_VECTOR_VALUE')) {
    const { vectorIndex, valueIndex } = e.details;
    console.error(`Non-finite value at [${vectorIndex}][${valueIndex}]`);
  }
  throw e;
}

When it happens

Trigger: Embedding model returned NaN (e.g. zero-division in custom embeddings), JSON round-trip produced nulls, dividing by a zero norm when normalizing, or Infinity from overflowing math.

Common situations: Custom embedding functions with unstable math; parsing embeddings from CSV/text where missing values become null; overflow in manual vector arithmetic.

Related errors

• Vectors array cannot be empty
• Metadata array length must match vectors array length
• IDs array length must match vectors array length
• Vector at index ${i} is null or undefined
• ${flag} must be a positive integer

AI-assisted analysis of mastra-ai/mastra@75dd419e61 (2026-08-30). Data as JSON: /api/errors/699cd7834303e498. Report an issue: GitHub.