mastra-ai/mastra · error · Error

ModelByInputTokens threshold keys must be positive numbers.

Error message

ModelByInputTokens threshold keys must be positive numbers. Got: ${limitStr}

What it means

normalizeThresholds converts the string keys of the upTo map into numeric token limits. Every key must parse to a finite number greater than zero; otherwise the processor would have unusable thresholds (NaN, 0, negative limits) and cannot order or compare them against input token counts.

Source

Thrown at packages/memory/src/processors/observational-memory/model-by-input-tokens.ts:38

    'modelId' in model ||
    'id' in model ||
    'providerId' in model ||
    'provider' in model ||
    ('doGenerate' in model && 'doStream' in model)
  );
}

function normalizeThresholds(config: ModelByInputTokensConfig) {
  const entries = Object.entries(config.upTo);

  if (entries.length === 0) {
    throw new Error('ModelByInputTokens requires at least one threshold in "upTo"');
  }

  for (const [limitStr, model] of entries) {
    const limit = Number(limitStr);
    if (!Number.isFinite(limit) || limit <= 0) {
      throw new Error(`ModelByInputTokens threshold keys must be positive numbers. Got: ${limitStr}`);
    }

    if (!isTieredModelTarget(model)) {
      throw new Error(`ModelByInputTokens requires a valid model target for threshold ${limitStr}`);
    }
  }

  return entries.map(([limitStr, model]) => ({ limit: Number(limitStr), model })).sort((a, b) => a.limit - b.limit);
}

export class ModelByInputTokens {
  private readonly thresholds: Array<{ limit: number; model: AgentConfig['model'] }>;

  constructor(config: ModelByInputTokensConfig) {
    this.thresholds = normalizeThresholds(config);
  }

  resolve(inputTokens: number): AgentConfig['model'] {

View on GitHub (pinned to 75dd419e61)

Solutions

1. Change the offending key to a plain positive numeric string, e.g. { '128000': model }
2. Remove unit suffixes and thousand separators from keys ('12k' -> '12000', '128,000' -> '128000')
3. Log/inspect the upTo keys and convert them with Number() yourself to find which one is invalid

Example fix

// before
new ModelByInputTokens({ upTo: { '12k': 'openai/gpt-4o-mini', '128000': 'openai/gpt-4o' } });
// after
new ModelByInputTokens({ upTo: { '12000': 'openai/gpt-4o-mini', '128000': 'openai/gpt-4o' } });

for (const key of Object.keys(upTo)) {
  const n = Number(key);
  if (!Number.isFinite(n) || n <= 0) throw new Error(`Invalid upTo key: "${key}"`);
}

function hasValidThresholdKeys(upTo: Record<string, unknown>): boolean {
  return Object.keys(upTo).every(k => { const n = Number(k); return Number.isFinite(n) && n > 0; });
}

try {
  processor = new ModelByInputTokens({ upTo });
} catch (e) {
  if (e instanceof Error && e.message.includes('positive numbers')) {
    const fixed = Object.fromEntries(Object.entries(upTo).map(([k, v]) => [String(Math.max(1, Number(k) || 1)), v]));
    processor = new ModelByInputTokens({ upTo: fixed });
  } else throw e;
}

When it happens

Trigger: `new ModelByInputTokens({ upTo: { '0': model } })`, `{ '-5000': model }`, `{ 'abc': model }`, or keys like `{ '': model }` / `{ '12e999': model }` that yield NaN or non-finite values when passed through Number().

Common situations: Typo in a threshold key (e.g. '1,28000' with a comma, or '12k' with a unit suffix); template-literal keys built from undefined variables producing 'undefined' or 'NaN'; locale-formatted numbers with dots/commas; copying config where a key was left as a placeholder.

Understand the failure class

Background: "Invalid value" and "allowed values are" config errors: what your library rejected and how to fix it — this error's family across 41 libraries.

Related errors

• Extractor "${name}" must include instructions.
• ModelByInputTokens requires a valid model target for thresho
• Cookie password must be at least 32 characters. Set WORKOS_C
• Factory rule version is required.
• ${label} must be an object.

AI-assisted analysis of mastra-ai/mastra@75dd419e61 (2026-08-30). Data as JSON: /api/errors/85ca2ae7c4951331. Report an issue: GitHub.