ModelByInputTokens requires inputTokens for resolution

Error message

ModelByInputTokens requires inputTokens for resolution

What it means

ObservationalMemory can be configured with a ModelByInputTokens, which selects a concrete language model dynamically based on the size of the current input (token count). During resolution via getConcreteModel, the caller must supply the inputTokens count; without it there is no way to pick the right model tier, so the library throws instead of guessing. This is a programming error in the internal call path, surfacing when the token count was never computed or propagated.

Source

Thrown at packages/memory/src/processors/observational-memory/observational-memory.ts:761

  /**
   * Wait for any in-flight async buffering operations for the given thread/resource.
   * Used by server endpoints to block until buffering completes so the UI can get final state.
   */
  async waitForBuffering(
    threadId: string | null | undefined,
    resourceId: string | null | undefined,
    timeoutMs = 30000,
  ): Promise<void> {
    return BufferingCoordinator.awaitBuffering(threadId, resourceId, this.scope, timeoutMs);
  }

  private getConcreteModel(
    model: ObservationalMemoryModel,
    inputTokens?: number,
  ): Exclude<ObservationalMemoryModel, ModelByInputTokens> {
    if (model instanceof ModelByInputTokens) {
      if (inputTokens === undefined) {
        throw new Error('ModelByInputTokens requires inputTokens for resolution');
      }
      return model.resolve(inputTokens) as Exclude<ObservationalMemoryModel, ModelByInputTokens>;
    }

    return model as Exclude<ObservationalMemoryModel, ModelByInputTokens>;
  }

  private getModelToResolve(
    model: ObservationalMemoryModel,
    inputTokens?: number,
  ): Parameters<typeof resolveModelConfig>[0] {
    const concreteModel = this.getConcreteModel(model, inputTokens);

    if (Array.isArray(concreteModel)) {
      return (concreteModel[0]?.model ?? 'unknown') as Parameters<typeof resolveModelConfig>[0];
    }
    if (typeof concreteModel === 'function') {
      // Wrap to handle functions that may return ModelWithRetries[]

View on GitHub (pinned to 75dd419e61)

Solutions

Ensure the call is made on a thread/context that has messages so inputTokens can be computed before model resolution.
If you call getConcreteModel yourself, always pass the computed inputTokens value.
Replace ModelByInputTokens with a concrete static model if you don't need dynamic tiering — then resolution never needs inputTokens.
Upgrade/check package version; if this occurs on normal empty-thread calls it may be a bug — file an issue with a reproduction.

Example fix

// before
const model = memory.getConcreteModel(new ModelByInputTokens({ small: m1, large: m2 }));
// after
const inputTokens = estimateTokens(messages);
const model = memory.getConcreteModel(new ModelByInputTokens({ small: m1, large: m2 }), inputTokens);

Defensive patterns

Strategy: validation

Validate before calling

if (model instanceof ModelByInputTokens) {
  const tokens = estimateInputTokens(thread.messages);
  if (tokens === undefined) throw new Error('Cannot resolve ModelByInputTokens: no token count available');
}

Type guard

function isResolvable(model: ObservationalMemoryModel, inputTokens?: number): boolean {
  return !(model instanceof ModelByInputTokens) || inputTokens !== undefined;
}

Prevention

Prefer a static concrete model unless dynamic tiering is required.
Only use ModelByInputTokens on threads that contain messages so token counts exist.
Never call getConcreteModel directly from application code.

When it happens

Trigger: getConcreteModel is called with a ModelByInputTokens instance and inputTokens === undefined. This happens when the internal concreteModel path resolves the model before any message/token counting has occurred (e.g. no stored messages, or a call path that skips token estimation), or if a custom subclass/caller invokes getConcreteModel without the second argument.

Common situations: Calling memory APIs (remember/recall) on an empty thread where no token usage has been recorded yet; custom code extending ObservationalMemory that calls getConcreteModel directly; a ModelByInputTokens configured without any messages in the thread so the resolver has no token count.

Understand the failure class

Background: "missing required argument" and "the following required arguments were not provided": what required-argument errors mean and how to fix them — this error's family across 20 libraries.

Related errors

AI-assisted analysis of mastra-ai/mastra@75dd419e61 (2026-08-30). Data as JSON: /api/errors/9126159eb724b056. Report an issue: GitHub.