Count Tokens in Node.js

Token counting in Node.js is useful for servers that build AI prompts dynamically.

What consumes the limit

Before sending a request, an application may want to determine whether the prompt fits within a model's context window, whether old conversation messages should be summarized, or approximately how much the request will cost.

The typical workflow is:

Leave room for a useful response

`javascript const tokenIds = tokenizer.encode(text); const count = tokenIds.length;

console.log(count); `

Check the model-specific constraint

The specific tokenizer depends on the model and package being used.

In real applications, count the assembled model input rather than only the latest user message.

Practical checks before you proceed

For example, your request might contain:

`text System instructions: 1,200 tokens Conversation history: 8,000 tokens Retrieved documents: 15,000 tokens User request: 300 tokens `

Practical checks before you proceed

The user may see only a 300-token question even though your application is preparing roughly 24,500 tokens of content.

This is particularly important for long-running chats.

Practical checks before you proceed

A common Node.js architecture calculates estimated context usage before each request. If usage approaches a configured threshold, it summarizes old history or removes low-priority information.

Avoid hard-coding the maximum context for every model. Store limits in configuration because model capabilities change.

Practical checks before you proceed

Finally, compare estimates with actual API response usage. Provider-reported usage is more reliable for billing because chat formatting, tools, reasoning, and other internal considerations can affect real counts.

Token counting should therefore be part of both prompt management and production observability.