context-priority

Sort and trim items by priority to fit a token budget. Built for AI context windows, prompt construction, and retrieval pipelines.

npm install context-priority

import { prioritize } from "context-priority";

const items = [
  { content: "User prefers dark mode", score: 0.9 },
  { content: "Session expired at 3pm", score: 0.2 },
  { content: "API key for stripe", score: 0.7 },
  { content: "Favorite color is blue", score: 0.3 },
];

const { result, totalTokens, omitted } = prioritize(items, { maxTokens: 20 });
// → highest-score items that fit in 20 tokens

Why

When building LLM applications, you often have more context than fits in the window. This library gives you a deterministic way to keep the high-value content and drop the noise — no model calls, no dependencies.

API

prioritize(items, options)

Return field	Description
result	Items that fit, sorted by score descending
totalTokens	Token count of the result
omitted	Number of items dropped

prioritizeContent(items, options)

Same as prioritize but returns only the content strings.

Custom tokenizer

prioritize(items, {
  maxTokens: 4000,
  tokenizer: (text) => text.split(/\s+/).length, // word count
});

Default tokenizer uses a rough heuristic (~4 chars per token). Pass a tokenizer function for model-specific accuracy.