@jeremysnr/snug-openai

Fit OpenAI chat messages into a token budget. Pass the result directly to client.chat.completions.create.

import { fitMessages } from '@jeremysnr/snug-openai';

const { messages } = fitMessages(conversationHistory, {
  model: 'gpt-4o',
  budget: 8192,
  reserve: 1024,
});

const response = await client.chat.completions.create({ model: 'gpt-4o', messages });

System messages are always kept. Other messages are prioritised by recency — older messages are dropped first when the budget is tight.

Install

npm install @jeremysnr/snug-openai

openai must be installed as a peer dependency.

API

fitMessages(messages, options)

| Option | Type | Default | Description | |--------|------|---------|-------------| | budget | number | — | Token limit | | reserve | number | 0 | Tokens to hold back for the model's reply | | model | TiktokenModel | 'gpt-4o' | Used to select the tiktoken encoding |

Returns

{
  messages: ChatCompletionMessageParam[];  // ready to send
  dropped: ChatCompletionMessageParam[];   // what didn't fit
  tokensUsed: number;
  tokensRemaining: number;
}

Part of the snug ecosystem

• @jeremysnr/snug — zero-dependency core primitive
• @jeremysnr/snug-tiktoken — snug with tiktoken, model-agnostic
• @jeremysnr/snug-anthropic — snug for the Anthropic SDK

Licence

MIT