prompt-template-lite

Lightweight prompt templating with {{var}} interpolation and version tracking — so logs and evals can record exactly which prompt version produced which LLM output.

Install

npm install prompt-template-lite

Quick start

import { defineTemplate } from 'prompt-template-lite';

const summarize = defineTemplate('Summarize this in {{sentences}} sentences:\n{{text}}', { version: 'v3' });

const { text, version } = summarize.render({ sentences: 2, text: doc });
// log `version` alongside the LLM response so you can trace which prompt produced it

Missing a required variable throws immediately instead of silently rendering {{name}} into the prompt:

summarize.render({ text: doc }); // Error: Missing required template variable: "sentences"

Why version tracking matters

Prompt text changes constantly during development, and a model's behavior shifts with it. Without recording which exact prompt version produced a given output, you can't tell whether a regression in an eval run came from a model change, a prompt edit, or noise. defineTemplate auto-derives a version hash from the template text (or accepts an explicit one), and every render result carries it — cheap enough to log on every call.

API

• defineTemplate(template, options?) — returns a PromptTemplate with .render(vars), .version, .source
• .render(vars) — throws if a required {{var}} is missing, otherwise returns { text, version }

License

MIT