Configuration

Where config lives

promptsize looks for configuration in this order:

1. The path passed to --config.
2. promptsize.config.{ts,mjs,js,cjs,json} in the current directory.
3. A "promptsize" key in package.json.

Globs and the baseline file are resolved relative to the config file’s location.

Shape

{
  "tokenizer": "o200k_base",
  "limits": [
    { "name": "system prompt", "path": "prompts/system.md", "limit": "2k" },
    { "name": "few-shot examples", "path": "prompts/examples/*.md", "limit": 8000 },
    { "name": "rag template", "path": "src/rag/*.txt", "limit": "1.5k", "tokenizer": "gpt-4" }
  ]
}

A JavaScript/TypeScript config works too — export the object as default:

promptsize.config.ts

import type { PromptSizeConfig } from "promptsize";

export default {
  tokenizer: "o200k_base",
  limits: [
    { name: "system prompt", path: "prompts/system.md", limit: "2k" },
  ],
} satisfies PromptSizeConfig;

Fields

Field	Type	Notes
tokenizer	string	Default encoding/model for all entries. Default o200k_base.
limits[]	array	One budget per logical prompt. Must be non-empty.
limits[].name	string?	Display name and baseline key. Defaults to the joined glob(s).
limits[].path	string | string[]	File path(s) / glob(s), relative to the config file.
limits[].limit	number | string	Token budget. 2000, "2k", "1.5k", "2 k tokens" all parse.
limits[].tokenizer	string?	Per-entry override of the tokenizer/model.

When an entry’s path matches multiple files, their tokens are summed into a single number and compared to one budget.

Limit syntax

A budget is a token count. You can write it as:

• a number — 8000
• a string with a k/m suffix — "2k", "1.5k", "1m"
• with an optional trailing word — "2k tokens"

Tokenizer & model names

Set tokenizer to an encoding name or a model name. Model names map to an encoding by longest-prefix match:

You write	Encoding used
o200k_base, gpt-4o, gpt-4.1, gpt-5, o1/o3/o4	o200k_base
cl100k_base, gpt-4, gpt-3.5, text-embedding-3-*	cl100k_base
claude-*, gemini-*	o200k_base (approximation — see Why)
text-davinci-003, code-davinci-*	p50k_base
davinci, gpt2	r50k_base

The encoding actually used is printed next to each result, so the number is always traceable.

Baseline file

Running promptsize --save writes .promptsize.json next to your config:

{
  "version": 1,
  "entries": {
    "system prompt": 1840,
    "few-shot examples": 8880
  }
}

Commit it. On later runs, promptsize shows the delta for each entry (+120 vs baseline), giving you regression visibility even when you’re still within budget.