manageprompt

JavaScript SDK for ManagePrompt — a local LLM call debugger.

Captures every LLM call with full prompt, response, token usage, cost, and latency.

Prerequisites

This package sends captured data to the ManagePrompt server. Install and start the server first:

brew install techulus/tap/manageprompt
manageprompt start

See the ManagePrompt README for other installation methods.

Install

pnpm add manageprompt

Usage

Vercel AI SDK Middleware (Recommended)

This integration supports Vercel AI SDK 7 and requires Node.js 22 or later.

import { generateText, wrapLanguageModel } from "ai";
import { openai } from "@ai-sdk/openai";
import { devToolsMiddleware } from "manageprompt";

const model = wrapLanguageModel({
  model: openai("gpt-4o"),
  middleware: devToolsMiddleware(),
});

const { text } = await generateText({ model, prompt: "Hello" });

Works with any AI SDK provider — OpenAI, Anthropic, Google, Mistral, etc.

Test the integration locally

Start ManagePrompt from the repository root:

go run ./cmd/manageprompt start

In another terminal, build the package and run the example with an OpenAI API key:

cd packages/manageprompt-js
pnpm build
OPENAI_API_KEY=your-key pnpm example:vercel-ai

The example makes one generated and one streamed request. Open http://localhost:54321 and verify both calls show the prompt, response, token usage, latency, finish reason, and correct streaming status.

capture()

Wraps any SDK call. Auto-detects provider, extracts tokens, cost, and latency.

import OpenAI from "openai";
import { capture } from "manageprompt";

const openai = new OpenAI();

const response = await capture(
  { model: "gpt-4o-mini", messages: [{ role: "user" as const, content: "Hello" }] },
  (input) => openai.chat.completions.create(input),
);

Works with OpenAI and Anthropic response formats.

log()

Manual logging for full control over what gets sent.

import { log } from "manageprompt";

log({
  model: "gpt-4o",
  provider: "openai",
  prompt: messages,
  response_text: "Hello!",
  tokens_input: 10,
  tokens_output: 5,
  latency_ms: 230,
});

Options

devToolsMiddleware({ url: "http://localhost:54321" });

capture({ model: "gpt-4o", messages, url: "http://localhost:54321" }, fn);

log({ model: "gpt-4o", provider: "openai", url: "http://localhost:54321" });

All functions default to http://localhost:54321.

License

MIT