open source

Local LLM call
debugger

Capture every LLM API call during development. Request/response bodies, token usage, cost estimation, and latency in a local web UI.

$ manageprompt start

Features

Everything you need to debug LLM calls

↔

Full capture

Request & response bodies with visual flow visualization.

Token tracking

Input, output, cache read & write tokens for every call.

Cost estimation

Real-time cost calculated via models.dev pricing data.

Streaming

Captures both streaming and non-streaming calls automatically.

•

Real-time UI

WebSocket-powered live updates as calls happen.

Project scoped

SQLite DB per project directory. Zero config needed.

Integration

Two lines to get started

AI SDK Middleware

recommended

pnpm add manageprompt

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

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

capture()

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

const openai = new OpenAI();

const response = await capture(
  { model: "gpt-4o-mini", messages },
  (input) => openai.chat.completions.create(input),
);

Works with any LLM API — OpenAI, Anthropic, Google, Mistral, and more.

Install

Get started in seconds

Homebrew

brew install techulus/tap/manageprompt

Go

go install github.com/techulus/manage-prompt/cmd/manageprompt@latest

npm