Documentation

Reference for integrating anytool into your LLM application. Start shipping faster today.

What is anytool?

anytool is a meta-tool for LLMs. Instead of building dozens of individual tools, add one tool that generates anything on demand. Think of it as "npm install" for LLM tools—you import capabilities instead of building them. Each generated tool runs in isolation, supports full NPM, and caches after first generation.

One Tool

Add anytool to your LLM app as a single tool

Generate On-Demand

Generates tools from prompts: QR codes, charts, validators, parsers

Cached & Fast

First generation: 3-5 seconds. Cached execution: <100ms

Secure Execution

All tools run in isolated sandboxed environments

Persistent State

Tools remember progress, cache results, resume work between executions

Edit & Iterate

Edit any tool's source code right in your browser. Changes regenerate instantly.

How It Works

Request received

Your LLM calls anytool with a prompt (e.g., "generate QR code") and input data

Code generation (first time only)

AI generates Bun-compatible code with npm imports. Takes 3-5 seconds. Code is cached by prompt hash.

First execution

Code runs in isolated Bun container. NPM packages auto-install on first use (cached globally). Takes 1-3 seconds depending on package count.

Subsequent calls

Same prompt hash? Code loads from cache. Packages already installed. Execution is <100ms (no generation, no installs).

Performance Breakdown

First generation + first execution: ~4-8 seconds

Cached tool execution: <100ms

Cache hit rate: ~95%+ (same prompt)

Get Started

Quick Start

Add anytool to your LLM app in 5 minutes. Copy the code example and run it.

Code Examples

AI SDK, LangChain, Python, custom implementations

Web Dashboard

Test tools manually in the browser

MCP Server (Advanced)

Native integration with Claude Desktop

Persistent State & SQL

Tools remember progress, cache results, and have isolated SQLite databases