GPTScript

Overview

GPTScript is a new scripting language to automate your interaction with a Large Language Model (LLM), namely OpenAI. The ultimate goal is to create a fully natural language based programming experience. The syntax of GPTScript is largely natural language, making it very easy to learn and use. Natural language prompts can be mixed with traditional scripts such as bash and python or even external HTTP service calls. With GPTScript you can do just about anything like plan a vacation, edit a file, run some SQL, or build a mongodb/flask app.

# example.gpt

Tools: sys.download, sys.exec, sys.remove

Download https://www.sqlitetutorial.net/wp-content/uploads/2018/03/chinook.zip to a
random file. Then expand the archive to a temporary location as there is a sqlite
database in it.

First inspect the schema of the database to understand the table structure.

Form and run a SQL query to find the artist with the most number of albums and output
the result of that.

When done remove the database file and the downloaded content.

$ gptscript ./example.gpt

OUTPUT:

The artist with the most number of albums in the database is Iron Maiden, with a total
of 21 albums.

Quick Start

1. Install the latest release

Homebrew (macOS and Linux)

brew install gptscript-ai/tap/gptscript

Install Script (macOS and Linux):

curl https://get.gptscript.ai/install.sh | sh

Manually

Download and install the archive for your platform and architecture from the releases page.

2. Get an API key from OpenAI.

export OPENAI_API_KEY="your-api-key"

3. Run Hello World

gptscript https://get.gptscript.ai/echo.gpt --input 'Hello, World!'

OUTPUT:

Hello, World!

The model used by default is gpt-4-turbo-preview and you must have access to that model in your OpenAI account.

4. Extra Credit: Examples and Run Debugging UI

Clone examples and run debugging UI

git clone https://github.com/gptscript-ai/gptscript
cd gptscript/examples

# Run the debugging UI
gptscript --server

How it works

GPTScript is composed of tools. Each tool performs a series of actions similar to a function. Tools have available to them other tools that can be invoked similar to a function call. While similar to a function, the tools are primarily implemented with a natural language prompt. The interaction of the tools is determined by the AI model, the model determines if the tool needs to be invoked and what arguments to pass. Tools are intended to be implemented with a natural language prompt but can also be implemented with a command or HTTP call.

Example

Below are two tool definitions, separated by ---. The first tool does not require a name or description, but every tool after name and description are required. The first tool, has the parameter tools: bob meaning that the tool named bob is available to be called if needed.

tools: bob

Ask Bob how he is doing and let me know exactly what he said.

---
name: bob
description: I'm Bob, a friendly guy.
args: question: The question to ask Bob.

When asked how I am doing, respond with "Thanks for asking "${question}", I'm doing great fellow friendly AI tool!"

Put the above content in a file named bob.gpt and run the following command:

$ gptscript bob.gpt

OUTPUT:

Bob said, "Thanks for asking 'How are you doing?', I'm doing great fellow friendly AI tool!"

Tools can be implemented by invoking a program instead of a natural language prompt. The below example is the same as the previous example but implements Bob using python.

Tools: bob

Ask Bob how he is doing and let me know exactly what he said.

---
Name: bob
Description: I'm Bob, a friendly guy.
Args: question: The question to ask Bob.

#!python3

import os

print(f"Thanks for asking {os.environ['question']}, I'm doing great fellow friendly AI tool!")

With these basic building blocks you can create complex scripts with AI interacting with AI, your local system, data, or external services.

GPT File Reference

Extension

GPTScript files use the .gpt extension by convention.

File Structure

A GPTScript file has one or more tools in the file. Each tool is separated by three dashes --- alone on a line.

Name: tool1
Description: This is tool1

Do sample tool stuff.

---
Name: tool2
Description: This is tool2

Do more sample tool stuff.

Tool Definition

A tool starts with a preamble that defines the tool's name, description, args, available tools and additional parameters. The preamble is followed by the tool's body, which contains the instructions for the tool. Comments in the preamble are lines starting with # and are ignored by the parser. Comments are not really encouraged as the text is typically more useful in the description, argument descriptions or instructions.

Name: tool-name
# This is a comment in the preamble.
Description: Tool description
# This tool can invoke tool1 or tool2 if needed
Tools: tool1, tool2
Args: arg1: The description of arg1
      
Tool instructions go here.

Tool Parameters

Tool parameters are key-value pairs defined at the beginning of a tool block, before any instructional text. They are specified in the format key: value. The parser recognizes the following keys (case-insensitive and spaces are ignored):

Name: The name of the tool.

Model Name: The OpenAI model to use, by default it uses "gpt-4-turbo-preview"

Description: The description of the tool. It is important that the properly describes the tool's purpose as the description is used by the LLM to determine if the tool is to be invoked.

Internal Prompt: Setting this to false will disable the built in system prompt for this tool. GPTScript includes a default system prompt to instruct the AI to behave more like a script engine and not a "helpful assistant."

Tools: A comma-separated list of tools that are available to be called by this tool. A tool can only call the tools that are defined here. A tool name can reference a name in the current file, or a file relative to the current directory of the tool file, a http(s) URL, or a file in GitHub using github.com/username/repo/file.gpt format. When referencing a file or URL the tool loaded is the first tool in the file. To reference a specific tool in a file or URL use the syntax tool-name from file-or-url.

Args: Arguments for the tool. Each argument is defined in the format arg-name: description. All arguments are essentially strings. No other type really exists as all input and output to tools is text based.

Max Tokens: Set to a number if you wish to limit the maximum number of tokens that can be generated by the LLM.

JSON Response: Setting to true will cause the LLM to respond in a JSON format. If you set true you must also include instructions in the tool to inform the LLM to respond in some JSON structure.

Temperature: A floating-point number representing the temperature parameter. By default the temperature is 0. Set to a higher number to make the LLM more creative.

Tool Body

The tool body contains the instructions for the tool which can be a natural language prompt or a command to execute. Commands must start with #! followed by the interpreter (e.g. #!/bin/bash, #!python3) a text that will be placed in a file and passed to the interpreter. Arguments can be references in the instructions using the format ${arg1}.

name: echo-ai
description: A tool that echos the input
args: input: The input

Just return only "${input}"

---
name: echo-command
description: A tool that echos the input
args: input: The input

#!/bin/bash
        
echo "${input}"

Built in Tools

There are several built in tools to do basic things like read/write files, download http content and execute commands. Run gptscript --list-tools to list all the built-in tools.

Examples

For more examples check out the examples directory.

Community

Join us on Discord:

License

Copyright (c) 2023 Acorn Labs, Inc.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.