Magically run typesafe functions by utilizing large language models (LLMs) as a runtime.
Warning
The code in this repository (including this README) was created with GPT-4. This is not a production repository and caution should be used when running this code.
This library uses a Typescript transformer to take the return type of a function, convert that type into a JSON Schema, and then replace the function body with code to query the OpenAI API and validate the response against the JSON Schema.
This library doesn't write code for your functions, it allows you to use LLMs as a runtime.
Transform your TypeScript functions by adding the //@magic
comment.
Here's an example function:
// @magic
async function example(): Promise<Mountain> {
//Return the 3rd highest mountain
}
When this function is called, it'll leverage an AI language model like GPT-4
and return the following result:
{ name: 'Kangchenjunga', height: { meters: 8586, feet: 28169 } }
Note: In this example,
Mountain
is defined as:interface Height { meters: number; feet: number; } interface Mountain { name: string; height: Height; }
export OPENAI_API_KEY=your_api_key_here
export OPENAI_MODEL=gpt-4 #optional
npm install --save @jumploops/magic
npm install -D ts-patch
ts-patch install
Now, add the plugin to your tsconfig.json
:
{
"compilerOptions": {
"plugins": [
{ "transform": "@jumploops/magic" }
]
}
}
Install ttypescript:
npm i -D ttypescript
Next, add the plugin to your tsconfig.json
:
{
"compilerOptions": {
"plugins": [
{ "transform": "@jumploops/magic" }
]
}
}
Compile your project using ttypescript:
ttsc
This package took inspiration from Marvin - build AI functions that use an LLM as a runtime
We welcome contributions! Please open an issue or submit a pull request if you'd like to help improve @jumploops/magic.
MIT License.