Getting Started

GenAIScript is a scripting language that integrates LLMs into the scripting process using a simplified JavaScript syntax. It allows users to create, debug, and automate LLM-based scripts.

GenAIScript brings the flexibility of JavaScript with the convinience of built-in output parsing to streamline the creation of LLM-based software solutions.

the script

The following script takes a file with text content (.txt, .pdf, .docx) as input and saves a summary of the file in another file.

summarize.genai.js

// metadata and model configuration
script({ title: "Summarize", model: "gpt4" })
// insert the context, define a "FILE" variable
const file = def("FILE", env.files)
// appends text to the prompt (file is the variable name)
$`Summarize ${file}. Save output to ${file}.summary`

GenAIScript will execute summarize.genai.js and generate the user message that will be sent to the LLM chat. It also populates the env.files variable with the files selected in the context (from a user UI interaction or CLI arguments).

user prompt

FILE ./lorem.txt:
```txt
Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua...
```

Shorten the following FILE. Limit changes to minimum.

System scripts

GenAIScript also automatically selects system scripts to support file generation and other features. Since we’re using files in this script, it will run the system.files script which teaches the LLM how to format files in a structured format.

system.files.genai.js

system({ title: "File generation" })
const folder = env.vars["outputFolder"] || "."
$`When generating or updating files you will use this syntax:`
def(`File ${folder}/file1.ts`, `What goes in ${folder}/file1.ts`, {
    language: "typescript",
})

And the resulting system prompt.

system prompt

When generating or updating files you will use the following syntax:
File ./file1.ts
```typescript
What goes in ./file1.ts
```

LLM invocation

All the generate prompts are formatted and sent to the LLM for processing. Typically, using the OpenAI Chat API (or compatible), this would be a JSON object with an array of messages.

llmrequest.json

{
    "model": "gpt4",
    "messages": [
        { "role": "system", "content": "When generating... " },
        { "role": "user", "content": "FILE ./lorem.txt:..." }
    ]
}

Output parsing

The LLM responds with a text which can be parsed for various micro-formats, like markdown code fences, files or annotations.

llmresponse.txt

File ./lorem.txt.summary
```
Lorem Ipsum.
```

GenAIScript automatically makes sense of the output and exposes it through a Refactoring Preview or directly saved to the file system.

Of course, things can get more complex - with functions, schemas, … -, but this is the basic flow of a GenAIScript script.

Next steps

Let’s start by installing the extension in Visual Studio Code.