Intervene Parser

Introduction

Parser to generate type-safe and context-safe API calls given a natural language query.

Here's a quick demo:

parser-demo-sped-up.mp4

Here's a sample output:

{
    "provider": "<API provider name>",
    "method": "<Http Method>",
    "path": "<API endpoint to call>",
    "bodyParams": "<eval'able function to return body params>",
    "queryParams": "<eval'able function to return query params>",
    "pathParams": "<eval'able function to return path params>",
    "requestContentType": "application/x-www-form-urlencoded",
    "responseContentType": "application/json",
    "responseSchema": "<schema of the Response>"
  }

It captures what inputs are needed and how to call the API, and gives you functions that can be eval'd on your end.

Problem

GPT4 currently hallucinates more often than not when dealing with APIs.

We take a user input and a corresponding OpenAPI spec, check which endpoints are the most feasible, pick one, verify the inputs and their types, and give you an expression (or code) that you can eval on your end.

How To

You can follow the guide to install dependencies and try the tool via our CLI.

To install dependencies:

pnpm i

npm install

CLI Usage

Running the CLI looks like below. You can run the help command to find out more.

pnpm cli help

npx tsx src/cli/run.ts help

Configure Command

This command allows you to set up the OpenAI API key and choose a vector database (either ChromaDB or Vectra, an in memory vector DB) along with its respective API key.

pnpm cli configure

npx tsx src/cli/run.ts configure

Follow the prompts to input or update the keys.

Parse Command

This command lets you parse a natural language query and give you an appropriate command. The OpenAPI spec augments the LLM's knowledge on the specific API providers needed for the task.

pnpm cli parse "[Your natural language query here]" "/path/to/file1.json,/path/to/file2.json"

npx tsx src/cli/run.ts parse "[Your natural language query here]" "/path/to/file1.json,/path/to/file2.json"

The first argument is your natural language query. The second argument is a comma-separated list of OpenAPI spec files' paths that you want to load.

For example:

pnpm cli parse "Fetch customer details from Stripe" "./specs/stripe.json"