The params property

Core to any chain is the inputs it receives from the user, or consuming application. For example, if you are building a chain that converts PDF to text and then summarises it, you will need to receive the PDF file as an input. Or if you’re building a chain that answers a user’s question, you will want to receive the user’s question as an input.

Define these in your Relevance AI chain via the params option.

You define each step using JSONSchema typing. This gets validated both on build time via our SDK and Typescript, as well as on run time via our API. This helps you create rock solid production ready features.

const chain = defineChain({
    params: {
        pdf_url: {
            type: 'string',
        },
        user_question: {
            type: 'string',
        },
        num_answers_to_generate: {
            type: 'number',
        },
    },
});

You are then able to retrieve these params in your setup function:

const chain = defineChain({
    params: {
        pdf_url: {
            type: 'string',
        },
        user_question: {
            type: 'string',
        },
        num_answers_to_generate: {
            type: 'number',
        },
    },
    setup({ params, step }) {
        const { pdf_url, user_questions, num_answers_to_generate } = params;

        const { text } = step('pdf_to_text', { pdf_url });
    },
});

All of your params will be typed, and warnings will be thrown if you try to use them in a transformation step inappopriately.