Based on the user's query, the LLM generates a list of Salesforce Object Query Language (SOQL) queries that are needed to retrieve all the relevant details from Salesforce. These queries are executed in a sequence, and the results are used by the Gemini Large Language Model (LLM) as context to answer the user's query.

The process receives a request from the user that includes system instructions for an LLM and an additional system instruction for executing SOQL queries. The LLM uses the instructions to generate a list of SOQL queries that need to be executed against Salesforce. The process sequentially executes each generated SOQL query against the Salesforce database. After each query execution, the result is used as context for the next query to the LLM.

The process continues this cycle using the response from each SOQL query as context data for the next LLM query, along with the initial user instructions. This loop continues until all queries generated from the user input have been processed. The maximum number of requests to be made to Salesforce is set to 5 by default. You can change the limit while invoking the process.

After all the SOQL queries have been executed and their results gathered, a final query is made to the LLM. The context for this final query consists of the results of all the executed SOQL queries.

The LLM uses the aggregated context to provide a response to the user's original question.

Watch an interactive demo to know more about how to use this recipe.