How It Works

Here's a high-level overview of how everything works together:

Build ReByte Table from Data Source

The purpose of the ReByte table is that, in most cases, the data is chaotic and its semantics are unclear. The table serves as a data-building process, extracting useful data from various sources and materializing it into a clean data view. Additionally, you can add metadata to this view, such as the relationships between columns, which helps large language models (LLMs) better understand the data.

Connect Table to LLM Workflow

Create a workflow that can interact with your tables, at its core, here's what happens:

1. take user's query, for example, "show me the sales number in the last three months"
2. load necessary tables, and output the schema. Essentially, this forms a virtual database with multiple tables.
3. feed the query and schema to LLM, and generate SQL query
4. query the virtual database with the generated SQL query, and output the result

Connect Workflow to Agent

Agent can run previously created workflows when needed, agent can look at the result from workflow, for example, if workflow produces an error, agent can retry the workflow until it succeeds, or agent can ask user for more information to help workflow succeed.