Main Components

Agent

ReByte Agent is an autonomous entity that end users can interact with. It can:

Has UI, user can interact with it
Know available tools, and call them to perform the task
Planning and reasoning and recover from failures
Interact with environment, mainly the user and the sandbox environment
Can run synchronously or asynchronously

Workflow

An orchestration of tools and LLM in a sequence to achieve a specific goal. Can be built by ReByte Workflow Builder

Workflow has several building blocks:

LLM: could be any LLM
Retriever: retrieve data from database or vector engine
Memory: each workflow run comes with a persistent memory, which can be used to store context information. Primarily used to store conversation history
Ability to run few lines of code: usually used to do data transformation
Common Tools: http requester, search engine, etc.

Abstract representation of two dimensional data, enriched with metadata. Table usually has business semantics, such as "sales number in last three months", "employee number in each department", etc. It is a partial view of your database. Table can be:

Federated: all queries are directly routed to the source, we don't store any data
Fusion table: user can define how data comes from different sources, and how to merge them. ReByte will store the fusion table, and future queries doesn't need to go to the source again.

Relationship between Agent, Workflow, and Table

Each Agent has a list of available workflows, which determines the scope of tasks it can handle.
Each Agent also equipped with a code agent, which can write/execute code in the sandbox
Workflow can interact with Table to retrieve data
Workflow can interact with other workflows as well