docs/development/workflows.mdx

---
title: "Workflows"
description: The workflow automation system within AgentGPT.
icon: "wave-sine"
---

A core function of Reworkd is AI powered workflow automation. This documentation covers key concepts within our workflow platform.

## Frontend models
The workflow hierarchy follows a graph-like structure. The frontend models only prescribe the front-end view of the workflow.

- A workflow is the graph itself. It represents the workflow in its entirety
- A node is a single element within a workflow. It has a position
- An edge represents a connection between two nodes of a workflow

## Backend models
The backend models represent the mechanisms to actually perform work for a given node.
Each frontend `Node` will have an associated `Block`.
`Node` represents the frontend view / position while the `Block` represents what will actually happen when that `Node` is run.
For example, a "SlackMessageBlock" is a `Block` that, when executed, would send a user a message on "Slack".

## Adding a new block
To add a new block, start by updating the frontend:
- open next/src/services/workflow/node-block-definitions.ts
- figure it out
- (soon block definitions on frontend will be set from backend and edits won't be needed here)

Then update the backend:
- open platform/reworkd_platform/schemas/workflow/blocks
    - add a new file for your block
    - define the block's input and output types as classes
    - add a class for the block with
        - attributes: type, description, image_url, and input
        - async method: `run`
- install dependencies for your block with `poetry add`
- open platform/reworkd_platform/web/api/workflow/blocks/web/__init__.py
    - import your block
    - add an if branch to get_block_runner
- `docker compose down; docker compose up --build`

## Node types
- Triggers: Designate how workflows are run
- Actions: Performs a concrete piece of "work"