Workflows

Prompt Chaining galore

!! WORKFLOWS ARE IN BETA !!

Overview

Workflows are how you can create complex prompt-chains and interact with live data, API's, and enhance your portals with RAG. This feature is under active development, but there's already enough there to bring value to our Beamers, so we thought we'd give y'all access.

Workflows function very similar to portals, and since they have the same inputs and outputs as a Portal can even be swapped in your application with very little work. You send variables and messages and then we respond with a new ChatMessage.

Start Beam

This first beam is where you define the global properties of your workflow, like it's name, or the variables that it can receive via the API.

Portal Beams

This works similar to the portal feature, except each version is a one-off, and not shared in the global list of Portals.

Inputs

Each beam has it's own set of variables defined by it's prompt. The input section allows for you to map the possible inputs available in the workflow at the stage of this Portal Beam to it's internal variables. The possibilities are,

• Output

  • This could be from a previous Portal Beam, or a API, or anything else that generates an output earlier in the workflow sequence

• Variable

  • This references the variables that are defined by the by top-level workflow in the Start Beam, and these values will later be "filled in" by your users when they interact with this workflow via the api

• Text

  • This is just static text that you define per Portal Beam.

You can also choose to not include all of the messages in the chat to the Ai in each Portal. This can be expecially useful when doing things like data-preprocessing, etc.

Previewing

You can also take this Portal Beam for a spin by just clicking the Demo/Preview Portal button, which allows you to explore how this works in isolation.

Decision Beam

These allow you to make branching decision logic based on, comparing variables, outputs, and static text.

Workflow Previewing

You can see how your workflow works at any point by clicking the "Demo/Preview Workflow" button. The output of the final Portal step is return to the chat at the end for you to interact with.

Get a chat completion from a workflow

This endpoint processes a chat interaction within a specified workflow. It supports both streaming and non-streaming operations based on the provided parameters.

POSThttps://api.easybeam.ai/v1/workflow/{id}

Authorization

BearerAuth

Path parameters

id*string

Unique identifier for the workflow

Example: "workflow-abcde"

Body

application/json

Payload containing the chat details and parameters for processing the workflow

variablesobject

Variables for the workflow

chatIdnullable string

Identifier for the chat session

Example: "chat-12345"

messagesarray of ChatMessage (object)

List of messages in the chat

streamboolean

Whether to stream the response

Example: true

userIdnullable string

ID of the user initiating the workflow

Example: "user-67890"

consolidateStreamnullable boolean

Whether to consolidate the stream

Example: true

Response

200400401

Workflow interaction successful

Body

application/json

newMessageChatMessage (object)

chatIdstring

Identifier for the chat session

Example: "chat-12345"

streamFinishednullable boolean

Indicates if the streaming has finished

Example: true

Request

JavaScriptCurlPython

const response = await fetch('https://api.easybeam.ai/v1/workflow/{id}', {
    method: 'POST',
    headers: {
      "Authorization": "Bearer JWT",
      "Content-Type": "application/json"
    },
    body: JSON.stringify({
      "variables": {
        "keyA": "valueA",
        "keyB": "valueB"
      },
      "chatId": "chat-12345",
      "messages": [
        {
          "content": "Can you help me with this workflow?",
          "role": "USER",
          "createdAt": "2024-04-27T13:00:00Z",
          "id": "msg-3"
        }
      ],
      "stream": false,
      "userId": "user-67890",
      "consolidateStream": true
    }),
});
const data = await response.json();

Response

200400401

{
  "newMessage": {
    "content": "Hello, how can I assist you today?",
    "role": "USER",
    "providerId": "openai",
    "createdAt": "2024-04-27T12:34:56Z",
    "id": "msg-1",
    "inputTokens": 150.5,
    "outputTokens": 200.75
  },
  "chatId": "chat-12345",
  "streamFinished": true
}