Actions and LLM Concepts

Overview

Conversations consist of a messageThread which contains dialogues. They each have a role and content property.:

Term	Description
Message Thread	A list/array of objects that comprise the conversation; each object is a dialogue
Conversation	Same as message thread
Dialogue	Each object is a Dialogue that has a role and a content property
Role	Will be one of `user`, `assistant`, or `system`.
Content	A String representation of the conversation
Agent	A Canopy bot that can take actions

Models

We current support the following models:

Model	Overview
`gpt-3.5-turbo`	Ultra low latency, not good for action calling or applications which require complex logic
`gpt-4-turbo`	~ +300 ms of latency. Good for agent like behaviour and action calling

CRUD conversations

You can run the CRUD operations:

Operation ID	Description
`ReadConversation`	Sends the `messageThread` to the client
`AppendDialogue`	Adds a `dialogue` to the `messageThread`
`UpdateDialogue`	Updates a `dialogue` in the `messageThread`
`DeleteDialogue`	Deletes a `dialogue` in the `messageThread`

Read Conversation

You can submit the ReadConversation event which will return the entire messageThread.

const canopy  = new Canopy ({ ... }) 
const { messageThread }  = await canopy.ReadConversation()

The response will be something like:

{
    success: true,
    timestamp: 1623346800,
    messageThread: [
         {
            role: "system",
            content: "You are a friendly bot"
        },
        {
            role: "user",
            content: "Hello"
        },
        {
            role: "assistant",
            content: "Hi there!"
        }
        ...
    ]
}

You can use ReadConversation to get the entire message thread, and parse and do what you want with it.

Append Dialogue

You can submit the AppendDialogue event which will, add the dialogue you provide to the end of the messageThread, and return the entire messageThread.

const canopy  = new Canopy ({ ... }) 
const { messageThread }  = await canopy.AppendDialogue({
    role: "system",
    content: "Here is information from the database:..."
})

The response will be the message thread with the appended dialogue. You can use AppendDialogue to add a message to the conversation. For example,if you were retrieving data from a database you could pass it as information in a system message. If you were

Update Dialogue

You can submit the UpdateDialogue event which will return the entire message thread.

const canopy  = new Canopy ({ ... }) 
const { messageThread }  = await canopy.UpdateDialogue({
    role: "system",
    content: "Here is information from the database:...", 
    dialogueIndex:4 // the index in the message thread array of this dialogue
})

The response will be the message thread with the appended dialogue. You can use UpdateDialogue to change a message in the conversation. For example, if you wanted to fill in information about the user in a system message as you learn it.

Delete Dialogue

You can submit the DeleteDialogue event which will return the entire message thread.

const canopy  = new Canopy ({ ... }) 
const { messageThread }  = await canopy.DeleteDialogue({
    dialogueIndex:4 // the index in the message thread array of this dialogue
})

The response will be the message thread with the appended dialogue.

Get Started

Documentation

API Reference

Actions and LLM Concepts

Overview

Models

CRUD conversations

Read Conversation

Append Dialogue

Update Dialogue

Delete Dialogue

​Overview

​Models

​CRUD conversations

​Read Conversation

​Append Dialogue

​Update Dialogue

​Delete Dialogue

Overview

Models

CRUD conversations

Read Conversation

Append Dialogue

Update Dialogue

Delete Dialogue