Class: Agent

Agent(config)

Agent

Constructor

new Agent(config)

Creates an instance of a LoopGPT Agent.

Parameters:
Name Type Description
config object

The configuration object for initializing the Agent class.

Properties
Name Type Attributes Description
keys keyConfig

The keys object containing model-related keys.
name string <optional>

The name of the agent. (optional)
description string <optional>

The description of the agent. (optional)
goals Array.<string> <optional>

The goals associated with the agent. (optional)
init_prompt number <optional>

The init prompt with general operating instructions for generating responses. (optional)
next_prompt number <optional>

The next prompt for evaluating responses and generating next responses. (optional)
model OpenAIModel <optional>

The OpenAI model for the agent. (optional)
tools Object <optional>

The tools associated with the agent. (optional)
embedding_provider * <optional>

The embedding provider for the agent. (optional)
temperature number <optional>

The temperature value for generating responses. (optional)
Source:

Members

_tools

Source:

constraints :string|Array.<any>

Type:
  • string | Array.<any>
Source:

history :Array.<{role: string, content: any}>

Type:
  • Array.<{role: string, content: any}>
Source:

plan :Array.<any>

Type:
  • Array.<any>
Source:

progress :Array.<any>

Type:
  • Array.<any>
Source:

Methods

_getNonUserMessages(n)

This function returns the last n non-user messages from a chat history, excluding any system messages that contain the phrase "do_nothing".

Parameters:
Name Type Description
n number

The number of non-user messages to retrieve from the chat history.
Source:
Returns:

This function returns an array of the last n non-user messages from the chat history, excluding any system messages that contain the phrase "do_nothing".

(async) chat(object)

This is a function for a chatbot agent that handles user messages, runs a language model to generate a response, and can stage and run tools based on the response.

Parameters:
Name Type Description
object Object

{message: string|null, run_tool: boolean,}
Source:
Returns:

The function chat() returns a Promise that resolves to the reply message from the agent's conversation with the user, or an error message if an error occurs during the conversation.

config(init_promptopt, next_promptopt) → {Agent}

This function takes in two boolean parameters and returns a modified configuration object with compressed history.

Parameters:
Name Type Attributes Default Description
init_prompt boolean <optional>
 false

A boolean value indicating whether or not to include the initial prompt in the configuration object.
next_prompt boolean <optional>
 false

The "next_prompt" parameter is a boolean value that determines whether or not a prompt should be displayed for the next input after the initial input. If set to "true", a prompt will be displayed for each subsequent input. If set to "false", no prompt will be displayed for subsequent inputs.
Source:
Returns:

The config object with the init_prompt and next_prompt properties removed if their corresponding arguments are false, and with the history property set to the compressed history obtained from the getCompressedHistory() method.

Type
Agent

constraintsPrompt()

The function generates a prompt message listing the constraints.

Source:
Returns:

The function constraintsPrompt() returns a string that lists the constraints of an object, with each constraint numbered and separated by a new line character.

(async) extractJsonWithGpt(s)

The function extracts JSON from a given string using GPT.

Parameters:
Name Type Description
s string

The parameter s is a string that represents the input data that needs to be converted to a JSON string.
Source:
Returns:

The function extractJsonWithGpt is returning the result of calling this.model.chat with a set of messages, a temperature of 0.0, and a maximum number of tokens to generate. The messages include a system message with a JavaScript function and a default response format, and a user message s.

getCompressedHistory()

This function returns a compressed version of a chat history by removing certain properties from assistant messages.

Source:
Returns:

The function getCompressedHistory() returns a modified version of the history array of messages. The modifications include removing all messages with the role of "user" and removing certain properties from the thoughts object of any messages with the role of "assistant". The modified history array is then returned.

getFullMessage(message) → {string}

This function returns a message with a prompt based on the current state of an agent.

Parameters:
Name Type Description
message string | null

The message parameter is a string that represents the user's input or response to the agent's prompt. It is an optional parameter that can be passed to the getFullMessage function.
Source:
Returns:

The function getFullMessage is returning a string that includes either the init_prompt or next_prompt property of the current object instance, followed by a new line and the message parameter (if provided).

Type
string

(async) getFullPrompt(user_inputopt) → {Promise.<{full_prompt: Array.<{role: string, content: string}>, token_count: number}>}

This function generates a full prompt for a chatbot conversation, including system messages, user input, and relevant memory.

Parameters:
Name Type Attributes Description
user_input string <optional>

The user's input, which is an optional parameter. If provided, it will be added to the prompt as a user message.
Source:
Returns:
  • An object with two properties: "full_prompt" which is an array of messages to be displayed to the user, and "token_count" which is the number of tokens used by the messages in the "full_prompt" array.
Type
Promise.<{full_prompt: Array.<{role: string, content: string}>, token_count: number}>

goalsPrompt()

The function generates a prompt displaying a list of goals.

Source:
Returns:

The goalsPrompt() function is returning a string that lists the goals of an object, with each goal numbered and separated by a newline character.

headerPrompt()

The function returns a string prompt based on the persona, goals, constraints, plan, and progress of a project.

Source:
Returns:

The headerPrompt() function is returning a string that includes prompts for the persona, goals, constraints, plan, and progress, joined together with line breaks.

(async) loadJson(s, try_gptopt)

The function attempts to parse a string as JSON, and if it fails, it may try to extract the JSON using GPT or return the original string.

Parameters:
Name Type Attributes Default Description
s string

The input string that contains the JSON data to be parsed.
try_gpt boolean <optional>
 true

A boolean parameter that indicates whether to try extracting JSON using GPT if the initial parsing fails. If set to true, the function will attempt to extract JSON using GPT if the initial parsing fails. If set to false, the function will not attempt to extract JSON using GPT.
Source:
Returns:

The loadJson function returns a parsed JSON object if the input string is in valid JSON format, or a string representation of the input if it cannot be parsed as JSON. If the input cannot be parsed as JSON and the try_gpt parameter is true, the function will attempt to extract JSON using a GPT model and retry parsing. If parsing still fails, an error is

personaPrompt()

The function returns a string that includes the name and description of a person.

Source:
Returns:

The function personaPrompt() is returning a string that includes the name and description of the object that the function is called on. The specific values of this.name and this.description will depend on the object that the function is called on.

planPrompt()

The function returns a string that displays the current plan.

Source:
Returns:

The planPrompt() method is returning a string that includes the current plan joined together with new line characters and preceded by the text "CURRENT PLAN:".

progressPrompt()

The function generates a progress prompt by iterating through a list of completed tasks and displaying them in a formatted string.

Source:
Returns:

The progressPrompt() function is returning a string that lists the progress made so far. The string includes a header "PROGRESS SO FAR:" and a numbered list of tasks that have been completed, with each item in the list formatted as "DONE - [task description]". The items in the list are separated by newline characters.

(async) runStagingTool()

The function runs a staging tool with specified arguments and returns the result or an error message.

Source:
Returns:

The function runStagingTool() returns different responses depending on the conditions met in the code. It can return a string response or an object response depending on the command and arguments provided. The specific response returned is indicated in the code comments.

toolsPrompt() → {string}

Displays the prompt for selecting tools.

Source:
Returns:

The tool prompt.

Type
string