​
AmazonBedrockSupportedModels

The AmazonBedrockSupportedModels class offers constants for models supported by Amazon Bedrock, simplifying model selection in your applications.

​
Usage

import { AmazonBedrockSupportedModels } from 'xpander-sdk';

const model = AmazonBedrockSupportedModels.ANTHROPIC_CLAUDE_3_5_SONNET_20240620;
ConstantDescription
ANTHROPIC_CLAUDE_3_5_SONNET_20240620Optimized for general use.
ANTHROPIC_CLAUDE_3_HAIKU_20240307Excels in summarization tasks.
COHERE_COMMAND_RIdeal for command-based instructions and NLP.

These constants help in selecting models that support function calling, reducing the need to memorize model names and IDs.

​
NvidiaNIMSupportedModels

The NvidiaNIMSupportedModels class provides constants for NVIDIA’s NIM platform models, aiding in model selection.

​
Usage

import { NvidiaNIMSupportedModels } from 'xpander-sdk';

const model = NvidiaNIMSupportedModels.LLAMA_3_1_70B_INSTRUCT;
ConstantDescription
LLAMA_3_1_70B_INSTRUCTSuitable for large-scale instructional tasks requiring deep context understanding.

​
OpenAISupportedModels

The OpenAISupportedModels class provides constants for OpenAI models compatible with the xpander SDK.

​
Usage

import { OpenAISupportedModels } from 'xpander-sdk';

const model = OpenAISupportedModels.GPT_4;
ConstantDescription
GPT_4Latest version of OpenAI’s GPT model.
GPT_4_OCost-effective optimized version.

​
ToolResponse

The ToolResponse class encapsulates the response from a tool invoked by an AI Agent, including details like tool call ID, role, and response message.

​
Initialization

import { ToolResponse } from 'xpander-sdk';

const response = new ToolResponse(
  'tool-call-id',
  'assistant',
  'Weather Tool',
  'Here is your weather report...',
  filteredToolObject,
  'payload request data'
);

​
Methods

  • toJSON(): Serializes the ToolResponse instance to a JSON object.
  • fromJSON(json: any): Instantiates a ToolResponse from a JSON object.

​
XpanderClient

The XpanderClient is the core class for interacting with xpanderAI tools, offering methods for managing sessions and invoking tools.

​
Initialization

import { XpanderClient } from 'xpander-sdk';

const client = new XpanderClient(
  'agent-key',
  'https://agent-url',
  LLMProvider.OPEN_AI,
  localToolsArray,
  toolsArray,
  customParams
);

​
Methods

MethodDescription
addLocalTools(tools: ILocalTool[])Adds tools for specific needs.
getGraphSessionParam(param: string)Retrieves a parameter for the current session.
startSession(prompt?: string)Initiates a new session with an optional prompt.
xpanderToolCall(toolSelectorResponse: any, llmProvider?: string)Invokes tools based on the selector response.

​
LLMProvider Enum

The LLMProvider enum allows selection of the desired LLM provider for interactions.

​
Values

ValueDescription
LANG_CHAINLangChain’s LLM models.
OPEN_AIOpenAI models like GPT-4.
NVIDIA_NIMNVIDIA’s LLM models.
AMAZON_BEDROCKAmazon Bedrock’s LLM models.

​
Example Workflow

To create an AI agent workflow interacting with multiple systems, follow these steps:

  1. Initialize the Xpander Client:
import { XpanderClient, OpenAISupportedModels } from 'xpander-sdk';

const xpanderClient = new XpanderClient(
  'your-agent-key',
  'your-agent-url',
  OpenAISupportedModels.GPT_4
);
  1. Add Local Tools:
xpanderClient.addLocalTools(localToolsArray);
  1. Start a Session:
xpanderClient.startSession('Start analyzing customer support conversations.');
  1. Invoke a Tool:
const response = xpanderClient.xpanderSingleToolInvoke('sentiment-analysis-tool-id', payloadData);
console.log(response);

​
Protocols & Interfaces

To extend the xpander SDK or create custom implementations, interfaces are provided:

​
ITool

The ITool interface represents a generic tool used by AI agents, including properties like:

PropertyDescription
nameThe tool’s name.
descriptionA description of the tool’s capabilities.
funcFunction for executing the tool.
parametersParameters required for execution.

​
Example Implementation

class CustomTool implements ITool {
  name = 'Custom Sentiment Analyzer';
  description = 'Analyzes sentiment of customer support messages';
  func = (params) => {
    // Custom logic here
  };
  parameters = {
    inputText: { type: 'string', description: 'Text to analyze' },
  };
}

​
ILocalTool

The ILocalTool interface allows the creation of tools for local use, maintaining security boundaries.

SDK Local functionsGet Tools