salesforce-agent-api-client v1.0.1
Node client for the Salesforce Agent API
See the API documentation and the Postman collection for more information on the Salesforce Agent API.
Quick Start Example
Here's an example that will get you started quickly with streaming events.
Install the client library and
dotenv
:npm install salesforce-agent-api-client dotenv
Create a
.env
file at the root of your project and configure it with the the following template:instanceUrl= clientId= clientSecret= agentId=
Create an
index.js
file with the following code:import * as dotenv from 'dotenv'; import AgentApiClient from 'salesforce-agent-api-client'; // Hard-coded prompt used for a single demo run const SAMPLE_PROMPT = 'What does "AI" stand for?'; // Load config from .env file dotenv.config(); const config = { instanceUrl: process.env.instanceUrl, clientId: process.env.clientId, clientSecret: process.env.clientSecret, agentId: process.env.agentId }; // Configure Agent API client const client = new AgentApiClient(config); // Authenticate await client.authenticate(); // Prepare SSE stream event handler function streamEventHandler({ data, event }) { const eventData = JSON.parse(data); console.log('Event: %s', event); console.log(JSON.stringify(eventData, null, 2)); // TODO: add custom logic to process events } // Prepare SSE stream disconnect handler async function streamDisconnectHandler() { // On disconnect, close the session await client.closeSession(sessionId); } // Create a new session const sessionId = await client.createSession(); const variables = []; try { // Sends an streaming message const eventSource = client.sendStreamingMessage( sessionId, SAMPLE_PROMPT, variables, streamEventHandler, streamDisconnectHandler ); } catch (error) { console.log(error); await client.closeSession(sessionId); }
Run the code with
node index.js
.If everything goes well, the output should look like this:
Agent API: authenticated on https://coralcloudresorts19-dev-ed.develop.my.salesforce.com (API endpoint: https://api.salesforce.com) Agent API: created session a4923398-0d60-4529-9f7a-91f021409875 Agent API: sending async message 1740068546539 with text: What does AI stand for? Event: INFORM { "timestamp": 1740068551742, "originEventId": "1740068546968-REQ", "traceId": "66b8d7d8f3aac7bb404730970c88659d", "offset": 0, "message": { "type": "Inform", "id": "f5d3d83f-3bc7-4d81-9f8f-4b7e75522aa3", "feedbackId": "12636229-b716-4fcd-ba0b-6a498e27caab", "planId": "12636229-b716-4fcd-ba0b-6a498e27caab", "isContentSafe": true, "message": "How can I assist you with any customer support issues today?", "result": [], "citedReferences": [] } } Event: END_OF_TURN { "timestamp": 1740068551746, "originEventId": "1740068546968-REQ", "traceId": "66b8d7d8f3aac7bb404730970c88659d", "offset": 0, "message": { "type": "EndOfTurn", "id": "696fd6f5-53bc-4366-b37c-fbf0a74ce992" } } SSE disconnected. Preventing auto reconnect. Agent API: closed session a4923398-0d60-4529-9f7a-91f021409875
Configuration
Object that describes the client configuration:
Name | Type | Description |
---|---|---|
instanceUrl | string | Your Salesforce Org domain. |
clientId | string | Connected app client ID. |
clientSecret | string | Connected app client secret. |
agentId | string | Agent ID. |
Logging
The client uses debug level messages so you can lower the default logging level if you need more information.
The documentation examples use the default client logger (the console). The console is fine for a test environment but you'll want to switch to a custom logger with asynchronous logging for increased performance.
You can pass a logger like pino in the client constructor:
import pino from 'pino';
const config = {
/* your config goes here */
};
const logger = pino();
const client = new PubSubApiClient(config, logger);
Reference
AgentApiClient
Client for the Salesforce Agent API
AgentApiClient(configuration, [logger])
Builds a new Agent API client.
Name | Type | Description |
---|---|---|
configuration | Configuration | The client configuration (authentication...). |
logger | Logger | An optional custom logger. The client uses the console if no value is supplied. |
async authenticate() → {Promise.<void>}
Authenticates with Salesforce.
Returns: Promise that resolves once the client is authenticated.
async createSession() → {Promise.<string>}
Creates an agent session.
Returns: Promise that holds the session ID.
async sendSyncMessage(sessionId, text, variables = []) → {Promise.<any>}
Sends a synchronous prompt to the agent.
Name | Type | Description |
---|---|---|
sessionId | string | An agent session ID. |
text | string | The prompt sent to the agent. |
variables | Object[] | Optional context variables. |
Returns: Promise that holds the agent's response.
async sendStreamingMessage(sessionId, text, variables = [], onMessage, onDisconnect) → EventSource
Sends an asynchronous prompt to the agent.
Name | Type | Description |
---|---|---|
sessionId | string | An agent session ID. |
text | string | The prompt sent to the agent. |
variables | Object[] | Context variables. |
onMessage | MessageCallback | Message callback function. |
onDisconnect | function() | Optional disconnect callback function. |
Returns: a SSE event source. See EventSource for implementation details.
MessageCallback(Object: message)
is a callback function from EventSource
. Notable message
properties are as follow:
Name | Type | Description |
---|---|---|
event | string | Event type. One of INFORM , ERROR or END_OF_TURN . |
data | string | The event data. Can be empty. |
async closeSession(sessionId) → {Promise.<void>}
Closes the agent session.
Name | Type | Description |
---|---|---|
sessionId | string | An agent session ID. |
Returns: Promise that resolves once the session is closed.
async submitFeedback(sessionId, feedbackId, feedback, feedbackText) → {Promise.<void>}
Submits feedback to the agent.
Name | Type | Description |
---|---|---|
sessionId | string | An agent session ID. |
feedbackId | string | Feedback ID. |
feedback | string | feedback type (GOOD or BAD ) |
feedbackText | string | Optional feedback text |
Returns: Promise that resolves once the feedback is saved.