Maintaining Conversations with Chat

While agent.prompt() is great for single, one-off questions, most interesting applications require a continuous dialogue where the agent remembers previous interactions. The SmythOS SDK handles this through the agent.chat() method.

The agent.chat() method creates a Chat instance, which automatically manages the history of messages between the user and the agent, providing the necessary context for follow-up questions.

The 03-chat.ts script demonstrates a simple in-memory conversation.

In-Memory Chat

By default, agent.chat() creates a temporary, in-memory conversation. The history is maintained for the lifetime of the chat object.

import { Agent } from '@smythos/sdk';

const agent = new Agent({
    name: 'Helpful Assistant',
    model: 'gpt-4o',
});

// Start a new conversation
const chat = agent.chat();

console.log('User: Hi, my name is Alex.');
// The agent will process this and store it in the conversation history.
const response1 = await chat.prompt('Hi, my name is Alex.');
console.log('Agent:', response1);

console.log('\nUser: What is my name?');
// Because the chat object maintains history, the agent can answer this question.
const response2 = await chat.prompt('What is my name?');
console.log('Agent:', response2); // "Your name is Alex."

// A new chat instance would have no memory of the name.
const newChat = agent.chat();
const forgottenResponse = await newChat.prompt('What is my name?');
console.log('\nAgent (new chat):', forgottenResponse); // "I'm sorry, I don't know your name."
Copy

Persistent Chat

For many applications, like a web server or a command-line tool, you need conversations to persist across different sessions. You can achieve this by providing an id and setting persist: true in the chat options. This tells the SDK to save the conversation history using the configured Storage provider.

The 04-chat-interactive-persistent.ts script shows this in action with an interactive loop.

Here's a simplified example of how it works:

// Assume 'agent' is already created.

const CHAT_ID = 'user-123-support-session';

// --- First run ---
console.log('--- Simulating first interaction ---');
// The first time we use this ID, a new chat history is created and persisted.
const chat1 = agent.chat({ id: CHAT_ID, persist: true });
const initialResponse = await chat1.prompt('Hello, I need help with my account.');
console.log('Agent:', initialResponse);
console.log('--------------------------------\n');

// --- Later, in a separate process or run ---
console.log('--- Simulating a follow-up interaction ---');
// By using the same ID, the SDK loads the previous conversation history.
const chat2 = agent.chat({ id: CHAT_ID, persist: true });
const followupResponse = await chat2.prompt('I mentioned my issue earlier, can you check the status?');
console.log('Agent:', followupResponse); // The agent will have context about the account issue.
console.log('--------------------------------');
Copy

When persist is enabled, every chat.prompt() call will automatically save the new user message and the agent's response to storage, ensuring a seamless conversational experience for your users across multiple interactions.

Next Steps

You've learned how to create stateful, conversational agents. Now let's look at how you can define and manage your agents declaratively using smyth Files.