Building a Conversational Assistant

This command-line assistant keeps conversation history in a named thread. Install and configure Marvin first.

import marvin

assistant = marvin.Agent(
    name="Support",
    instructions="Answer briefly. Ask for missing details rather than inventing them.",
)
thread = marvin.Thread(id="support-demo")

while True:
    message = input("You: ").strip()
    if message.lower() in {"exit", "quit"}:
        break
    if message:
        response = assistant.say(message, thread=thread)
        print(response)

Restarting with the same thread ID and database resumes the history. Use a distinct thread ID for each conversation in an application. The agent can refer to earlier messages in this thread. For facts that should be searchable across conversations, attach a memory with a persistent provider. A thread is enough for this example; the default vector-memory provider is ephemeral. To expose external data, add a real Python function to the agent’s tools. A prompt that asks for current weather or web research does not provide access to those sources. For async applications, use await assistant.say_async(message, thread=thread).

Get Started

Concepts

Functions

Patterns

Guides

Community

Building a Conversational Assistant