Agents

An Agent groups a model, instructions, tools, and optional memory. Calling agent.run creates and runs a task with that agent.

import marvin

editor = marvin.Agent(
    name="Editor",
    instructions="Edit for clarity. Preserve the author's meaning.",
)
revision = editor.run(
    "Edit the supplied sentence",
    context={"sentence": "We are in receipt of your message."},
)

Assign an agent to a task

import marvin

editor = marvin.Agent(name="Editor", instructions="Write concise prose.")
task = marvin.Task(
    "Rewrite the supplied sentence",
    agents=[editor],
    context={"sentence": "We are in receipt of your message."},
)
revision = task.run()

Use agents=[editor] with marvin.run or Task, and agent=editor with helpers such as cast and extract.

Models and tools

Set model="provider:model-name" or pass a Pydantic AI model instance. Pass ordinary Python functions in tools=[...]. See model configuration and tools and context. An agent does not store conversation history itself. Reuse a thread to retain messages across calls, or attach memory for searchable facts. For multiple agents, coordinate their calls explicitly in Python. The team classes are deprecated.

Get Started

Concepts

Functions

Patterns

Guides

Community

Assign an agent to a task

Models and tools

​Assign an agent to a task

​Models and tools

Assign an agent to a task

Models and tools