Pydantic AI Backend

File storage, sandbox backends, and console toolset

Multiple backend options (in-memory, local filesystem, Docker sandbox) with a console toolset (ls, read, write, edit, grep, execute) and fine-grained permission system.

Installation

pip install pydantic-ai-backend

Choose from three backend options depending on your needs: StateBackend (in-memory, ephemeral — perfect for tests), LocalBackend (real filesystem with a sandboxed root directory), or DockerSandbox (full Docker container isolation with pre-configured runtimes for Python, Node.js, and data science). The console toolset provides 7 tools (ls, read_file, write_file, edit_file, glob, grep, execute) with a permission system offering 4 presets: DEFAULT, PERMISSIVE, READONLY, and STRICT.

Features

Quick Start

from dataclasses import dataclass
from pydantic_ai import Agent
from pydantic_ai_backends import LocalBackend, create_console_toolset

@dataclass
class Deps:
    backend: LocalBackend

agent = Agent(
    "openai:gpt-4o",
    deps_type=Deps,
    toolsets=[create_console_toolset()],
)

backend = LocalBackend(root_dir="./workspace")
result = agent.run_sync(
    "Create a Python script that calculates fibonacci and run it",
    deps=Deps(backend=backend),
)

Use Cases