pydantic-work 0.1.10

A local-first chat UI for Pydantic AI agents

Pydantic Chat

A local-first chat UI for your Pydantic AI agents.

uvx pydantic-work <module>:<agent variable>

# e.g. uvx pydantic-work src.my_agent:agent

What is it?

Pydantic Chat provides a beautiful web interface for interacting with your Pydantic AI agents. Your agent runs locally on your machine, and the chat UI can be accessed either via localhost or through a secure remote URL.

Important: your messages never leave your machine and never touch the internet. Remote URLs are a convenience for you to run multiple agents on your machine without picking and keeping track of ports. Anyone visiting one of your URLs will get an empty chat window and a message like below ↓

Installation

# Install with uv
uv tool install pydantic-work

# Or install in development mode
cd agent
uv pip install -e ".[cli]"

Usage

# Basic usage
pydantic-work module.path:agent_variable

# Example
pydantic-work src.golden_gate_bridge:agent

# Localhost-only mode (skip remote registration)
pydantic-work --localhost src.golden_gate_bridge:agent

# Custom port
pydantic-work --port 8000 src.golden_gate_bridge:agent

How It Works

1. Local Server: Your agent runs on your machine with a FastAPI server
2. Remote Access (Optional): On first run, you'll be prompted to choose a project slug (e.g., my-project)
3. Two URLs: Access your chat via:
   • Local: http://127.0.0.1:PORT
   • Remote: https://your-project.pydantic.work/ (if registered)

Your agent code and data never leave your machine. The remote URL just provides the frontend.

Note: The localhost UI is served via CDN (jsdelivr) at a pinned version, while the remote UI is served from the Cloudflare Worker and may be on a different version. Both UIs are compatible with the same backend API.

Example Agent

# src/my_agent.py
from pydantic_ai import Agent

agent = Agent(
    'anthropic:claude-sonnet-4-0',
    instructions="You are a helpful assistant."
)

@agent.tool_plain
def get_current_time() -> str:
    """Get the current time."""
    from datetime import datetime
    return datetime.now().strftime("%I:%M %p")

Run it:

pydantic-work src.my_agent:agent

Configuration

On first run, Pydantic Chat creates .pydantic-work/config.json in your project directory with your slug, token, and port. This folder is automatically added to .gitignore.

Patterns

Multiple Projects

Each project directory gets its own config, so you can run multiple agents with different slugs:

cd project-a && pydantic-work agent:agent  # -> project-a.pydantic.work
cd project-b && pydantic-work agent:agent  # -> project-b.pydantic.work

Troubleshooting

Failed to load agent: Check that your module path is correct (module.path:variable_name).

Registration failed: The server will automatically fall back to localhost mode. Use --localhost to skip registration entirely.

Slug already taken: Choose a different slug when prompted. Slugs are globally unique.

Links

• Pydantic AI Documentation
• Pydantic