Getting started

reflex-agentkit needs Python 3.11, 3.12 or 3.13. The package is reflex-agentkit; the import is reflex_agentkit. The agent frameworks are optional extras:

Extra	Installs	For
(none)	Reflex, ag-ui-protocol, httpx	AG-UI servers over HTTP (HttpAGUIClient), your own AGUIAgent
pydantic-ai	pydantic-ai-slim[ag-ui,anthropic,openai]	PydanticAIAgent
langgraph	ag-ui-langgraph, langgraph	LangGraphAGUIAgent
all	both

Install

uvpip

uv init my-agent-app && cd my-agent-app
uv add "reflex-agentkit[pydantic-ai]"

python -m venv .venv && . .venv/bin/activate
pip install "reflex-agentkit[pydantic-ai]"

Pre-releases

Until 0.1.0 is out, add --prerelease=allow (uv) or --pre (pip) to get the latest pre-release.

A first app

Create a Reflex app (uv run reflex init, or reflex init with pip, and pick the blank template), then replace the app module with this:

from typing import ClassVar

import reflex as rx
from pydantic_ai import Agent

from reflex_agentkit import AgentState, agent_chat
from reflex_agentkit.adapters.pydantic_ai import PydanticAIAgent

travel_agent = Agent(
    "anthropic:claude-sonnet-5-5",
    instructions="You are a helpful travel assistant.",
)


@travel_agent.tool_plain
def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    return f"It is sunny and 22 °C in {city}."


@travel_agent.tool_plain(requires_approval=True)
def book_hotel(city: str, nights: int) -> str:
    """Book a hotel room. Requires human approval."""
    return f"Booked {nights} nights in {city}."


class ChatState(AgentState, rx.State):
    agent: ClassVar = PydanticAIAgent(travel_agent)


def index() -> rx.Component:
    return agent_chat(ChatState)


app = rx.App()
app.add_page(index)

The components are built on Radix Themes, so enable the plugin in rxconfig.py:

import reflex as rx

config = rx.Config(
    app_name="my_agent_app",
    plugins=[rx.plugins.RadixThemesPlugin()],
)

Run it with your Anthropic key:

ANTHROPIC_API_KEY=sk-ant-... uv run reflex run

Ask it to plan a trip. The get_weather call appears as a card, and the approval dialog opens before book_hotel runs. You can approve, edit the arguments or reject the call.

No API key?

Use a local model: see Local models with Ollama. To check that the app compiles with no model at all, PydanticAI's TestModel works too.

The three pieces

1. The agent. Anything that produces AG-UI events: a PydanticAI agent wrapped in PydanticAIAgent, a LangGraph graph wrapped in LangGraphAGUIAgent, or an AG-UI server reached through HttpAGUIClient.
2. The state. A class that mixes AgentState into rx.State. Configuration such as agent is class-level and annotated ClassVar, so it never becomes a state var that Reflex would send to the browser.
3. The UI. agent_chat(ChatState), or your own components reading the state's vars and calling its event handlers (Your own UI).

Next steps

• Concepts explains runs, threads, tool calls, interrupts and shared state.
• Human in the loop covers approvals and questions.
• The Dojo examples show every feature with PydanticAI and with LangGraph.