agent-state-bridge

agent-state-bridge

Full-stack bridge for sharing app state between frontend and AI agents. Includes React hooks (npm) and Python backend utilities (PyPI) for FastAPI, Flask, and Django.

🚀 Build AI-powered apps with seamless state synchronization. New in v1.0.0: Centralized custom hooks, {messages, actions, context} model is now required, breaking changes, RAG-ready architecture. See CHANGELOG.md.

📦 Packages

This monorepo contains two complementary packages:

Package	Platform	Description
agent-state-bridge	npm	React hooks and UI components for frontend
agent-state-bridge	PyPI	Python utilities for FastAPI, Flask, Django

🎯 Frontend (React/TypeScript)

Installation

npm install agent-state-bridge

Quick Start (v1.0.0)

1. Centralized custom hook pattern

See the examples for the recommended pattern. All agent integration logic (chat, actions, context, CRUD) is now centralized in a single custom hook (e.g., useTodoAgent).

// Example: useTodoAgent.jsx
import { useAgentChat, useAgentCRUD } from "agent-state-bridge";
import { useTodos } from "../context/TodoContext.jsx";

export const useTodoAgent = () => {
  const { todos, addTodo, toggleTodo, deleteTodo, getStats } = useTodos();
  const { messages, sendMessage, loading, error } = useAgentChat({
    getContext: () => ({
      todos: todos.map((t) => ({ id: t.id, text: t.text, done: t.done })),
      summary: getStats(),
    }),
    getActions: () => [],
    onActionsReceived: (actions) => {
      actions.forEach((action) => {
        switch (action.type) {
          case "post":
            if (action.payload?.text) addTodo(action.payload.text);
            break;
          case "put":
            if (action.payload?.id) toggleTodo(action.payload.id);
            break;
          case "delete":
            if (action.payload?.id) deleteTodo(action.payload.id);
            break;
        }
      });
    },
    initialMessages: [],
  });
  // ...useAgentCRUD for each action...
  return { messages, sendMessage, isLoading: loading, error };
};

2. Use the hook in your UI

import { useTodoAgent } from "../hooks/useTodoAgent";

export const AgentChatContainer = () => {
  const { messages, sendMessage, isLoading, error } = useTodoAgent();
  // ...
  return (
    <AgentChatSidebar
      messages={messages}
      onSend={sendMessage}
      loading={isLoading}
      error={error}
      // ...
    />
  );
};

New in v1.0.0:

• All agent logic must be centralized in a custom hook (see examples)
• {messages, actions, context} model is required
• No more direct use of removed hooks/components (see CHANGELOG)

Features

✅ State-agnostic: Works with Zustand, Redux, useState, useContext, etc. ✅ Centralized: All agent logic in one hook ✅ UI included: Pre-built chat component with markdown support ✅ Customizable: Use hooks only or customize the UI ✅ TypeScript: Fully typed

📖 Full Frontend Documentation →

🐍 Backend (Python)

Installation

# For FastAPI (recommended)
pip install agent-state-bridge[fastapi]

# For Flask
pip install agent-state-bridge[flask]

# For Django
pip install agent-state-bridge[django]

Quick Start

FastAPI

from fastapi import FastAPI
from agent_state_bridge.fastapi import create_agent_router
from agent_state_bridge.models import AgentResponse, Message, Action

async def my_agent(messages: list[Message], actions: list[Action], context: dict) -> AgentResponse:
    """
    v1.0.0 model: {messages, actions, context}
    - messages: Conversation history
    - actions: Recent CRUD operations (post, put, delete)
    - context: App state + RAG data
    """
    cart_items = context.get("cart", {}).get("items", [])
    last_msg = messages[-1].content if messages else ""

    response_text = f"You have {len(cart_items)} items. You said: {last_msg}"

    # Optionally return actions for the frontend to execute
    return AgentResponse(
        response=response_text,
        actions=[Action(type="post", payload={"product": "suggested_item"})],  # Optional
        context={"updated": "data"}  # Optional
    )

app = FastAPI()
router = create_agent_router(my_agent, tags=["agent"])
app.include_router(router)

Why this model?

• ✅ Separates state mutations (actions) from context
• ✅ Ready for RAG: Add vector search results to context
• ✅ Bidirectional actions: Agent can return actions for frontend to execute
• ✅ Scalable: Easy to add query agents and semantic search

Flask

from flask import Flask
from agent_state_bridge.flask import create_agent_blueprint

def my_agent(message: str, state: dict) -> str:
    return f"Processed: {message}"

app = Flask(__name__)
bp = create_agent_blueprint(my_agent)
app.register_blueprint(bp)

Django REST Framework

from agent_state_bridge.django import agent_api_view

@agent_api_view
def my_agent(message: str, state: dict) -> str:
    return f"Processed: {message}"

Agent Framework Support

✅ LangChain: Full async support
✅ Microsoft Agent Framework: Azure AI integration
✅ CrewAI: Multi-agent orchestration
✅ Custom agents: Bring your own logic

📖 Full Backend Documentation →
📁 Examples with LangChain, Agent Framework, etc. →

🚀 Complete Example (see examples/)

See the examples/ directory for full working codebases with the new v1.0.0 pattern.

📚 API Reference

See the examples/ and CHANGELOG.md for the latest API usage and migration notes. The previous hooks/components like useAgentChat and AgentChatSidebar have been removed or replaced by centralized custom hooks.

🏗️ Architecture

graph LR
    A[React App] -->|HTTP POST| B[Backend API]
    A -->|state + message| B
    B -->|response| A
    B --> C[Agent Logic]
    C --> D[LangChain/Agent Framework/Custom]

Key principles:

• Frontend is the source of truth for state
• Backend is stateless (no session storage)
• Separation of concerns: Actions (CRUD) vs Context (state + RAG)
• Messages, actions, and context sent with every request
• Agent can return actions for frontend to execute
• Ready for RAG: Add vector search results to context
• Works with any AI framework

📖 Architecture Documentation →

📚 Examples

Check out complete working examples in the examples/ directory:

• Shopping Cart - Full-stack e-commerce with AI assistant
  • React + Zustand + LangChain
  • Demonstrates {messages, actions, context} model
  • RAG-ready architecture

View all examples →

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

MIT License - see LICENSE for details

🔗 Links

• npm package: npmjs.com/package/agent-state-bridge
• PyPI package: pypi.org/project/agent-state-bridge
• GitHub: github.com/SergioCantera/agent-state-bridge
• Issues: github.com/SergioCantera/agent-state-bridge/issues

Made with ❤️ for the AI development community