AI Study Coach (Cloudflare Agents)

A stateful AI study planning application built on Cloudflare Agents. The system generates personalized study plans, persists the latest plan using Durable Objects, supports context-aware plan refinement across conversations, streams responses in real time, and schedules study tasks through workflow-style tool execution.

Designed to simulate how real-world AI assistants manage evolving user context and long-running workflows.

---

🧠 AI generates a personalized study plan

---

⚡ Quick-start prompts for easy interaction

---

⏰ Built-in task scheduling & reminders

---

Why this project

Most AI applications today are stateless — they lose context across interactions.

This project demonstrates how to build a stateful AI system at the edge, where:

• user context persists across sessions
• plans can be incrementally refined
• tasks can be scheduled and executed asynchronously

---

Features

• Conversational study planning using Workers AI (LLM)
• Persistent memory using Durable Objects
• Iterative plan refinement (context-aware updates)
• Task scheduling and execution (workflow support)
• Real-time streaming chat UI
• Stateful Plan Memory (WOW factor):
  • Stores the latest study plan in agent state
  • Allows retrieval via natural language ("What’s my current plan?")
  • Enables structured plan evolution across multiple turns

---

Stateful Plan Memory (Key Highlight)

Unlike traditional chat-based apps, this system persists the latest generated plan explicitly in agent state.

This enables:

• retrieving the current plan at any time
• updating plans incrementally
• maintaining continuity across interactions

Example: User: Make me a plan → Plan generated

User: Make Wednesday lighter → Plan updated

User: What’s my current plan? → Retrieved from persistent state

---

Architecture

User → Pages UI → Agent (Durable Object) → Workers AI ↳ Persistent State (latestPlan) ↳ Scheduling (tasks/workflows)

---

Example Flow

1. User: “I have an exam in 5 days…”
2. Agent generates structured study plan
3. User: “Make Wednesday lighter”
4. Agent revises plan using memory
5. User: “What’s my current plan?”
6. Agent retrieves plan from persistent state
7. User: “Remind me tomorrow at 9 AM”
8. Agent schedules task using workflow

---

Tech Stack

• Cloudflare Agents (stateful AI agents)
• Durable Objects (persistent memory)
• Workers AI (LLM inference at the edge)
• Vite + React (frontend)
• Zod (tool validation)
• AI SDK (streaming + tool execution)

---

Key Engineering Concepts

• Stateful agent design using Durable Objects
• Tool-based LLM interaction (getCurrentPlan, scheduleTask)
• Context pruning for efficient inference
• Separation of conversational memory vs persistent state
• Event-driven workflow execution (scheduled tasks)

---

Why Cloudflare

This project leverages Cloudflare’s edge-native architecture:

• Durable Objects → stateful agents
• Workers AI → low-latency inference
• Built-in scheduling → long-running workflows

It demonstrates how AI systems can evolve from:

stateless APIs → persistent, interactive agents at the edge

---

How to run locally

Clone the repository:

git clone git@github.com:ShireenMeher/cf_ai_study_planner.git

Navigate into the project directory

cd cf-ai-study-planner

Install dependencies

npm install

Start the development server

npm run dev

Open in browser

Visit: http://localhost:5173

Live Demo

Here is a link where you can access the agent live:

https://cf-ai-study-planner.shireenmeher296.workers.dev

Future Improvements

• Plan versioning (compare old vs new)
• Recurring schedules (daily study reminders)
• Browser notifications for reminders
• Multi-user plan isolation

Summary

This project demonstrates how to build production-style AI systems, not just chatbots — combining:

• memory
• workflows
• tool usage
• and edge-native execution

into a cohesive, stateful experience.