FocusFlow - An Adaptive AI Productivity Agent

Inspiration

Planning a busy day is easy in theory but difficult in practice. Deadlines overlap, priorities shift, and one missed task can disrupt an entire schedule. I built FocusFlow to explore how AI agents can make personal productivity more adaptive, helping people turn everyday requests into actionable plans rather than just generating advice.

What It Does

FocusFlow is a lightweight productivity web application where users can create and manage tasks, organize their day, track progress, and adapt their schedules when plans change. Its conversational interface supports natural-language task management, while its scheduling logic helps organize tasks around deadlines and priorities.

The application includes a simulated voice-assistant experience, clearly labeled as a simulation, and a mock mode that enables local demonstrations without requiring cloud AI credentials.

How It Works

FocusFlow uses Next.js and TypeScript for its frontend, FastAPI for its backend, and SQLite for persistent task storage. A self-hosted Model Context Protocol (MCP) server exposes task-management functionality over Streamable HTTP, allowing compatible clients to interact with the application's tools.

The architecture is designed to support Amazon Bedrock for cloud-based AI reasoning. However, live Bedrock execution has not yet been verified, so the current demo relies on deterministic mock-mode behavior where cloud AI is unavailable.

Challenges and Learnings

A key challenge was connecting conversational interactions to real application state while keeping task operations reliable. Implementing a tool-based architecture also highlighted the importance of validating inputs, testing API behavior, and distinguishing successful connectivity from verified end-to-end execution.

Building FocusFlow helped me understand the difference between a chatbot and an agent-oriented application, particularly how APIs, persistent storage, scheduling logic, and MCP can work together.

Current Status

FocusFlow is a locally runnable, single-user demo. The MCP connectivity handshake has been verified, and mock mode supports demonstration without AWS credentials. Live Bedrock execution, actual Alexa+ deployment, speech recognition, text-to-speech, and production deployment are not currently verified or implemented.

The next steps are to complete manual UI testing, verify real model invocation when credentials are available, and improve the application based on testing and feedback.

Built With

• ai
• fastapi
• generative-ai
• mcp
• model-context-protocol
• next.js
• open-source
• productivity
• python
• react
• rest-api
• sqlite
• streamable-http
• tailwind-css
• task-management
• typescript