Agentbook

An adaptive AI study workspace that remembers what you learn, identifies where you struggle, and turns persistent learning memory into grounded guidance and confirmed study actions.

Comment

Inspiration

Most AI study tools forget the learner between sessions. We wanted memory to directly improve what the AI recommends next.

What it does

Agentbook connects study materials, quiz mistakes, learner memory, and tasks to provide grounded, personalized study guidance.

How we built it

We used React, FastAPI, CockroachDB, distributed vector search, Managed MCP, ccloud CLI, and AWS.

Challenges we ran into

Designing safe agent actions, persistent learner memory, and strict workspace isolation without turning the project into a simple RAG chatbot.

Accomplishments that we're proud of

We built a complete learning loop where mistakes become memory, memory changes recommendations, and agent actions require explicit confirmation.

What we learned

Good agentic systems need reliable memory, clear tool boundaries, and safe actions—not just better prompts.

What's next for Agentbook

Public deployment, stronger long-term personalization, richer learning analytics, and expanded agent workflows.

Built With

Share this project:

Updates