OmniMind: Distributed Agentic Memory

An AWS-powered AI agent using CockroachDB's distributed SQL for persistent, row-level TTL memory. It enables highly contextual, long-term conversational recall across multiple cloud regions.

Comment

Inspiration

Modern AI agents often suffer from "amnesia" or rely on centralized, single-region databases that introduce high latency for global users. We wanted to build a globally distributed memory system where an AI agent can maintain long-term contextual recall across continents without compromising speed or data residency.

What it does

OmniMind is a globally distributed AI memory system. It uses an AWS-powered AI agent network that writes conversational context (user preferences, events, temporary states) directly to a unified CockroachDB cluster spanning multiple cloud regions.

It features a custom Persistent Memory Console UI that tracks two types of agentic memory:

  • Pinned Memories: Core user preferences that persist permanently.
  • Expiring Context: Temporary conversational states managed entirely by CockroachDB's Row-Level TTL (Time to Live), which automatically purges outdated context without heavy background sweeping.

How we built it

  • Database: We utilized CockroachDB Serverless to act as our Distributed SQL engine, taking advantage of its multi-region survivability and native Row-Level TTL functionality.
  • Cloud Infrastructure: The AI Agent processing layer is hosted on AWS across us-east and eu-central regions to ensure low-latency interactions globally.
  • Backend & Frontend: The orchestration layer is built with Python/Flask, serving a sleek, dark-mode desktop UI constructed with HTML5, CSS3, and JavaScript.

Challenges we ran into

Orchestrating state synchronization across multi-region AWS deployments while ensuring CockroachDB's TTL configurations seamlessly aligned with our Python agent's logic was tricky. We had to carefully structure our SQL schema to ensure the TTL expirations didn't prematurely wipe out critical conversational context mid-session.

Built With

Share this project:

Updates

posted an update —

OmniMind is officially submitted for the CockroachDB × AWS Hackathon!

I just wrapped up the final build for OmniMind, a globally distributed AI memory system. I am incredibly proud of the custom Persistent Memory Console UI I built from scratch to visualize how CockroachDB's Row-Level TTL can seamlessly manage agentic context and expiring memories without heavy background sweeps.

Check out the architecture diagram and the full codebase in the project gallery. This was an amazing learning experience combining AWS orchestration with the power of Distributed SQL!

Log in or sign up for Devpost to join the conversation.