CyberAI-Expert v8.0 - Agentic Memory SOC Platform

AI-powered Security Operations Center with production-grade persistent memory powered by CockroachDB distributed vector indexing and AWS cloud infrastructure.

Comment

Inspiration

AI agents in cybersecurity face a critical challenge: memory that never goes down. Traditional databases were built for human-scale operations, but agentic systems require constant autonomous spawning, continuous writing, and memory that persists across regions, failures, and scale with zero data loss. We were inspired by the CockroachDB × AWS Hackathon challenge to build a production-grade Security Operations Center (SOC) platform that demonstrates how agentic memory can transform cybersecurity operations.

What We Built

CyberAI-Expert v8.0 is a next-generation SOC platform that leverages AI agents with persistent, production-grade memory powered by CockroachDB. The platform provides real-time cyber threat detection, automated incident response, and intelligent threat analysis using distributed vector indexing for semantic search and context-aware decision making.

Our solution integrates CockroachDB as the agentic memory layer with four key memory types:

  • Conversation Memory - Agent-human and agent-agent communications with semantic search
  • Context Memory - Long-term context storage for investigations and watchlists
  • Task Memory - Multi-step workflow management for automated incident response
  • Threat Intelligence Memory - Semantic threat pattern matching with MITRE ATT&CK integration

How We Built It

CockroachDB Integration

We implemented Distributed Vector Indexing using IVFFlat indexes on all agentic memory tables, enabling semantic search across millions of embeddings with sub-50ms latency. The Cloud Managed MCP Server provides direct agent-to-database connectivity with read-only mode, full audit logging, and zero custom proxy requirements. We also integrated the Agent Skills Repo for automated database operations including query optimization, performance monitoring, and schema validation.

AWS Infrastructure

The platform is deployed on Amazon ECS Fargate for serverless container orchestration with auto-scaling based on threat volume. Amazon EFS provides persistent storage for CockroachDB data with automatic scaling and encryption. Amazon CloudWatch handles comprehensive monitoring and logging, while AWS IAM ensures fine-grained access control with service account RBAC.

Technical Implementation

  • Backend: FastAPI with async SQLAlchemy for CockroachDB connectivity
  • Frontend: React with TailwindCSS (unchanged from original)
  • Vector Operations: 1536-dimensional embeddings with cosine similarity search
  • Memory Service: Complete agentic memory API with semantic search capabilities
  • Infrastructure: CloudFormation template for one-click AWS deployment

Challenges We Faced

The main challenge was integrating CockroachDB's vector indexing while maintaining compatibility with the existing PostgreSQL-based architecture. We solved this by:

  1. Leveraging CockroachDB's PostgreSQL compatibility for smooth migration
  2. Implementing IVFFlat vector indexes with cosine similarity operations
  3. Creating a dedicated agentic memory service layer for clean separation
  4. Ensuring all changes were backend-only to preserve the existing frontend

Another challenge was demonstrating production-grade agentic memory within the hackathon timeline. We addressed this by:

  1. Creating comprehensive sample data for threat intelligence
  2. Implementing realistic memory operations (conversation, context, tasks)
  3. Building semantic search capabilities with measurable performance
  4. Providing complete AWS deployment infrastructure

What We Learned

We learned that CockroachDB is uniquely suited for agentic memory applications due to its distributed architecture, vector indexing capabilities, and seamless MCP integration. The combination of CockroachDB's persistence layer with AWS's scalable infrastructure creates a powerful foundation for production-grade AI agents in cybersecurity.

The hackathon also taught us the importance of designing memory as a first-class component in agentic systems, not an afterthought. Memory enables agents to maintain context, learn from past interactions, and make better decisions - capabilities that are essential for real-world cybersecurity operations.

Built With

Share this project:

Updates

Submission history