Cult Memory

Inspiration

Inspiration AI agents are becoming more capable, but many still treat memory as temporary context rather than durable application state. I wanted to explore what happens when an agent can remember previous decisions, user context and task state across sessions — while still operating within explicit capability and human-approval boundaries. What it does Cult Memory is an agentic application powered by the Cult runtime. It stores persistent memory in CockroachDB, retrieves relevant context for future tasks, and uses that memory before deciding what action to take. The agent can remember user preferences, previous instructions, decisions and task state across sessions. Sensitive actions can require explicit human approval before execution. How I built it The application uses Cult as the agent/runtime layer, CockroachDB as the persistent memory system, and AWS for deployment/execution. CockroachDB stores structured memories and semantic context so the agent can retrieve information from previous sessions instead of starting from zero every time. The project is designed to demonstrate a complete loop: request → retrieve memory → reason → choose action → approval if required → execute → store new memory CockroachDB The project uses CockroachDB as the system of record for agent memory. The implementation will use at least two hackathon-supported CockroachDB tools, including Distributed Vector Indexing for semantic memory retrieval and one additional CockroachDB agent tool for database interaction/operations. AWS The agent application is deployed using AWS infrastructure so the memory-backed agent can run as a real hosted service rather than only as a local demo. Challenges The main challenge is making memory useful rather than simply storing conversation history. The agent needs to determine what should be remembered, retrieve only relevant context, and avoid letting stale or irrelevant memories distort future decisions. Another challenge is combining autonomous behavior with explicit capability boundaries and human approval. What I learned Persistent agent memory changes the architecture of an AI application. Once an agent remembers previous state, reliability, access control, retrieval quality and auditability become much more important. What's next I want to expand Cult Memory into a reusable persistent-memory pattern for applications built with Cult, including multi-agent workflows, task history, richer semantic retrieval and longer-running autonomous systems.

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Cult Memory

Built With

• agentic
• agents
• ai
• amazon-web-services
• cockroachdb
• cult
• distributed
• go
• indexing
• memory
• persistent
• search
• vector