Enterprise AI in Production

Production AI agents need more than retrieval and a larger context window. Once an agent starts returning to the same user, acting on company data, and carrying context across sessions, memory becomes production state. That state needs boundaries: what is personal to a user, what is shared across the organization, who can change it, and how teams can inspect what happened when the agent behaves unexpectedly.

This session looks at agent memory as an infrastructure problem rather than a prompt trick. I will walk through a practical design for durable agent memory using explicit tools, scoped user/shared knowledge, revision history, access logs, and controlled context injection. I will use MemexAI, an open-source Postgres-backed memory layer for agents, as the demo system to make the patterns concrete while keeping the focus on design trade-offs teams can apply in their own stacks.

Takeaways:

• How to think about agent memory as governed production state, not just chat history or vector search.
• A practical architecture for inspectable memory: scoped paths, explicit memory tools, revisions, access logs, and safe prompt injection.

This session is useful for AI platform teams, enterprise architects, fullstack engineers, data/ML engineers, product engineers building internal copilots, and leaders evaluating how to move agent prototypes into production safely.

Sooraj Sanker is a Bengaluru-based fullstack and AI engineer, and the founder of MemexAI, an open-source Postgres-backed memory layer for long-horizon AI agents. He spent roughly six years at Hasura and PromptQL, where he worked on conversational analytics, AI agent interfaces, tool-based workflows, artifact-aware chat, and state management for distributed agent execution. His work sits at the intersection of agent infrastructure, context engineering, data UX, and production reliability.