Files, Context Windows, or Memory: Choosing Persistence for Production Agents with Memori Labs

September 2 - 11 a.m. ET

Building a proof-of-concept agent is easy. Building one that can safely remember, reason, and improve over thousands of real interactions is not. The hard part isn’t the model—it’s where the agent’s memory lives, how it’s governed, and how much it costs to run in production.

In this session, Mikiko Bazeley, Staff Applied AI Engineer at MongoDB and Adam Struck, Co-Founder and CEO of Memori Labs will walk through the full agent memory lifecycle, exploring three concrete options for persistence: bigger context windows, files on disk, and a dedicated memory system running on MongoDB. You’ll see where each approach works, where it breaks, and how to pick the right one for your own workloads.

You’ll learn how to:

• Decide when you actually need a memory system at all, using clear break signals around task horizon, corpus size, tenancy, and cost.
• Compare context windows, file-based memory, and a dedicated memory layer across concurrency, isolation, retrieval precision, and governance.
• Design memory around the full lifecycle loop—from state inside a single run to durable, decaying, cross-session knowledge.
• Implement MongoDB-backed agent memory in practice, with a live build of Memori Labs writing trace and conversational memory into an Atlas cluster you already run.
• Understand real benchmark results, including how Memori Labs’ approach delivers up to 20x token savings at near full-context accuracy on a public agent-memory benchmark.

You’ll also see how Memori Labs and MongoDB together address the production concerns teams hit first—supersession, per-tenant isolation, auditability, retention, and deletion—so memory becomes an asset instead of a liability.

Join to learn how to instrument, measure, and evolve memory for your next generation of AI agents.