Twelve Agents, One Memory: Persistent Recall for an Autonomous Agent Fleet
Architecture deep-dive by Empire Labs Pty Ltd - the team that runs a 12-agent autonomous fleet in production.
We run twelve autonomous AI agents in production. They generate leads, parse tenders, monitor compliance, review code, and keep each other honest. They run on a cron scheduler, work from the same codebase, and - until we built it properly - they forgot everything between runs.
Ask any agent "what were we working on last?" and you'd get a blank stare. Each run was a fresh context window, a fresh identity, a fresh amnesia. The first time one of our agents lost track of an external thread (a vendor application, an email sequence, a live finding), we learned the hard way that "I'll remember that" is not a memory architecture.
This is the memory stack we built instead - four layers, each with a specific job, plus the discipline rule that makes the whole thing hold together. It's what lets an agent pick up a thread days later and say, "Here's where we left off, here's what's changed, here's the blocker."
