Memory is tricky. Too much, and your agents drown in irrelevant context. Too little, and they lose the thread on long-running projects. We just shipped a set of enhancements to Strug Recall that thread this needle: archive controls, salience-based filtering, and search across both active and archived memories.
What Changed
The core addition is lifecycle management. Memories now have an explicit archived state, controlled through both the backend router (lib/agent/routers/memory.py) and a new UI in the Strug Recall browser (src/app/dashboard/memory/page.tsx). Teams can now configure archive policies, manually archive low-value memories, and—most importantly—search across both active and archived sets without polluting day-to-day agent context.
We also added salience controls. Not all memories are created equal. A high-confidence insight about your deployment pipeline deserves different treatment than a transient note about a one-off bug fix. The new filtering lets agents and humans alike prioritize what matters.
Why It Matters
Autonomous systems accumulate knowledge fast. Without lifecycle management, memory becomes a liability—agents waste tokens on stale context, and humans lose trust in the recall surface. This update gives both parties control. Archive what's done. Surface what's relevant. Keep the history when you need it.
The audit tests we added (ArchiveConfig.audit.test.tsx) ensure that archive actions are logged and reversible. This isn't just feature work—it's infrastructure for long-term agent reliability.
What's Next
This is the foundation, not the finish line. We're already looking at automatic archival policies based on memory age, access frequency, and confidence decay. Imagine agents that proactively retire low-salience memories after 30 days of inactivity, or that surface archived context only when explicitly relevant to a new mission.
We're also considering memory snapshots—checkpoints that let you roll back to a known-good state before a major refactor or deployment. Think of it as version control for organizational knowledge.
If you're running Strug Works, the new archive controls are live in Strug Recall now. If you're curious about how we're thinking about agent memory at scale, we'd love to hear from you.