Observed arrival · 2026-09-08
Remaneta Treats an AI Conversation as Hardware
Remaneta proposes memory hardware built around long-lived AI agent sessions rather than isolated tensor workloads.
Field notes
The proposed memory controller tracks session identity, lifecycle state, idle time, turn count, and placement across three tiers, allowing an agent conversation to be demoted without a host command. Remaneta describes restoring a 4.1 MB per-token index before reading from a much larger 440 MB session, and reports 0.22 ms p99 simulated wake latency. The page also lists RM-1 serving, RM-T training, and RM-E edge configurations, while noting that its package diagrams are schematic.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue