LLM-driven agents need to remember prior plans so they're consistent on similar tasks. Mneme provides task-description embedding → plan lookup with confidence-gated reuse. Two agents (alice, bob) get separate namespaces so their memories don't bleed.
LLM endpoint: {{ spark_url }} · model: {{ model }}
Re-run the same task on the same agent → exact-match cache hit. Try a paraphrase → semantic match. Switch agents → miss (separate memory).
| # | Agent | Layer | Latency | Task |
|---|