Similar Issue. While executing Agent. What I found is that ML common plugins memory index got corrupted. Indexes ( .plugins-ml-memory-meta & .plugins-ml-memory-messages) are not accessible to agent hence failing.
[2025-06-20T08:53:07,302][DEBUG][o.o.m.t.MLTaskRunner ] [ip-10-37-4-61] Execute ML request 32de7f34-eede-4b26-a4b6-045d672d3bae locally on node xpuzY8k1RimsoxphGsYRiQ
[2025-06-20T08:53:07,304][DEBUG][o.o.m.e.a.a.MLAgentExecutor] [ip-10-37-4-61] Completed Get Agent Request, Agent id:ZKxtjJcBI4w65yUkV-_b
[2025-06-20T08:53:07,311][ERROR][o.o.m.e.a.a.MLAgentExecutor] [ip-10-37-4-61] Failed to read conversation memory
java.lang.NullPointerException: null
at com.google.common.base.Preconditions.checkNotNull(Preconditions.java:903) ~[?:?]
at org.opensearch.ml.engine.memory.MLMemoryManager.createInteraction(MLMemoryManager.java:110) ~[?:?]
at org.opensearch.ml.engine.memory.ConversationIndexMemory.save(ConversationIndexMemory.java:116) ~[?:?]
at org.opensearch.ml.engine.algorithms.agent.MLAgentExecutor.saveRootInteractionAndExecute(MLAgentExecutor.java:300) ~[?:?]
at org.opensearch.ml.engine.algorithms.agent.MLAgentExecutor.lambda$execute$2(MLAgentExecutor.java:234) ~[?:?]
Can we delete these indexes to reset ? Is there any other ways to reset memory for Agents ?
PS: Observation, it does not matter which agent in used, even after creating new agents it seems there is issue with the Memory access process.