Merge pull request #16209 from Security-Onion-Solutions/cogburn/memory-fixes

Batch Size and Retries

salt/soc/defaults.yaml

@@ -1547,6 +1547,8 @@ soc:
           maxGlobalMemoriesToInclude: 5
           maxUserMemoriesToReconcile: 20
           maxGlobalMemoriesToReconcile: 20
+          memoryExtractBatchSize: 5
+          maxMemoryRetries: 2
           memoryModel: gemma@SOAI
           embedModel: amazon.titan-embed-text-v2@SOAI
           reconcileModel: gemma@SOAI

salt/soc/soc_soc.yaml

@@ -908,6 +908,14 @@ soc:
             description: Milliseconds to wait between attempts to look up a tool request that has finished streaming but has not yet been saved to the database. The API retries up to toolUseTurnAttempts times, so this value times the attempt count is the maximum wait before auto-approval fails. Increase it if tool auto-approval fails intermittently.
             global: True
             advanced: True
+          memoryExtractBatchSize:
+            description: Max number of a session's messages that will be processed for memory extractions at a time. Larger values are more efficient but risk exceeding the memory extraction LLM's size limits, due to the variable and unpredictable size of each message.
+            global: True
+            advanced: True
+          maxMemoryRetries:
+            description: The number of times to retry extracting memories from a session if errors occur.
+            global: True
+            advanced: True
       client:
         assistant:
           enabled: