Batch Size and Retries

2 new config fields. Batch size is used to limit how many message turns we put in the transcript when we ask the memory agent to extract facts. The retries helps limit how many times we ask the memory agent to process a problematic session.
This commit is contained in:
Corey Ogburn committed 2026-09-04 12:18:29 -06:00
1 parent 64771e9e1d
commit 30574fdbb9
2 files changed
+10

No files matched your search

+2
View File
@@ -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
+8
View File
@@ -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: How many messages to include in the transcript that the Memory agent will extract from.
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: