Semantic recall cleaning
Telemetry
Quick question,. I want to do Semantic recall cleaning, semantic recall have tool-calls embedded in them, these can be very large, and overflow the context window, I have tried using the guardrails (https://mastra.ai/en/docs/agents/guardrails) I try to implement cleaning of this large tools calls. but i cannot do it with the processors.
Memory Processors run on messages retrieved from storage via memory.query(), which happens BEFORE semantic recall embeddings are added to system messages. The semantic recall content is embedded into the system message during prompt generation (after memory processors run), so my EmbeddedConversationCleaner will never see or clean these embedded conversations. Is my understanding wrong?
Question. How can i make this happen? is there a way to clean the system messages?
https://discord.com/channels/1309558646228779139/1309558648476930100/1431223615738613781
Memory Processors run on messages retrieved from storage via memory.query(), which happens BEFORE semantic recall embeddings are added to system messages. The semantic recall content is embedded into the system message during prompt generation (after memory processors run), so my EmbeddedConversationCleaner will never see or clean these embedded conversations. Is my understanding wrong?
Question. How can i make this happen? is there a way to clean the system messages?
https://discord.com/channels/1309558646228779139/1309558648476930100/1431223615738613781
Learn how to implement guardrails using input and output processors to secure and control AI interactions.
