[BUG][BETA] semantic recall embedding are triggered for each step and history messages
VectorAgentMemory
I debugged doEmbed and got all the way to listResolvedInputProcessors.
memory.getInputProcessors returns new semantic memory class every time and so the internal caching is not really doing anything (this.embeddingCache - is new instance everytime and so if (cachedEmbedding) is never true in embedMessageContent)
this increases the latency of responses and cost unnecessary money for embedding that already were embedded before...
memory.getInputProcessors returns new semantic memory class every time and so the internal caching is not really doing anything (this.embeddingCache - is new instance everytime and so if (cachedEmbedding) is never true in embedMessageContent)
this increases the latency of responses and cost unnecessary money for embedding that already were embedded before...
