M

Mastra•6mo ago•

3 replies

Felix G

Why does `memory_messages` always recreate the `memory_messages_vector_idx`?

We use Mastra with memory and it correctly creates the

memory_messages

memory_messages

table. However, there is no way to specify which index to use and always defaults to ivfflat with cosine distance, even though the vector code itself can take these options.

The main problem then stems from the fact that it calls the

setupIndex

setupIndex

regularly and tries to rebuild the index with different size. Unfortunately, we have so many messages that this times out and the chat regularly breaks down because of this issue. It would be great if we could configure which index the memory should use, so we could use

hnsw

hnsw

and also inner product because OpenAI embeddings are normalized and performance would be better.

https://github.com/mastra-ai/mastra/blob/03a908b6ef003c00ad4274eba9d25455c80c43eb/stores/pg/src/vector/index.ts#L558

GitHub

mastra/stores/pg/src/vector/index.ts at 03a908b6ef003c00ad4274eba9d...

The TypeScript AI agent framework. Assistants, RAG, observability. Supports any LLM: GPT-4, Claude, Gemini, Llama. - mastra-ai/mastra