I upgraded my Langgraph-based multiagent from version 0.6.7 to 1.0.2.
This multiagent works like a chatbot, and I can easily ask the same question multiple times.
Comparing the agent’s resolution times for the same question, I noticed that the new version is much slower.
I’ve performed all the necessary migrations.
Is there any step to check besides the migration already documented for the create_agent part?
I’m attaching the comparison extracted from Langsmith.
It looks like most of your new latency is coming from the search biome tool - this doesn’t seem like it’s architecture related, and moreso something to do with your vector store retriever. Did the indexing of the vector store update at all, or was the result cached in the faster run?
Here I am again, I completely refactored the bot and its tools, switching completely to the new version.
Regarding the RAG part, I can confirm that even with 10 tests in parallel, the highest latency remains on version 1.x.
I ran a second multiple test with a very simple tool that retrieves a user’s timezone. Here too, the difference is very large.
dataexpert tool: 2.62 second with 1.x, 1,47 second with older (I can’t attach the 2 images directly here. I’ll post them directly in the next messages).
