Run deep research queries in parallel

[chetanunadkat]

I was wondering how to make deep research faster. I have connected it with single mysql table having 7 -8 million rows. deep research on this take around 20 mins on local macbook pro m3. is there a way we can run these queries in parallel ? or some other improvements ?

[rishsriv]

Thanks for the issue. We'll attempt to add this functionality.

Currently, all tool calls happen sequentially so that they are data-aware. I.e., subsequent questions asked can be much more intelligent – as they are based on data from prior questions asked.

Having said that, we don't have to have every call be sequential. Instead of a "one-node-at-a-time" graph, we can try and make this a "multiple-nodes-that-interact-at-once" graph.

We'll experiment with ways to make this happen, though no guarantees as this might cause regressions.

[dyrtyData]

@chetanunadkat

• Though I wouldn't recommend calling the tools in parallel within each query for the reasons @rishsriv outlined above since this is a self reflective/adaptive RAG chain where each response is included in the context for further tool calls.
• However, you CAN run separate queries in parallel by just submitting separate queries one after another (I do it through their API http://localhost:1235/docs#/) although you may have to space out each subsequent request in order to avoid hitting Anthropic API rate limits depending on which Tier you are on.