Structured Data Analysis (SDA) Plugin Level Toggle

This is helpful feedback – we're hearing this from a few customers who want more control over where SDA activates.

Just to provide some context on what SDA is doing: SDA is a sensor that helps the reasoning engine understand large volumes of data. Today, if data from a plugin is too large and doesn't fit in the context window, it gets truncated. SDA was created to help parse and consume larger amounts of data that don't naturally fit inside the context window.

So if you turn off SDA on a plugin, that plugin will likely start losing data (anything that doesn't fit gets truncated), which leads to inexplicable experiences where the agent just doesn't "see" parts of the data it needs.

That said, we don't want SDA to hurt the experience, and we'd love to better understand why there are certain plugins you don't want this to run on. Could you share specific examples?

• Which plugin(s) are problematic?
• What queries or use cases trigger the unwanted SDA behavior?
• What's the impact when SDA runs on these plugins? (performance, incorrect results, security concerns, etc.)

Having concrete examples will help @anayak (our SDA PM) understand the pattern and figure out what's going wrong.

Thanks for raising this!