Brim May 2026 Release: Intelligent Generation, Enhanced Validation, and A New Assistant
June 2, 2026
Every abstraction project depends on getting the right data from the right records with the confidence that the results are accurate.
This month's release is about giving you more control at every stage of that process. The May 2026 release introduces New Generation Capabilities that let you target exactly what gets run. Validation V2 makes it easier to build pipelines you can actually trust. Brim Assistant Beta is a conversational guide built right into the product to help you get answers fast.
New Generation Capabilities: More Control, Clearer Flow
Complicated abstraction projects demand fine-tuned generation control. This release gives you more options to describe exactly what labels you want to generate.
With this update, you can filter generation by document upload, so instead of running across your entire record set, you can target exactly the data you care about. You can also generate only new values, skipping records that already have labels and running only on new patient-variable combinations, a significant time and token saver on large projects.
The generation setup itself has also been reorganized so each step is clearer and easier to follow, whether you're launching a first-time project or running your tenth re-abstraction.
Finally, we've added an estimation of the Maximum Token Usage while setting up a generation, so you can understand the upper bound of cost before spending the tokens.
Learn more in our documentation.
Validation V2: Build More Reliable Pipelines
Validation has been rebuilt from the ground up to give your team more flexibility and more control.
Keeping your optimization set and your holdout evaluation set distinct matters a lot in practice. Without that separation, it's easy to overfit your variables to the examples you've already seen, and wind up surprised when agreement scores drop on new records. You can now separate your Training and Holdout data, keeping the data used to optimize variables distinct from the data used to measure their performance. This produces more honest accuracy metrics and helps you build pipelines that hold up in production.
You can also combine uploaded data with manually labelled records when understanding agreement for Training Data. Whether your golden data comes from a CSV import, hands-on labelling in Brim, or both, Validation V2 works with it all.
The result is a validation workflow that's even better prepared for complex projects and workflows.
Learn more in our documentation.
Brim Assistant Beta: Ask Anything, Anytime
Sometimes you need a quick answer without digging through data or documentation. That's what Brim Assistant is for.
Look for the chat bubble in the bottom right corner of Brim. Click it and ask whatever's on your mind: how a feature works, details of your project setup, or questions about your abstracted data. Brim Assistant draws on knowledge of the product to give you a direct, relevant answer right in context.
Brim Assistant is in Beta, and we want to hear what you think. As you try it, let us know where it helps and where it falls short. Reach out at support@brimanalytics.com or reply to this email. Every piece of feedback shapes where we take it next.
Learn more in our documentation.
How to Get the May Release
This release is available as:
Version 2026.05.29
You can check your current version in Settings → About.
As always, thank you for building with Brim.
We're excited to keep making abstraction faster, clearer, and more trustworthy, and grateful to support the important work your teams are doing every day.