Learning Hub + Reasoning Effort

Learning Hub

See exactly how your agent is performing and improve it in a few clicks. The new Learning Hub turns the feedback your agent collects from you and from its own self-checks into a clear list of recurring issues, then drafts fixes to the instructions behind them. You review every change before it goes live.

  • See overall accuracy and the feedback collected over time, at a glance

  • Recurring problems are grouped into plain-language issues, worst first

  • Fix an issue with one click Beam drafts the improvement for you

  • Review a clear before/after of every change and approve it before it goes live

Reasoning Effort for LLM Models

Choose how hard a model thinks on each step. You can now set a reasoning level on any model that supports it trading speed for depth where it matters, and staying fast everywhere else.

  • Set reasoning to Minimal, Low, Medium, or High on any reasoning-capable model

  • The control appears only for models that support reasoning no clutter elsewhere

  • Works on standard steps and decision points; each model keeps its own level

  • Existing agents are unchanged your current defaults stay as they were

Improvements

  • Fail a task automatically when its accuracy stays below your set threshold, even after retries

  • Give feedback on a whole task in plain words Beam routes it to the right step for you

  • See tokens used by each optimization in the Learning Hub, for clear cost visibility

  • Merge duplicate issues on the same tool to keep your list clean

Bug Fixes

  • Fixed the Outlook trigger throwing an error when a test email included CC recipients

  • Fixed the HubSpot “Update contact” action failing partway through a task

  • Fixed deleting a decision point leaving the following steps disconnected in the workflow

  • Fixed re-running a single step not updating the task view until a manual refresh

  • Fixed users invited as admins showing up as members instead

  • Fixed being unable to leave a workspace when its owner couldn't be identified

  • Fixed the Test Output button not returning results

  • Fixed evaluation criteria disappearing when you regenerated them

  • Fixed profile pictures not appearing in the side navigation