Your AI Steps Now Have Tools

Your AI steps can now use tools — not just generate text. Give one a goal, assign the tools you want it to work with, and let it figure out the rest. It decides what to call, evaluates what comes back, and tries a different approach if it needs to.

Feature screenshot

  • Use it for tasks that require multiple steps or can't be resolved with a single response

  • The step evaluates its own output and tries a different approach if the result isn't good enough

  • Outputs connect to the rest of your flow like any other step — conditions, loops, and beyond

Improvements

  • Custom integrations now reject duplicate names, so you can't accidentally create two that clash

  • Steps fall back to your default model automatically when a previously selected model is removed

  • Deleting an agent from copilot now takes you straight back to the home page

  • Clearer error messages in chat when something goes wrong

  • Beam AI now adapts cleanly to smaller screen sizes

Bug Fixes

  • Fixed .eml email attachments producing garbled text when converted for agent input

  • Fixed the Trigger Agent Task tool not generating the entry step's output values

  • Fixed connections being removed when you delete a node

  • Fixed the Outlook CC test trigger throwing an error

  • Fixed consent-required tasks being falsely marked as failed after 30 minutes

  • Fixed evaluation criteria disappearing when you regenerate them