Integration Auto Fix + Agent Monitoring

Integration Auto Fix

When an integration step fails, Beam now diagnoses it and fixes it for you. Click "Fix with AI" on the failed step and copilot opens with the error already explained in plain language, a suggested repair, and the reasoning behind it. Nothing changes until you approve.

  • "Fix with AI" button on any failed integration step in the task view

  • Copilot opens with the failure already diagnosed in plain language, so there are no logs to read

  • One suggested fix at a time, with the reasoning and how confident Beam is in it

  • Approve in chat and Beam applies the fix and reruns the step. Decline and nothing changes

  • Successful fixes are remembered per agent, so the same failure is resolved instantly next time

Agent Monitoring

Every agent now has a Monitoring tab where you set rules for the moments your team needs to know about: a task ends in a status you choose, an evaluation score falls outside the range you set, or failures pass a number you set. Each rule sends to Slack, Microsoft Teams, or email, either the moment it happens or as a scheduled report.

  • Set rules from the Monitoring tab in any agent's configuration, and pause or edit them from the list

  • Watch for a status change, an evaluation score outside your range, or a failure count you set

  • Send to Slack, Microsoft Teams, or email, in real time or as a daily, weekly, or monthly report

  • Combine conditions in one rule, so an alert fires only when a task fails and scores low

  • Smart monitor, on Pro: an AI model watches the agent, tracks escalations to resolution, and feeds Agent learning

Improvements

  • Learning Hub can now improve integration steps, not only prompts, with a rewrite you approve first

  • New Gmail and Outlook actions fetch only the unread emails received since a time you choose

  • The Outlook Draft Reply action now lets you override the recipient, CC, and BCC

  • Code Executor steps show a distinct alias in the link picker, so you no longer wire the wrong one

  • The Give feedback button has a new design, explains itself when disabled, and is hidden on test tasks

Bug Fixes

  • Fixed the Google Ads "create report" action failing after Google retired the API version it used

  • Fixed an agent trigger step not creating its output values

  • Fixed the Learning Hub review screen showing an empty "before" prompt

  • Fixed being unable to remove an input variable from a Code Executor step

  • Fixed wrong status counts on the task list after applying a date filter