I’m excited to share that Healing Agent is now generally available, offering:
Intelligent self-healing to ensure your UI automations continue running uninterrupted, even when unexpected application changes occur;
Actionable recommendations for fixing automation issues—without the need for additional troubleshooting or debugging.
Demo:
By default, Healing Agent is enabled for all new processes to provide you with repair recommendations. You can manage these settings and the ‘self-healing’ capability at the process and job level in Orchestrator.
Check out our new documentation page for more details and licensing information: Agents - May 2025
On behalf of our engineering and product team, I wanted to say a big thank you for all your continuous feedback throughout this journey. From the beginning, we’ve felt so much support and excitement around Healing Agent—and we truly appreciate it. We hope you enjoy using Healing Agent and, as always, feel free to ping me here if you have any questions or feedback.
According to the documentation, add-on licenses are required here (except for Enterprise / Advanced platform SKU). Which is unexpected because the feature is enabled by default for new process deployments. Or are these only required for self-healing?
The documentation explaining the data collection is not meaningful due to the broken link: Agents - Data collection
This is great, one of the best updates to UiPath ever! I am wondering, how does the Healing Agent work when we are using libraries? For example, let’s say I have a REFramework bot running, and it uses code from a UiPath library of mine. If one of the selectors in that library fails, can the Healing Agent overcome that? Does it “temporarily” or “soft” update that selector?
Or does it actually go into the library itself and update the selector in there?
HA is an insurance for the robots. We recommend customers fix the problems immediately they become aware of them, using the HA fix recommendations, and not let it run on a ‘flat-tire’.
Licensing summary:
‘No Platform SKU’ or ‘Standard Platform SKU’ customers - need to acquire the HA Add on - (10k per year, includes 5k or heals)
‘Advanced Platform SKU’ customers have the HA add-on included (5k heals per year)
Consumables (15 Agentic Units / heal) can be used as fallback once exhaust the included heals
All the details here.
HA recommendations option (and not self-heal) it’s enabled by default for each new deployed process that meets the prerequisites. If the licensing conditions are not met, the process runs normally, without HA intervention.
Regarding Data collection page, we’re fixing that as we speak. Thanks for reporting the issue.
The HA solve the problem, at execution time. No matter where the selector/target configuration is stored (within the project on in a dependency library), if the activity it’s about to fail, it will fix the selector just in time and use the info to
offer a fix recommendations
self-heal the activity
HA does not permanently fix the code and let’s the developer review the proposed code updates.
I’m sorry, but let me clarify something:
The HA recommendations, which involves analyzing the UI by an AI-agent like functionality outside of the server that the automation is running on (I’m assuming it’s outside), and for which Data collection documentation is not available (checked now - still not working), is enabled by default for new process deployments?
If yes:
A. You’re playing with fire from GDPR and general compliance perspective.
B. Have this not been a lesson learned before with the “closest match” selector error messages that were exfiltrating confidential data into the Orchestrator logs?
We all know that whatever we send to Automation Cloud needs to be prescreened, but there was only 1 instance in all the years that I could remember where there was an “unintended leak” of data from Robots into the Orchestrator side - and that was the closest selector error messages (which were capturing whatever the algorithms deemed “close”, which in certain contexts did include things that could not land in the Orchestrator). From our context, when that thing occurred, we were still steering away from the Automation Cloud, and caught it very early, so nothing really happened.
But that would be very far from a short encounter if we were on Automation Cloud back then.
So I’d very kindly ask you to please not enable by default things that make data leave robot servers. We do not need more rounds with compliance teams than we already have.
And yes - our OPS did catch that it’s enabled by default when doing a deploy today (that’s why I’m posting here), and nothing actually happened, but it was a bit too close for comfort.
As a sidenote - your documentation on how to disallow Healing Agent on organization level is either confusing, or I’m just dumb with Automation Ops. Probably the latter, but for silly old folks like me (or people who haven’t really used Automation Ops before [also partially me]) it would be nice to have a very clear “put this here” step by step, especially since it’s on by default and we have to block it. A global “Healing Agent On/Off” would be nice, but I guess instead we need to scramble around in different places and hope that it’s enough and we didn’t miss something.
Enabling by default HA Recommendations only (not self-heal) for new deployed processes is meant to help you, our customers, to fix faster UI automation issues and save time.
We offer multiple ways of opting out of Healing Agent specific features or even, disable it entirely in your tenants:
if you want to use Healing Agent without allowing it to make any Gen AI calls, we offer AITL integration - data will not leave the robot machine
if you want to govern the Healing Agent at the org level, use Automation Ops Robot runtime governance
**Update the Robot Policy to 24.10.0 template
** Configure Robot policy/the HA tab, and deploy it on your robots.
[Allow Fix Recommendations] If disabled, this flag is completely shutting off HA at runtime (does not influence the OR/Process level configuration).
[Allow Self-Healing] allows self-heal, only if [Allow Fix Recommendations] is enabled
[Allow Screenshot Saving] allows screen saving for debugging purposes, only if [Allow Fix Recommendations] is enabled
forbids/allow HA at the application level, only of [Allow Fix Recommendations] is enabled
We even offer an option at the UIAutomation activity level:
Thank you. This is the step-by-step for dumb-dumbs like me I was talking about. Would be a great addition to the docs pages.
Also good to see the Data collection page being up.
I’m having issues getting the Healing Agent to run. I’ve been using it since it was in public preview with no issues. Now when I try to enable it in a process, it says: ‘Update Needed’: Upgrade or allocate more units
Upgrade the Healing Agent add-on to use this feature, or allocate more units. Please consult our documentation for more details.
