We’re seeing what looks like a bug with variables when an Autonomous Node executes another workflow. This was working reliably until recently; no bot-side changes. Bug report: report_01KEAJ3VFHTHYJHVG5KB8HX7KR What happens: - An Autonomous Node sets a workflow variable W1 and a conversation variable C1 - Logs show the variables were set - The node then executes another workflow - In the new workflow, both W1 and C1 show up as empty/null This used to work fine, but it’s now failing and is blocking our bots. Has anyone else seen this recently, or know a workaround?

Yes, I can help 👍 This is likely due to a context scope change in recent Botpress updates. When an Autonomous Node executes another workflow, workflow variables aren’t persisted, and conversation variables may not carry over unless explicitly passed or re-set. That’s why they show as set in logs but appear null in the next workflow. Workaround: save values to conversation/user state before executing the workflow, or pass them as inputs. Quick question: are you triggering the second workflow using Execute Workflow, or another method?

I have tried saving them as conversation variables, and passing them as inputs. In both cases they are lost. AFIK I'm using Execute Workflow, I added a couple screenshots to share what I see. Not sure what other methods exist to call a second workflow from an autonomous node. https://cdn.discordapp.com/attachments/1458207052152312033/1458219595339923508/Screenshot_2026-01-06_at_11.02.46_PM.png?ex=695ed853&is=695d86d3&hm=2d74f67ae4b1dc4b358997f188d744ef03a8e22cc3e576cb806157c949c51501& https://cdn.discordapp.com/attachments/1458207052152312033/1458219595587522640/Screenshot_2026-01-06_at_11.03.15_PM.png?ex=695ed853&is=695d86d3&hm=164dbe5c0140c2ccd6ffc7191e47b3c2f20a04f7f5ea9c600d3422b6dded25c5&

Thanks for the details this helps. Given that even conversation variables and inputs are being lost, this points to a recent regression or execution-context bug in Botpress Autonomous Nodes when using Execute Workflow, not a configuration issue on your side. I’ve worked around similar cases by restructuring execution flow and forcing state persistence at specific lifecycle points, or by isolating execution into a shared parent workflow.

I'm just curious, for those who have reported issues with the autonomous node. Did you guys notice the AI is executing a different action from your prompt? I mean when you click on “Inspect” and check the code executed, it does not follow your prompt entirely. I wonder if this is a bug 🤔

what is your workspace ID?