Lemar has taken another step towards working as a complete operator: it can now plan sequences of actions, not just react to a single alert. And when it does not know how to solve a case, it tells you instead of making something up.
What changed
- Multi-step plans. When an insight needs more than one action, Lemar orders up to four steps using real catalog actions. For example, it can check the repair, notify the customer and create an internal task in the right order.
- No invented actions. The planner can only choose actions that exist on the platform. If none apply, it runs nothing.
- Fewer repeats. If a proposal is already waiting for you, Lemar will not re-plan the same case or create duplicate proposals.
- It learns with you. When a capability is missing, Lemar closes the alert, records the gap for review and notifies you in your language. The case is flagged for training or for a new action in the future.
Why it matters
Before, a case that did not fit a single action would stall or be escalated without context. Now Lemar tries to solve it step by step and, when it reaches the limit of what it knows, turns the failure into a clear training request. You stay in control: you still decide what is autonomous, assisted or observation-only.
How to use it
- Review pending proposals in the Autonomy tab. Anything that needs you is still approved one by one.
- When you see the "Lemar needs to learn something new" notice, open the Training tab and add the missing answer, FAQ or note.
- Adjust autonomy per action based on how much you trust each flow.
None of this publishes knowledge on its own or runs arbitrary code. Lemar proposes, you validate.