Handle model truncating the returned answer from the CLI

- In case of tools that support multi VMs ID, parse the XML to return the minimum number of info needed to avoid truncating the message. If it needs more info, the model can call the tool again but with single ID
- Avoid vector db and adding other dependencies, work on the tool and on filtering the XML returned by the CLI