I reviewed HKUDS/nanobot v0.3.0 against its repository, release notes, and docs. No published prompt caching benchmark exists in the sources; an open issue discusses prompt-prefix preservation. Here is what the evidence does and does not support for self-hosted AI automation.