It would be useful to have the option to branch to a new context e.g. nb model inputcontext -o newcontext eval do the thing Along the same lines, being able to read __multiple__ inputcontexts and write to a new one would be useful. Agents could then summarise existing conversations, or draw related ones together.