In addition to the current functions exposed in the Anthropic implementation, we should allow users to specify subsets of functions to be exposed to the LLM from our API. We may want to group this around entities like repos, issues... Ideally we also let the user pick their own functions to export, to help minimize context usage.