In a best-case scenario, I can do the following things from within IDEA
- connect to a running FE
- talk to the FE, inject cells, evaluate stuff
- talk to the running kernel that the FE is using (so I can load packages automatically or get evaluated results back to IDEA)