I would personally like to have a command line tool as well (similar to e.g. 'ri' for Ruby). And it would be nice to be able to look up the same stuff from a Scheme REPL even when the REPL is not running in a text editor.

I think the command line option is already given for any VM/runtime based Scheme, the command line tool could then just be a small wrapper script invoking the whole thing. In addition it can be an interesting option for one of the executable-building Schemes, so we could e.g. have a Gambit/Bigloo based standalone executable, which does not require any installed Scheme and still allow access to the documentation. IIRC for both Gambit and Bigloo the REPLs are limited/stripped off for the executables, hence we might loose some features, but the convenience of a simple executable can very well be worth it. And as you write below, if we're smart with our Scheme sources (common, R6RS, R7RS, implementation specific), the executable build can rather be a deployment option than a different source tree.

If someone wants to use the API from an editor without running a Scheme REPL, there could also be an Emacs mode to talk to the API directly.

That's what my option 1 means. And as you note, this does not necessarily need to be an either/or, the options can also can also be mixed. I think we're now getting into a quite abstract/fuzzy area, where many thinks "could be done". I would like to map the options to the individual features, and then try to find a decision per each feature and architecture option. I already mentioned a list of features before - if you have more, please add, otherwise I'll start there.