At Singly, we raised about $9mm of VC with 100% of our code open source. We saw that as absolutely essential in creating an ecosystem and it was a huge benefit in reducing customer concerns around lock in.
I like what you guys are doing and would be happy to chat if it's helpful!
We are also struggling with the word Context. The thought was "a Connector is a dumb pipe and it needs a Context in which to run" (e.g. the auth credentials, etc.). Not perfect, and possibly something that will change in the not too distant future, but it works for now.
The whole project is still very young, but we are incredibly excited about all the cool apps that will be created on top of it.
It's not really a separate pipe, its's simply the bits of information that tell the dumb pipe where/how to run. For example, the Google Docs connector knows generically how to connect to a hypothetical person's Gdocs account, but the auth credentials provide the information that is required to connect to a real person's account.
A "smart pipe" is actually a lot more like the Collections. They bring data in from sets of similar Connectors (dump pipe) and collect, de-dupe, and normalize it.
I think the statistical importance is by far the most interesting feature. Our inboxes are really priority queues, but we are currently spending a great deal of time and energy doing the prioritization ourselves. The idea of seeing my inbox sorted by importance is game changing.
Hilary, what are the aspects of the email that influence importance? Is it mostly the text content of the email, or are there things like sender, number of emails in the chain, number of recipients, etc?
I agree. For example, why do we look at email ranked chronologically? I'd much rather rank it by relevance or importance (or my likelihood to actually answer it).
The feature set includes sender, recipients, whether I'm the only recipient, time of day, words in the subject, words in the body, and whether I've exchanged messages with the sender recently.
I wouldn't be surprised if GMail gets a feature like this at some point. Google Reader already has an option to rank RSS feeds by what it thinks you're most likely to read, so similar technology already exists at that company. The main (massive) issue would be teaching the filter.
Do the scripts have the ability to change the weight of features on the fly? For example, sender becomes the most important feature when the sender is, for example, my boss or particular end users. Otherwise, other features may be more dominant.
Greplin does sounds interesting. I actually applied with essentially the same idea - cloud data indexing and search. Check it out and let me know if you want a beta invite:
Yeah, I actually pitched the exact same thing to PG a year ago (in interviews) and YC thought no one needed search beyond e-mail. Then again I hear Greplin was accepted to YC with a completely different pitch :)
Frankly, it seems hard to imagine NOT needing search beyond email. I really only user a handful of web services (I'm sure that number is only increasing) and I can barely remember whether I sent something in an email, as a facebook message, saved it to delicious, etc.
I started creating Nutmeg Search just so that I could find my own stuff, but the idea seemed useful so I'm trying to make it available to the world.
I like what you guys are doing and would be happy to chat if it's helpful!