Your post does a good job describing what is stored in a readable format on your servers. One thing I didn't understand is where the data sent to users not yet on Keybase is stored. It sounds like my client is responsible for storing this unsent message, either locally or on Keybase's server encrypted for myself, and occasionally querying the DB to see if the target user exists and re-encrypting it. Is that right?

Finally, any thoughts on spam prevention for this protocol?

Encrypted messages waiting for others are stored on Keybase servers. But they're encrypted only for the sender. The important requirement of this protocol is the removal of extra human steps, especially the ones before composing a message. The only thing a person should have to do is (a) write a message and (b) maybe tell the recipient it's waiting for them on keybase.

It's the worst to ask someone for their number or email before you get to compose a message.

There are still computer steps, however, and Keybae hides them from you. If you post an encrypted message for foo@reddit, then you sign a statement to yourself that foo@reddit is an intended recipient. When foo@reddit proves herself by announcing a key, your own client verifies the announcement and rekeys the content for her, assuming her key proof matches your signed assertion. This is no work for you, although it does require that one of your devices comes online. If they try to get to the content and all of your devices are off, they'll see a message that the original sender needs to come online before it's available for the first time.

The story with phones will be even better, as they're easier to reach through push notifications.

As with most large data encryption, this is performed by rekeying a symmetric key. So you can leave a huge attachment for foo@reddit and you don't have to download/upload the data all over again just to make it available.

As for spam prevention...we're still discussing. We obviously don't have the ability to study message contents. At the very least, we will need to have easy/clear blocking features

I hope this is an internal nickname ;)

Thanks for expanding, and I wish you luck on the spam prevention. We often have conversations in the office about implementing a cost-based anti-spam system[1] for email, but I think a lack of universal usage kills that idea before it can begin. Keybase may be in a position to implement it before the protocol and user interfaces have ossified. Something to think about.

I find this stuff fascinating. Deciding not to interview with you guys because I didn't want to move remains one of the biggest "what if"s of my career so far :)

[1] https://en.wikipedia.org/wiki/Cost-based_anti-spam_systems

This immediately jumped out at me as a potential problem with the system. I think you had better get jumping on it quickly. Preventing abusive messages is much easier if mitigation is built in from the start.

What prevents hundreds of spam messages hitting a new user as soon as they sign up? An automated system could create a spam message for every github user account (just as a way to get likely signup identities) and keep a client running and they would be delivered to new users as soon as someone verifies themselves. A group of people could target one individual so that they are flooded with unwanted messages upon joining.

It'd be like arguing that by enabling PFS also creates a social contract that the receiver guarantees that their device isn't compromised. Or that webservers that use PFS in TLS don't log most/all of the details for the requests they receive. I don't believe either of these to be true.

You seem to have extrapolated some social properties from a purely technical property and assumed everyone must think about it in the same way as you. But in doing so, you've made your protocol weaker against realistic attacks.

Oh, and the likelihood of an attacker stealing a device with WhatsApp installed and only being able to extract the identity keys and not the cached messages seems absurdly improbable.

Personally, I wish you'd either tried to work with OWS/WhatsApp/whoever to integrate in what makes keybase.io great (providing identity verification - something that all these could use) and not gone about adding to the already crowded area of chat providers.

https://news.ycombinator.com/item?id=13605873

Nothing stops you from going full PFS.

In fact, you could even flip it around and let a shared notepad connected to the chat represent what you want to remember permanently, while the chat could remain ephemeral.

This makes for more accurate expectations and less risk of user error.

[1]: https://keybase.io/encrypt#oddevan

wcBMA+CNQSDgk7JsAQf/bOJdpfJ6Uzt2/dwlQrgYnOOEFVDoFnMYpLDU3R41U4EI vG5CuW3BtpUZyQI+W/sgVXJoaj/HHtc4H+Kj3lqDizdUBansIYBbjtMZaJ/hN/cm zNIa6NlUZqmGKOYt0CKJhINnTp0dQJeVlKJQHjOBwvaqpPw0jfYLqLwfkR5y55RO ++WwyHSLXHQYTd6vX/CYtVAYIEU8NiP3s/FraLtYSF1OWMDITV22vyw5NSg9BLiK SI2vL74hKT5W8AvMBWdMIbFvlC6ef5e8BXABPLVXUtxuHg9t/uCM5ZQ6gnegn54R K39EA4bOwWuCE5XJK1h2RfCB6k8FG7nl9SpjTU6GW8HBTAMv22o+4yM4xAEQAIVZ Szs2mKfbkMKoR1jr6lTvfjm3wpd95ykOMZtKNd1YPPjqbGpAVii/SPqlTMwsgaj3 uuGSZlIQMcjUB/QixaeJowddBQxZzJHS04Cg0t+44MlaVczMe0gE3dyEY8swmKbM OFBq3nPDPuEyFV1LckFUuQwnwfoexvAgzEbW6m6qWt7DR+xKOW6DwkBL2aztWtJc A2s+JBRFw58PY9aeUIku6L0v3+tHLMKAQoXnmwFVA277CpG2OH7sk6rLq85y5JrW fXJ4SuHyIW7mLjYKm9PX6+PhQLH9KzqYlWM+08p/IoXR7DHYSWHRvk7zSruJgHqi I8WJTDdRghfUcbeedQRH8fx7MZiBwLQeHg0kBEEn6ajGr8trbKzdgw82FqPwXIfL 09Df0pcfXhJimfQ8XRVeNPK4+DVHUCEWc4cavm8ZaNSA8r1vRTytNtI3h4eOtM9G FZh/eaxl7hX8KxNGuK0clY+oMshUiNRIVScEUtsgpH8actXSdP16Wv/+H+51Tg9G 8rdNyeRMMBqKn+fDAAQFKbSe7n4GyNcg+dbHFuVzaF1fNq2LbGPOtnKb+Xbfwg0O WMLCui3TXNre30Jv5iorSTNLu3xP9X0LEDWLsgBeqgYW/YFhtwBLjeFUV5lmB4YS UVCDxepIb+yGDHESm9rwb/1Na0kslNPh9/18ngZ40sHBAerxcxFVGav1JBqpModE usf64H3xiBzIyxHbCzPhiSKZjuP2WXNV5k153rCG8bbreTfUXxpO6nIBHT2ie6sA xo7tN66AmN2fouxMIFYERIbL7S5S8fBkO1tsaF+VSveBx2heTQ84k+P4TCVD7PQB IU+TPXg+Vrz15tiVSoSI+Ep+rV2pzKOFQXl0adOUisV8tMMFJD3YSwbUoTreM8tq 7CAyWhCxjnbgiedG1qF/J55WYdNyc9PorK7Kjx6wTRHLVQhF1W80OT8X5gvFbUoe PnzNTp2zi/4hKZfl/8G2tN0VZqhEynCT+iaDO0hG7VVz8DJfDz+9699KpoSVbiad pHiRORulU3/gYKLAYYpOo1fx6AP1utcmDfwEiJEROfZqkMI+CW5HTXGZavr3eImq JhZFsrtSvitXGwcGW6bYDoR56NaV9+A2PUlVt9PrlAonus06tuj2/o47jrdxmX1K aEp3uWVh0ZFSjc2YazmaN/zJyECP7b5Ig2wvhjDh39QqlYPocSZnNnZkWlUi6x3r OQ7ffkqepcFD+Zafm8vU95xs8ehFY+Jivdbbj4P3kff1LBv8NQ6KDKA4eiliDrnO Ml/B1TwipND2PVg8qEq2n64cYJpeGD/Hrcr8PlAvH+v8Xc9VJaerB8yeg9FqyJS7 7dAAkJsgWXBx1bfK6BUkJ/CefNjOYjtizwhVa1uQFOZjpg057WEo0m+dw7nCEG5M 3VexMu/Kx96dI5+hqs6SWSf6rC+Jj74J5cyvtoZZ2XdHl8GzrKt4mHf+XFpqTefG TjwNkYPnofKXfXe/R14wifThs2thVmM6tY5TJBtGdqOVVhJ2wbsbByQ2ivVYBCW3 hAA= =1dpX -----END PGP MESSAGE-----

https://saltpack.org/in-the-wild

There's one hurdle I need to work through to get going on chat. Thus far I've avoided uploading my private GPG key to my Keybase profile, or even copying it to other devices (call me paranoid). Unfortunately this apparently means I can't authorize any other devices (see error message: http://imgur.com/a/UOftN). I assumed device keys were meant to solve this problem, but maybe not. Is there a supported way to make a subkey (GPG or otherwise) of my primary private GPG keypair, so other devices can securely authenticate against my KB profile?

EDIT: I haven't yet started using device keys. Maybe they would work?

You'll see this policy in action when you install Keybase on a 2nd computer. It'll make you either (a) type something on your first computer, or (b) enter a paper key. This isn't just two-factor auth with server trust. The old key is signing a statement about the new key, and the new key is countersigning.

    Install Keybase on a different machine that has your PGP key

I've set up Keybase on my trusted machine with my GPG keypair, and now have a device key on that machine. When I go to Devices -> Add new... -> New Computer in the GUI I'm told to "Type in text code" (along with the note "In the Keybase app on your computer, go to Devices > Add a new device"). I find this confusing because I'm already there. I tried using the only paper key I have, the one corresponding to my first device key, but there's no response when I click Continue. This is the Linux client, by the way. I'm guessing this is a bug, but I'm not sure. Can you confirm this is the same process you went through to generate your second device key?

When I try to log in on the secondary computer, which doesn't have the GPG keypair or a device key, I'm brought to the same error shown in the screenshot.

  > Undocumented in the post: you can invent channels
  > for app-to-app communication from the JSON API.

Wondering if you can clarify a bit—what parameters cause a chat to/not to show up in the GUI? Different topics? Different channel names? I have a use for this immediately and am excited to give it a shot!

For example:

  keybase chat send friend1,friend2\
   --topic-type=dev\
   --topic-name=KEYBASE-SYSOPS\
   "server-restart [reason:hot and bothered]"

----

If you're using the JSON api - which makes more sense if you're programming it - take a look at `keybase chat help api` to see some examples of how to structure the JSON going in. Anywhere you see a `channel` object, you can add `topic_type` and `topic_name` to them. Both should be strings.

You said something about violating expectations when you sync PFS'd chat contents, but I don't see why that's relevant unless people were promised otherwise.

There's nothing about the encryption algorithm itself that dictates how data going over it should be handled at the endpoints.

Just let users have a regular mode and an off-the-record mode where nothing is kept. Both PFS protected.

Additionally I am curious why on Reddit people have to post a wall of text, while on Twitter you only have to post a small tweet? Is there some reason people can't just paste the tweet version onto Reddit too?

Secondly, I think this is incredibly important moving forward with the proliferation of messaging channels. I want to be able to use one program for all of my different chat sessions. Whether that be adium or pidgin or mIRC or even weechat/bitlebee in a docker container. It would be great if there would be common plugins for common chat clients provided off the shelf. :D

As it is right now, I imagine it would be very easy to write a library that acts as a bridge to other interfaces, if that's of interest.

There are calls to see into your "inbox" which is the set of conversations, read messages, even peek at unread messages. And of course send messages, too.

• a Matrix gateway server

• a libpurple plugin

Nice. I'll be looking at what source of tools that use this style of communication on github.