A TL;DR after analyzing Clubhouse app | Hacker News Reader

A TL;DR after analyzing Clubhouse app

twitter.com

Serious question: Why is an app like Clubhouse using external services such as Agora and PubNub and relinquishing control over their core app functionality (audio rooms)? Wouldn't that be something they would want to engineer in-house?

Clubhouse atm is probably an MVP, and they most likely don't have the infrastructure talent to build/manage an audio and streaming backend.

Seems like they shouldn't have a billion dollar valuation if that's the case.

Why not? If they have product-market fit, they can build the infrastructure they need.

I don't think they have product-market fit. It's not even public, you need an invite to use it.

Regardless, you wouldn't do this as it introduces massive risk to the business and puts you 100% at the mercy of your technology providers (who I hope are currently raking Clubhouse over hot coals for all the cash they have).

> It's not even public, you need an invite to use it.

It’s not clear what this has to do with anything, or whether it’s even realistic to say it’s not ‘public’.

All existing users have invites to give out. Part of it is a strategy to manage scaling, but the other part is about building a coherent network.

It’s not obvious that they ever need to change this.

> Regardless, you wouldn't do this as it introduces massive risk to the business and puts you 100% at the mercy of your technology providers (who I hope are currently raking Clubhouse over hot coals for all the cash they have).

I assume you don’t have a startup valued in the billions of dollars, so it seems like both they, and then investors know something about this that you don’t, but I could be wrong about this assumption! Feel free to correct me.

As for being at the mercy of their providers - normally in business these problems are dealt with by using a contract that sets out what the costs will be and what the service will be for some time period. It’s also controlled by there bing more than one supplier, and supplier reputation.

In short, business just doesn’t work the way you are implying it does.

If their suppliers are able to ‘rake them over the coals’, that would be a huge failure of their lawyers, and it would be hugely damaging to their suppliers, who would not be trusted by new clients.

Exposure to suppliers is also a normal risk which is checked during due diligence. That means their investors attorneys would also have failed.

They have a small, private audience that they've extrapolated into a mass market audience. That does not justify a one billion dollar valuation. They do not have the usage to back that up.

VCs make tons of awful bets, I would not say they have any particular insight into what will be big and what will not. They fail more than they succeed.

Regarding vendors, you're definitely incorrect. Clubhouse has exposed themselves to great risk by relying on another platform. Just a reminder:

https://newatlas.com/skype-joltid-litigation-shut-down/12390...

> They have a small, private audience that they've extrapolated into a mass market audience. That does not justify a one billion dollar valuation. They do not have the usage to back that up.

How do you know? Have you seen their data?

Again you say the audience is ‘private’ but that isn’t true.

> Regarding vendors, you're definitely incorrect.

I’ve explained my position. You have not explained how it’s wrong. I have to assume you can’t.

> Clubhouse has exposed themselves to great risk by relying on another platform.

You don’t know what risk they have exposed themselves to unless you have seen the contracts and know what the switching costs will be. You also don’t know anything about where in their timeline their own infrastructure development is.

> Just a reminder: https://newatlas.com/skype-joltid-litigation-shut-down/12390...

That link proves my point, and disproves yours. eBay didn’t do appropriate due diligence, and didn’t have an appropriate contract. That is a risk Clubhouse can easily avoid, and the eBay/Skype case is now a well known piece of business lore taught in business schools.

I assume you don’t know anything about what contracts Clubhouse has with their suppliers. Again, correct me if I’m wrong about that.

Reminder that valuations are a measure of what investors think you might exit for, not an assessment of whether your technology is any good.

Even in "tech" startups, the marketing, growth strategy, investor marketing, and userbase are incredibly important. If all those are amazing enough you can IPO before you even launch a product.

> If all those are amazing enough you can IPO before you even launch a product.

No, you actually cannot.

For a one billion dollar valuation, it should be expected that the company owns the technology stack used to power the product. If I were Clubhouse's vendor I would be absolutely holding them over a barrel right now. I would have jacked up my costs 10X the second they closed the round.

That would surely cause all of your other customers to freak out that if they became successful you'd do the same thing to them.

Your reputation for fair prices would be ruined and all of your customers would now be building their own versions of your platform.

Effectively your startup would be over, since nobody would ever trust you again. And, presumably your investors wouldn't be very pleased with you, so I guess you'd be looking for a new line of work, too...

(That's not say that vendors won't increase their prices, but they will have to do so slowly. Meanwhile the founders of Clubhouse will be aware of this and no doubt building the replacement platform that has everything they will need...)

I don't think anyone would fault a vendor for charging a startup that built a billion dollar valuation off of their tech stack more money. In fact, I'd question the business acumen of a vendor who didn't do that.

Of course, most people know better than to use someone else's technology to build their business.

The demand and growth for the social network is what the valuation is based off. Marketing/Growth is really expensive and more expensive than actual infrastructure spend for most companies. Users of the social network don't care what the infrastructure is based off of.

Because it's the social network that's generating hockey-stick growth, not whatever boring technology enables it. The alternate universe Clubhouse whose developers decided to build their own audio stream multiplexing technology first is still working on it while the actual Clubhouse is rapidly grabbing market share and securing their position.

It might make sense for them to acquire their infrastructure providers or clone their technology in-house, but only later when they have the money to actually do that.

> would want to engineer in-house?

It's quite expensive to do so so why not buy a reasonable working solution until you become big enough?

Clubhouse is in the end a life style app which from the get to go uses the snake oil of pseudo elitism and exclusiveness to create a flair of it being a high quality serious product independent of weather or not it actually is one.

So they do a clever slow grow strategy:

- start with invites to grow slowly and with this only have slowly increasing costs of operation

- only start with one platform to keep initial cost down

- buy in/rent all the tech know-how to reduce initial costs and bring initial products out faster even if it has some quality or security issues

- (maybe) hype highly to grow investments fast

- later expand to other platforms

- later replace bad dependencies with custom solutions

Engineers engineer, product creators create products. Therefore, product creators are running Clubhouse.

Interestingly enough, I believe you inadvertently highlighted a major reason why many projects never manage to see the light of day.

They probably realized it's not worth the engineering effort to bootstrap an in-house audio/voice backend until they know if their idea is viable.

Also: For all we know they've already hired engineers who are diligently working on this right now.

I did a deep dive last week on this, more details here http://atyesh.com/how-clubhouse-app-is-built.html I dont see them using Agora infrastructure when you join from US

This is great analysis!

thanks

Do you see them transmitting audio data? Because if you don't, you're probably missing something. (In your writeup it looks like you're only looking at HTTP requests.)

I didnt look at websocket data, i looked at all the https traffic, there is just no mention of Agora at all there. On the API call where they return the token and keys for the room, there is this flag agora_native_mute that is set to True. Other than this its all pubnub

it is possible that if you join from China/Taiwan, then they might use Agora to host those rooms.

I looked at PubNub's site and it doesn't appear that they do the type of audio tech that CH would need, or at least if they do, they don't advertise it very clearly.

Do you know if PubNub has the tech capability to support ~7k people in one audio conference?

Looking here https://www.pubnub.com/blog/introducing-voice-and-video-to-e... Maybe i can do a packet dump to confirm this.

Which would mean they're at the base using Vonage's services, which I think Vonage bought Tokbox. I still believe that they're probably using Agora thru UDP as someone else mentioned, but I don't know enough networking to really know what that means.

Uh... I think you completely skipped the UDP portion.. Agora (hence CH audio) uses UDP and obviously you wouldn't see the traffic on HTTPS or Websocket.

In so far as I can tell Pubnub does not do A/V streaming. Instead they partner with others such as Vonage and Agora: https://www.agora.io/en/blog/agora-and-pubnub-collaborate-to.... They're quite explicit here: https://www.pubnub.com/developers/demos/webrtc/

I don't get why startups have such horrible security practices so often. It's much harder when you have to improve a legacy application, but these guys started from scratch! Does it really take that much more time to at least follow basic security practices? Even if one wants "to move and break things fast" I don't see how it would take that much more effort...

Not enough appsec people relative to engineers?

Security <---------------------------X-> Convenience

The 'X' is where I imagine most startups are focusing for usability and time to market so they survive.

Moving the 'X' farther left delays time to market and increases overhead for the company, tech and users.

We hear the version of every day as "why does Signal still doesn't have feature X? whatsapp, telegram and most other messengar have it since ages."

And that's why telegram has at least 10x and whatsapp 100x more users.

My guess is that there are plenty of startups that put a lot of time and effort into security, you just have never heard of them because they never shipped a product

If an app like Clubhouse wanted to hide communication with Chinese servers would it not be trivial for them to proxy all of the telemetry calls through a US-based AWS or Azure server?

Clubhouse will be another snapchat when FB/Twitter clones it.

So Clubhouse will have a market cap of $96 billion? Seems worth it for them then.

Snapshat is far more impressive wrt tech and scale. Clubhouse will never get that far honestly - simply too easy to clone even without Facebook and Twitter.

Snapchat was the same though which is why I jokingly posed the counter point.

Snapchat was just an app to send photos (and later videos) back and forth. Facebook copied it with their Poke app that failed. Maybe the answer is integration with existing apps instead of having a dedicated app? Instagram stories was relatively effective

Way ahead of you https://help.twitter.com/en/using-twitter/spaces

repeat of https://en.wikipedia.org/wiki/Vine_(service)

The ClubHouse price isn't the technology (as it isn't nothing more than audio in some setup). It's in the userbase and attraction: influent people getting onboard, demand based on invitation only, exclusivity, iOS-only...

If they don't lose their hands, they'll keep this as a desirable product, and the value will reflect so.

1. ToS violation to record and publish shows without 'written' consent of all the people in the room (which is effectively not enforceable)

2. iOS only

3. Invite only

4. Not ADA compliant (leaves out deaf folks)

5. Buggy

6. Centralized

7. You can't easily delete your account

8. Pushes hard to ingest your contacts, expanding their social graph

I personally tried it and deleted it. No thanks.

>Pushes hard to ingest your contacts, expanding their social graph

This is the #1 thing that new apps seem to do. Is it just a growth hack, or is there something nefarious about it?

For example, I find it mindblowingly irritating that Telegram, of all things, blasts a notification announcing you've joined, the instant you join, to every other Telegram user who has your number saved in their phone. You can't opt out of this. And that's without even allowing access to your own contacts!

> Is it just a growth hack, or is there something nefarious about it?

Yes. It's a nefarious growth hack.

Are we really gonna complain that audio-only app leaves out deaf folks? Come on!

I think we have to consider whether the audio is the defining feature. Clearly with Spotify, it is, and an expectation that it works for deaf users would be ridiculous.

I think with Clubhouse it's not. The defining feature is the network and social interaction, the audio is the format in the same way that on Facebook it's text. We expect Facebook to support screen readers for their text posts, so I think it's reasonable to expect that Clubhouse do something for deaf users.

Whether they should be required to do this at an early stage or not is a different question in my opinion, but I think it's fair to consider eventual support the Right Thing To Do.

providing real time translations (and transcripts) is something skype has solved years ago. instead of reinventing conference-calls it wouldn't have hurt to at least provide one feature that justifies the data harvesting to the user.

I feel that your comment is insensitive. For context, this was posted in the Badger DAO discord by a deaf user:

"I would like to recommend that we stop using Clubhouse. It has already been identified as one of the most inaccessibility platforms for individuals with disabilities. I am Deaf and I cannot access it - if it is going to continue to be used, it should have a transcript and coin giveaways should not happen unless it is on a platform that allows accessibility for all individuals with disabilities. It is a matter of time before Clubhouse catches up with accessibility but right now it is pretty much a fast moving platform because they aren't following accessibility requirements. This usually happens to new emerging companies and they move fast because they avoid accessibility. I hope @Spadaboom | BadgerDAO and @DeFiFry will give this some serious consideration and move to another platform that allows accessibility.- @Gabrielhaines is pretty good at this and could give some ideas for clubhouse alternatives."

Clubhouse will soon have major competition with Twitter Spaces too. It already has closed captioning, tweet based Q&A highlighting, and is integrated with Fleets. Looks to be the most meaningful update in a very long time for them.

I think all of these products will quickly find that there is no mass-market demand for an audio only social network. This is a tool for self indulgence among "thought leaders" not something people actually want.

Even podcasts, which I'd argue are more accessible than live rambling, are not a mass market product after more than a decade of trying.

Radio was one of the most mass market products, wasn't it? Not trying to be mean with that but to say that audio-only can be big as well. Maybe talk radio declined precipitously with video on TV yet I think there's still quite a huge market for listening to people talk and/or joining the conversation.

I think that's the theory behind why people keep chasing this market, but radio's success was likely due to lack of competition. People have been banging on this door since Odeo.

What I've seen is that the barrier to join CH as a speaker is lower than being on video or maybe even writing in text. Audio only is much easier to produce, especially when one doesn't have to edit it, than video I'd say by orders of magnitude. Also, for people who don't like to read and write so much, speaking can be much easier.

Will audio-only eat up video and text platforms? No. Video can provide much more rich info and text can be much quicker to read and in loud places. But I do think there is a spot for live audio-only content, just don't know how big it is and how that changes when the pandemic subsides.

I totally agree that it's super easy to produce, that's one part of the problem with it. No barrier to entry, means a lot of trash content.

I think the real challenge is on the listening end. I cannot for the life of me imagine why anyone would sit around listening to people talk. If there's important information I need, I'll read it or watch a Youtube video. Audio is the least useful medium to absorb info. I'm sure some people feel otherwise, but we haven't seen a success in this space yet, and not for lack of investment/trying.

Fair point on being too easy to produce content, yes, can lead to lots of noise.

I think many people listen to people talk, especially at conferences. Imagine a panel discussion: if one is not in the first few rows, the visual info is quite little (can't see facial expressions much) and so it's mostly audio.

I guess for me, I like listening to live conversation, where it has the unpredictability of where it will go. I like to watch live interviews on TV and yet believe that many of them are stale, scripted, and too short, vs CH which yes can be stale but rarely feel scripted or too short. What I like about interviews or live conversations is that things may arise that one wouldn't think to produce without the external stimulation.

I wonder if it hasn't succeeded before because it didn't have the initial crowd. CH is basically just group conversations and if I go to a party where I think the people are not interesting (or more so, famous), I may not be so excited to talk/listen. But if I go to a private party, almost like going to a TED conference, I'm quite interested to hear those people talk to each other and myself try to contribute if I can. That's what it is, based on the audience it initially curated, it kinda felt like being at a TED or other similar conference. I wonder how that changes over time. But still, it lets ad-hoc conversations arise, sometimes with combinations that have not really happened on other platforms, e.g., Shane Battier, former NBA player, asking a question to Chris Voss, former FBI hostage negotiator. Again, sometimes the interactions bore me, but sometimes they thrill me, and that feeds to the variable ratio reward addiction mechanism of the app.

Just like horses.

I agree completely with your first paragraph.

> Even podcasts, which I'd argue are more accessible than live rambling, are not a mass market product after more than a decade of trying.

In the U.S., 37% of Americans surveyed by Edison Research listened to at least one podcast during the last month. (It was 32% the previous year.) That means podcasts are about 50% more popular than Twitter or LinkedIn. Is that not a mass medium?

http://www.edisonresearch.com/wp-content/uploads/2020/03/The...

That research firm seems heavily invested in podcasts succeeding by the looks of it. If podcasts were actually that popular, wouldn't there be a Twitter/Instagram/LinkedIn equivalent site for podcasts? To the best of my knowledge it's just a feature inside of iTunes for the most part.

I also think we'd see a big advertising industry around podcasts. I have to admit to not being an expert in that particular market, but it sure doesn't seem like podcasts have the scale of other consumer social media.

How does UDP work for broadcasting audio? I can understand UDP from a speaker's phone to a CH server, but how would UDP packets be sent to thousands of listeners, many of them hidden in NATs?

What would be a better architecture? If I had to build this I'm thinking a fleet of Phoenix/Erlang servers accepting websocket connections and distributed-ly broadcasting each packet over them, think there's some WebRTC tools for this.

Any better ideas? Go with memberlist UDP transmission maybe?

Any serious audio transmission is going to be done over UDP as otherwise you get ridiculous head of line blocking issues where a single lost data packet can cause a blip in the audio at least as long as a full round trip. NAT--even broken symmetric NAT--is not an issue, as the client is establishing the channel to a fixed, public endpoint.

For sending the audio from the speaker to Clubhouse, sure.

What about the broadcast to all the listeners? Here the server is sending UDP packets to thousands of clients? Not sure how NAT works with UDP incoming into the NAT - isn't UDP connection-less and there's no indication from the server side what the internal network layout is?

You use ICE/StUN servers to solve this.

The trick to punching UDP through NAT is to make the client do an outbound transmission first.

Now your NAT is keeping an eye on this port and will forward incoming packages to said client.

And what the sibling says : all solved with STUN/TURN/ICE. This is decades old tech.

no one should use this app.

quite honestly Clubhouse should pull their servers down until these issues are fully mitigated. What a shitshow.

Obligatory unrolled thread:

https://threadreaderapp.com/thread/1361866446427488256.html

Link to the original (Korean) article:

https://theori.io/research/korean/analyzing-clubhouse/

The real tl;dr:

Clubhouse made some weird security decisions and seems like you shouldn't trust them with state secrets. Also, weird traffic going to China.

Reply on news.ycombinator.com