I just learned in the documentation that
I just learned in the documentation that Auth is not charged by Convex at all, you'll pay the provider directly if you go above the limits right?
42 Replies
I'm not sure which limits you're referring to, but there's no usage charging specifically for Convex Auth
I mean MAU
like the limits are the provider limits, you don't charge additional costs
I've seen you asking about MAU. There's no MAU concept to Convex Auth
This is different from say Firebase Auth, which does have MAU
Yes that's what I was saying thanks
There are no limits to user counts in Convex Auth
the limits are the Google provider limits for example
As an oversimplification, Convex Auth is the infra around a database table
which is 10.000 free then you pay if i remmeber correctly
Yeah, GCP has an auth solution like that. In my experience, Firebase Auth is not really innovating anymore, and it's being converted to GCP's user identity solution, which is significantly more expensive
I dont know about firebase i don't use it.
I use directly Google OAuth https://developers.google.com/identity/protocols/oauth2
Which is 10k users free then 0.1$/MAU i think
Google for Developers
Using OAuth 2.0 to Access Google APIs  | Authorization  | Google ...
Convex have a great potential i like the direction so far
I feel like I need to be careful about terminology here. OAuth is a protocol about asserting an identity. OAuth is not about counting users. I think you're referring to Google Identity, or something like that (I forget the brand name), which is about counting users and charging you for it
Yeah exactly I don't know what's the name either, i just use it form the Google console
0.1$/MAU doesn't sound right. That's crazy expensive. Even more expensive than Auth0, which is 7 cents a user
let me check
Like I don't know what is this called ?
That's still about OAuth setup
I think you are looking for Google IAM pricing
is it the same as "Google Identity" you were refering to ?
yeah
I'm pretty sure "Google OAuth" is free (although I can't find a clear reference)
I don't think it's free https://cloud.google.com/identity-platform/pricing
can you check please ?
That's not Google Sign In
To my understanding, that's a competitor to Firebase Auth / Clerk etc.
Wow, I hadn't looked at GCP's Identity Platform in a while. It's MAU price is basically a fourth of Clerk's. That's nice.
But not really a fair comparison. I don't think one can readily integrate GCP Identity Platform with say, a Next.js app. There's also no UI SDK, or authorization logic
I found this on the console
Nothing is clear at Google 😅
Oh I found it !
by default there is a threshold of 10k token limit per day to avoid abuse
that's a fundamentally different concept to MAU
Yeah
All right, well good to meet Axibord. I see you're new to the server. Best of luck with Auth!
Thanks a lot Matt and Michal
yeah, basically, auth costs what it takes to run the infra to serve it... on convex
there is no additional charge for auth
and we don't really plan to charge for any future components we release either. they cost what it takes to run/store them in terms of convex platform resources
I hope it will get widely adopted to maybe allow you to get more advantageous pricing.
Can you share why the Vector storage cost is so high, I'm a former data scientist, I guess it's the dimensions of the vectors that makes it inherently take a lot of space?
relative to other providers? or relative to like, other database data, and search indexes, etc?
not relative to other providers, but in general, like why it's way more?
ah gotcha
what makes it cost that much
yeah, it's pretty resource intensive to make really fast vector indexing compared to most other things
Ohhh so it's on the CPU, not the storage itself, it's the action of storing it?
at the end of the day, usually higher costs are a function of how memory intensive things are more than anything else
b/c on the CPU side, the challenge is always: make this fast enough to serve for live traffic
ergo, the CPU requirements end up being similar for most things. they are sort of "held constant"
then the rest of the design space is, "how many resources does it take to make that thing hit the universal latency need for live applications?"
and the biggest lever there is usually memory
how memory intensive are the indexes necessary to achieve that performance target
I understand way better now, thank you so much. I didn't thought that way about memory, and now I can see why