3 ms·

This is the idea of local-first software[0]. Imagine you owning your own database and the only thing you get when you go to a website is the software and none o

This is the idea of local-first software[0]. Imagine you owning your own database and the only thing you get when you go to a website is the software and none of the data (it accesses your database instead). Projects like Textile[1] are building out tools that could help with that sort of project. I really hope it takes off, the cloud sort of freaks me out. [0] https://www.inkandswitch.com/local-first.html https://www.inkandswitch.com/local-first.html [1] http://textile.io/ http://textile.io/

yrgulation 6y ago
Either that or each user’s records are encrypted using their own password and forgot password recovery question/s answers. Here is an example of how it can be done: https://security.stackexchange.com/questions/30193/encrypting-user-data-using-password-and-forgot-my-password https://security.stackexchange.com/questions/30193/encryptin... Although this approach still requires trust that service providers do indeed encrypt user data.
Scene_Cast2 6y ago
So the core problem is that when something is running on someone else's server, you can't even verify what they're running. E2E encryption avoids that by not trusting whatever is running on the servers Local DBs avoid that by not giving up local data. However, it would be quite interesting to have a way to remotely know that a certain service is running the code you think it's running.
- jaywalk 6y ago
  > it would be quite interesting to have a way to remotely know that a certain service is running the code you think it's running. It sure would, and it would be a hell of a discovery if someone could come up with it. Because I sure can't think of a way that I can't easily debunk.
  - Scene_Cast2 6y ago
    Intel has done some work on this front that can hypothetically be used this way, but I wouldn't say it's practical, for various reasons.
- DSingularity 6y ago
  You mean like intel TXT remote attestation?
  - Scene_Cast2 6y ago
    The trick here would be integrating that to play nicely with load balancers, REST-style APIs, and reducing overhead from establishing yet another secure connection.
hezag 6y ago
Related: Solid Project - HN discussion: https://news.ycombinator.com/item?id=25989698 https://news.ycombinator.com/item?id=25989698 - A great article about the project: https://ruben.verborgh.org/blog/2020/12/07/a-data-ecosystem-fosters-sustainable-innovation/ https://ruben.verborgh.org/blog/2020/12/07/a-data-ecosystem-...
- Jugurtha 6y ago
  Slightly related, we're working with a similar philosophy. As a machine learning consultancy that has done many learning projects for enterprise, we're building our machine learning operations, "MLOps", platform (https://iko.ai https://iko.ai) to simplify our work. However, what we're doing is working from the architecture level to have as little and preferrably no sensitive information on our service. We're architecting it so that you give us specific access to deploy on your cluster, and everything happens there: the notebook servers are there, your data is where you choose to put it, your training jobs are there, your experiments are tracked there. Your models are deployed there. I have a saying that the platform should be able to run on a Raspberry PI. One of my personal pet peeves working with the team is to be able to disappear without impacting them, and it has become the same with our platform: it must be able to disappear users having to scramble to exfiltrate or export their work or data from our infrastructure, because it simply is not there.

yrgulation 6y ago

Either that or each user’s records are encrypted using their own password and forgot password recovery question/s answers. Here is an example of how it can be done: https://security.stackexchange.com/questions/30193/encrypting-user-data-using-password-and-forgot-my-password https://security.stackexchange.com/questions/30193/encryptin... Although this approach still requires trust that service providers do indeed encrypt user data.

Scene_Cast2 6y ago

So the core problem is that when something is running on someone else's server, you can't even verify what they're running. E2E encryption avoids that by not trusting whatever is running on the servers Local DBs avoid that by not giving up local data. However, it would be quite interesting to have a way to remotely know that a certain service is running the code you think it's running.

jaywalk 6y ago
> it would be quite interesting to have a way to remotely know that a certain service is running the code you think it's running. It sure would, and it would be a hell of a discovery if someone could come up with it. Because I sure can't think of a way that I can't easily debunk.
- Scene_Cast2 6y ago
  Intel has done some work on this front that can hypothetically be used this way, but I wouldn't say it's practical, for various reasons.
DSingularity 6y ago
You mean like intel TXT remote attestation?
- Scene_Cast2 6y ago
  The trick here would be integrating that to play nicely with load balancers, REST-style APIs, and reducing overhead from establishing yet another secure connection.

hezag 6y ago

Related: Solid Project - HN discussion: https://news.ycombinator.com/item?id=25989698 https://news.ycombinator.com/item?id=25989698 - A great article about the project: https://ruben.verborgh.org/blog/2020/12/07/a-data-ecosystem-fosters-sustainable-innovation/ https://ruben.verborgh.org/blog/2020/12/07/a-data-ecosystem-...

Jugurtha 6y ago
Slightly related, we're working with a similar philosophy. As a machine learning consultancy that has done many learning projects for enterprise, we're building our machine learning operations, "MLOps", platform (https://iko.ai https://iko.ai) to simplify our work. However, what we're doing is working from the architecture level to have as little and preferrably no sensitive information on our service. We're architecting it so that you give us specific access to deploy on your cluster, and everything happens there: the notebook servers are there, your data is where you choose to put it, your training jobs are there, your experiments are tracked there. Your models are deployed there. I have a saying that the platform should be able to run on a Raspberry PI. One of my personal pet peeves working with the team is to be able to disappear without impacting them, and it has become the same with our platform: it must be able to disappear users having to scramble to exfiltrate or export their work or data from our infrastructure, because it simply is not there.