A few current features
- The most basic model info page: https://t.co/nj6UJoFGC5
- View all models from a specific provider: https://t.co/efb8OsALgi
- View which providers offer a given model: https://t.co/7YprMJCzTi
Built a model database / information hub
https://t.co/uTJDM1YPxv
The processing scripts and data are in https://t.co/mPyZSkpX4G. If a program wants to integrate with it, it can just directly use the JSON files there.
Also, if you look at the JSON data, there is some other information as well (not yet shown in the UI), such as what reasoning effort levels different models support, or which providers offer the same protocol, and so on.
Another reason is that there are now more and more Conditional Pricing rules, but various model libraries do not include them, which makes it very difficult to implement pricing properly. So I decided to build one myself.
The main data sources are https://t.co/ZV72fuSoEU and LobeHub, plus manual corrections. The original reason I built this was that the flat list on the https://t.co/ZV72fuSoEU website was really unpleasant to browse.
I just think, you need to use AWS (or other EC2 services) only if you need HA (or upgrade but can’t bear downtime).
Durable (or backup) is easy, and is pretty cheap on a VPS. But HA is really expensive if you want to achieve without managed services.
You can't run a business on a VPS.
It's unreliable.
You need AWS.
You need microservices.
You need Kubernetes.
(This server costs me 13 EUR/month on @Hetzner_Online and has been running for 3 years.)
@_arupbasak_@jetbrains Oh no. Different languages need different presets. When using the cursor, one of my most disliked things is to create different workspaces and configure them.
@jarredsumner I believe this shouldn’t be in runtime. I believe it’s valuable to provide an interface to reflect file system (like golang FS interface https://t.co/2kLk0nBSFx) but not directly provide a third-party implementation.