Fetching models/prices/best model from github instead of more migrations

[drnic]

It's another week and there are new openai/anthropic/groq models available and some price changes, etc.

Currently we have LanguageModel (per User) and I think the only way to programmatically populate this model is via migrations.

I'd propose that we document the available models in this repo, in a models.json file, which could then be fetched via HTTP at any time, by any running hostedgpt app to get latest values.

For example, to fetch the raw value of fly.toml in the root of the repo, visit https://raw.githubusercontent.com/AllYourBot/hostedgpt/refs/heads/main/fly.toml

So a models.json would be HTTP retrievable at https://raw.githubusercontent.com/AllYourBot/hostedgpt/refs/heads/main/models.json

An airlocked HostedGPT could just load the models for new Users from this file.

An internet-based HostedGPT could fetch the file from the trusted main branch from github.

The Language Models page could have a "Sync from Github" button

[krschacht]

That’s a great idea. I never considered using the repo to serve a fetchable file like that.

There is one thing we have to keep in mind. Changes to the models.json file are not handled solely with INSERTs. Some existing lines will change, like the default model, and those require UPDATEs to existing records. But that won’t be too tricky the sync code could just iterate through all of them. We have a few PRs where we’ve done this (for the past model changes) so we can use these as a guide.

Is this something you could take the lead on creating a PR for? I can help review and test it out.

[drnic]

First draft at #536

[drnic]

In this PR, it provides tools for insert/update from a local models.yaml file; and another rake task to regenerate this file.

Future work could allow this models.yaml file to be fetched remotely via a button in the UI.