add initial documentation for available integrations

[davanstrien]

[WIP] PR to explore some options for documenting tools that integrate with/can use Inference Providers. The actual content is mostly a placeholder. Mostly want feedback on the structure.

The idea is to highlight all the tools that can use Inference Providers with a small guide on our side and links to the tools.

If we do one page per integration, it will look something like this, i.e. more below/above MacWhisper.

@Wauplin, we originally discussed doing a table. This would look something like:

The left column should link to the one-pager on the Hub docs, and the second column should contain an external link. I already have quite a long list of things that could go here, so this table might end up a bit long but probably it's okay to start.

Another option is not to use a table and do some light grouping, but this can get a bit messy with overlapping categories.

@Wauplin @hanouticelina let me know what you think. If you are happy with the structure of doing table + one pager, I can tidy up PR to just add one or two integrations to start. I am slightly worried that we end up maintaining docs for all of the integrations too, so we could also drop the one-pager, i.e. only do open code as a full code and don't do a page for every integration.

also cc @julien-c @pcuenca for viz.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[Wauplin]

Hey @davanstrien , thanks for starting this up and sharing your thoughts! I was first in favor of a table + no categories to avoid the categorization issues. But now that you've written both version, I do like the index-simple.md one with categories. Next step is to correctly define categories but I feel that's you've started a great job here.

3 things that comes to mind:

• IMO if an integration can go in two categories, let's list it twice
• inside a category, let's list integrations alphabetically
• maybe no need for a one-pager for each integration if we want to avoid maintenance issues. If we feel that the integration is already well documented on their side and that a one-pager wouldn't bring much value, we can always skip it. What I do like though with the one-pager is if it can bring better SEO (but hard to tell...)

So my opinion would be 1. lists in categories instead of table 2. ok for duplicate in lists 3. let's not force ourselves to make 1 page per integration but would still be nice for the main ones.

Let me know what you think!

[julien-c]

agree with those comments!

[pcuenca]

I like the rationale and explanations in index.md that could be carried over to index-simple.md if we go with categories. Personally, I find a table easier to scan for a relatively small number of items, perhaps we can use categories and a summary table that is exhaustive?

In favor of adding one-pagers for quick configuration with inference providers, when we can. If we link to the official docs, we could try to link to a section that describes this use. But agree there's no need to force us to always provide a dedicated page per tool.

[davanstrien]

Thanks for the feedback/suggestions! Have tidied this so it should be ready for a review of the actual content.

I have added both the overview table and the category list. We can revisit this at some point if the index page gets very long.

There are a few follow-up PRs I'd do separately:

• We can add a bunch more integrations (this last was a very quick pass) to the table/list (and refine the categories as we add more integrations).
• There are some tools for which a page in our docs would be worthwhile.
• A "how to integrate with Inferencre Providers" guide. This will be a bit more focused on building integrations versus apps, the document cheapest, and how to retrieve supported via the API etc. Some of this info is in other pages, but it would be nice to have a single source for potential integration to refer to.
• It might be good to have a short guide on "how to use any tool with Inference Providers via its OpenAI-compatible endpoint". Again, this information is already in the documents, but it's nice to have a single page that quickly covers this.