Hey everyone! Still getting my feet wet with all these AI/MLOps terms, so please bear with me 😅
I've been exploring Relevance AI for a potential side project. I keep seeing the term "tool" everywhere in their docs and UI. From my DevOps background, my mind immediately goes to "this is just an API endpoint wrapper," but that feels too simplistic.
Can someone explain what makes a Relevance AI "tool" more than just an API call? Like, if I use their pre-built Google Search tool, is it literally just hitting the search API with my query, or is there extra orchestration or logic happening under the hood they manage for me? Trying to understand the abstraction here. Thanks in advance!
> "this is just an API endpoint wrapper"
I had the same thought when I first saw it. Coming from a general IT background, a "tool" sounded like a fancy name for a function call. But then I tried their pre-built Slack tool and noticed it handles auth tokens, rate limiting, and error formatting automatically. So it's definitely more than a raw HTTP request.
What I'm still fuzzy on is how the orchestration works. Is Relevance AI running some kind of lightweight workflow engine behind the scenes? Like if I chain two tools together, does it handle retries or timeouts for me? Or is that all on me to code?
CloudNewbie
You're right to be skeptical. It's mostly just an API wrapper, but with the vendor's specific glue logic baked in. The "extra orchestration" you're asking about is usually just their own error handling and request formatting, which locks you into their platform's way of doing things.
So yes, their Google Search tool hits the API. The abstraction is that they manage the API key and parse the response format for you. You trade control for a bit of convenience.
If you ever want to move off their platform, you'll have to rewrite all that "tool" logic yourself. It's a calculated bet.
Your vendor is not your friend.