Show HN: Use LLMs to Interact with APIs
github.com
Recently featured in a LangChain blog https://blog.langchain.dev/empowering-development-with-flowt... , use LLMs to construct an API first runnable workflow with an IDE experience.
github.com
What joshuanapoli said is absolutely true, "user has a chance to review and debug the workflow before using it"
but majorly we have put guardrails that filters out the part of the output that LLM hallucinated or is incorrect
LLM-to-automation-api seems really great on paper but the statistics are simply against any sort of wide adoption
at least that is the general consensus speaking to clients
Thank you for voicing your opinion. We actually talked to a lot of individuals and we found that particularly people who are part of frontend team or part of a QA/QE or CRE team who are not the primary developers of the backend/api were most interested in the LLM to API workflow feature, partly because going through the OpenAPI spec and understanding each endpoint is a lot of work. What our tool does it that it not only identifies the right endpoints and chain them but also pre-fills the request parameters or request body which gives them a chance to quickly experiment with backend.
Of course you always have the option of drag and drop. But we hope that one day using LLMs will get commoditized with models being quite good and then that day a natural language translation will seem obvious
Aside from that one of the major problems that we found that not a lot of folks want to expose their api authentication credentials or in-development/private apis to tools like postman or zappier. So giving an IDE like experience that works as a local desktop app helps alleviate a lot of privacy issues.
Graphical representation of end to end tests was also a big ask