HACKER Q&A
📣 parthkm

How are people building SaaS integrations for AI agents?


How are people building SaaS integrations for AI agents?


  👤 parthkm Accepted Answer ✓
AI systems would need full control over the Integration APIs. Thus none of the existing types of closed integration providers would be able to help: Unified APIs (merge.dev), Ipaas workflows (workato) and ETL (fivetran). It has to work with open source integrations.

However, all open source integration repositories are filled with custom code. Each repository choosing how to code in their own format and structure for authentication, api configuration and several of them mapping to their own formats. Not seen a single one of them actually use the OpenAPI Specification (OAS) to create their repository of integrations. You cannot really convert OAS to custom code specific to your application and vice versa.

Some of the Open Source Integration repos I have looked at: cptn.io, windmill, cloudquery, airbyte, revert, n8n, active pieces, singer, rudderlabs, handshake(only for oauth.

I believe AI systems and companies in general looking to work with integrations would really benefit from configuration objects that define how to use the apis of all saas platforms, especially those that do not have the OAS.

Would an open source standard to define Oauth (like the gist link) be useful? Furthermore a configuration object and parser that can be reproducible from OAS to be used in interacting with saas apis? Lastly an open source mapper (based on jsonata) to seamlessly convert one schema to another.