Hey everyone, I've been deep in a trial with Amazon Q Developer for the past few weeks, trying to integrate it into our team's workflow. The promise of an AI assistant that understands our specific tech stack is incredibly appealing, especially for onboarding new devs. 😅
However, I've hit a major snag that I know others in regulated industries or with sensitive IP will face. Our core value is wrapped up in our internal APIs and proprietary services. I want Q to help developers generate code that calls these internal endpoints correctly—think authentication flows, specific query parameters, and response handling—but the documentation is clear that you *cannot* upload your source code to AWS for custom training. The standard knowledge base connectors are for public URLs, Confluence, SharePoint, etc.
So my big, burning question is: **How are teams handling this?** I'm looking for practical, real-world patterns.
I've been experimenting and taking notes, and I see a few potential paths, but each has trade-offs:
* **Heavy Prompting in the Chat:** This is the most obvious. You paste your API documentation (or relevant snippets) directly into the chat prompt every single time. I've found you need to be *extremely* detailed—providing example curl requests, response JSON schemas, and auth headers. It works for one-off questions, but it's verbose, repetitive, and doesn't scale for a team.
* **Creating "Pseudo-Public" Documentation:** The idea here is to stand up an internal documentation portal (like a read-only Sphinx or Redoc site) on a VPN or behind IP whitelisting, but then... Q can't access it unless it's publicly reachable, which defeats the purpose. So this seems like a non-starter unless I'm missing a connector type.
* **Leveraging Generic Patterns with Placeholders:** Training the team to ask Q for generic code patterns (e.g., "generate a Python function for a POST request with OAuth 2.0 client credentials grant") and then manually filling in our specific endpoints and secrets. This is safer but loses a lot of the "assistant" magic.
Is there a fourth option? Has anyone from AWS hinted at future on-premises or VPC connector options? I'm particularly worried about the risk of developers accidentally pasting large chunks of our actual codebase into the chat, which would essentially be the "uploading" we're trying to avoid.
Would love to compare notes and see what workarounds the community has engineered. The tool feels powerful, but this internal knowledge gap is a significant hurdle for adoption in my environment.
~TrialHunter
Another trial, another spreadsheet