What is 'Bring-Your-Own-AI'?
Bring-Your-Own-AI is not a feature of QC Database, but rather a way of describing how it is used. As the name implies, QC Database is primarily a database or in other words, a data storage for your construction quality documents and tables. QCDatabase.AI is the version of this database that is built for the AI age, but AI is not required. Currently, you can use QC Database in three ways in terms of Artificial Intelligence:
AI-Free
Essentially the same as the legacy QC Database app. Everything is manual data entry. Users fill out forms and click submit buttons. Nothing changes except some design improvements over the old system. The CSA Admin will have to turn off Cloud AI tools like drawing BOM extraction and AI document extraction.
Cloud AI aka as a 'GPT-wrapper'
QCDatabase.AI has some "GPT-wrapper" features, meaning we have plugged an LLM into some features on the backend. Things like auto-tagging NDE reports to welds, populating the welder list from uploaded WPQ's, pulling heat numbers from MTR's, etc. Your CSA will have to turn these features on and configure them properly for them to work. In this use case, we are treating AI like a dumb machine. We feed it one document at a time and we always ask it to do the same thing with that document.
Bring-Your-Own-AI aka 'Agent Workflows'
If you have Codex or Claude Desktop installed on your laptop, you have something called an 'AI Agent' that can do work for you. The definition of 'AI Agent' is not quite settled in 2026, but for our purposes here, an AI Agent can make decisions for you because you give it a degree of AGENCY. You can use permissions to limit how much agency your AI agent has, but in this style of work, AI is treated more like your employee instead of a dumb mechanical tool.
For an example of Agentic workflows, you can give Claude a stack of documents including RT reports, PQR's, and hand-filled torque reports, and Claude will file these into QC Database using our MCP server. Claude is Agentic because it decides where to put the files. This is what we could call mild agency. The AI Agent is still treated like a low-level clerk, not a competent employee. A child could figure out how to file documents properly.
For a more advanced agentic workflow, I could tell Claude "Turn in those last few socket welds for Steve". In this scenario, Claude can look up my past actions and see which drawing I was working on, make sure it has socket welds that aren't turned in, look up the welder ID for "Steve", and even if it finds two or SEVEN welders named Steve, it can still determine that the drawing I was just working on is in fact being welded by Steve Smith, welder No. 1234. In this case, it is obvious who I am talking about. Here the AI can be more of an "Agent" and decide to put welder ID 1234 in on the welds that it thinks I am talking about.