When we set the architecture for Parseloom in mid-2024, we had a choice: build a strong GUI workflow and add an API later, or build the API first and layer the GUI on top. We chose API first. That choice has shaped everything since.
The problem with GUI-first document tools
Most document extraction tools were designed for a person to sit at a screen, upload a PDF, review the extracted fields, and click save. The API is added later, often as a wrapper around the GUI workflow. That means the API is slow, has awkward state management, and returns data in shapes that make sense for a UI but are clunky to consume programmatically.
More importantly, GUI-first tools get stuck on the human review step. They're designed around the assumption that someone will always review every extraction. That assumption is fine for low-volume workflows but doesn't scale.
What API-first actually means in practice
For us, API-first means three things. The extraction endpoint is designed for machine consumption: deterministic request format, predictable JSON response shape, no UI state dependencies. Every feature we build has to work via the API before we add a GUI control for it. And the confidence scoring is built to support automated downstream decisions, not just to give a human a review queue.
The result is that our early-access teams mostly use Parseloom without ever opening the web interface. They submit documents from their own systems, parse the response, and push data into their ERP in one automated flow. The web interface exists for configuration and debugging, not for daily document processing.
The tradeoff
API-first made the initial product harder to demo. A GUI walkthrough is easier to understand in a sales call than a curl command and a JSON blob. We've learned to show the JSON response alongside a visual representation of the document with the extracted fields highlighted, which bridges that gap somewhat.
The teams that get the most value from Parseloom are the ones with a developer who can spend a day wiring the API into their existing document intake flow. If that's not you yet, the GUI is there and it works. But the API is the real product.