Set Up OCR in Your No-Code App: One Integration, Zero Backend Code
A small post called "OCR It" has been making the rounds on Hacker News this week. It does one thing: pulls text out of un-copyable documents so an LLM can actually read them. Simple idea, huge demand, because half the useful data in the world still lives inside PDFs, scans, and screenshots that won't let you Cmd+C anything.
If you're building a business app, you'll hit this problem eventually. A user uploads a receipt. A tax form. An old invoice scanned by a photocopier from 2004. Your app either reads it or your users retype it.
Here's how to wire OCR into a DontCode app without touching backend code.
Step 1: Add the integration
Open Integrations in your project. New Integration, pick your OCR provider. Google Cloud Vision, AWS Textract, and Mindee are all solid options with a simple HTTP API and a key on signup.
DontCode asks for three things: a name (like "OCR"), a base URL (the provider's API root), and credentials. Paste the API key into the credential slot. It's encrypted with pgcrypto and never readable again from the UI. At deploy, DontCode injects it into your app as INTEGRATION_OCR_API_KEY.
Step 2: Write the intent, not the code
Add an outbound function to the integration. Instead of writing code, you write an intent. Something like:
When called with { fileUrl }, download the file, POST it to the provider's OCR endpoint, and return { text, confidence }.
DontCode's build agent turns that sentence into the actual function file. Change the intent later and it regenerates. No copy-pasting from Stack Overflow, no dependency updates, no CORS surprises.
Step 3: Add it to a page
Open the App chat and say: "Add a page where users upload a receipt. After upload, run the OCR function on the file and show the extracted text next to a Save button that stores it in a receipts table."
That's the whole prompt. DontCode uses storage (private bucket, signed URLs), calls your new OCR function, wires the result into the DB, and drops a page into your Todo List when the build is done. Review, keep what you like, roll back what you don't.
The shape scales
Most integrations follow the same three steps: credentials, one intent, one page. Stripe, Slack, a shipping API, a background check service, anything with a REST endpoint. If you can set up OCR, you can set up any of them.
The interesting part of an AI app is the plumbing, not the LLM call. DontCode is our take on that plumbing. Less config, more shipping.
Try it at dontcode.co. More walkthroughs on the blog.