OneTag AI

Free OCR API: image to text from a web page, no backend

Is there a free OCR API that turns an image into text without a server?

Yes. OneTag AI reads images with a vision model, and each email address gets a key with 1,000,000 free tokens a month and no credit card. In our tests a full page of text cost about 2,200 tokens, so the free key reads about 450 full-page images a month, or about 1,900 small ones such as receipts. The key is safe to leave in page source, so the page itself can send a photo and get the text back, with no backend.

It returns plain text, or JSON in the shape you ask for. It does not return word positions and it does not read PDFs; the comparison below lists the free OCR APIs that do.

Steps

1. Get a free key

   Enter your email at onetagai.com. The key (it starts with ot_) arrives by email. One key per email, no card.

2. Read the text in an image

   Pass the image file as the input. The script scales it down to at most 1,600 pixels on the long side before it is sent:

   <input type="file" id="photo" accept="image/*">
   <pre id="text"></pre>
   
   <script src="https://onetagai.com/v1.js" data-key="YOUR_ONETAG_KEY"></script>
   <script type="module">
     const photo = document.getElementById("photo");
     photo.onchange = async () => {
       document.getElementById("text").textContent = await onetag.ask(
         "Transcribe all the text in this image exactly. Keep the line breaks. Return only the text.",
         photo.files[0],
         { maxTokens: 2000 }
       );
     };
   </script>

   The input can also be an <img> element that is already on the page.

3. Get fields instead of text

   Describe the JSON you want and set json: true. The answer is a parsed object:

   const receipt = await onetag.ask(
     'Read this receipt. Return JSON {"merchant": string, "date": "YYYY-MM-DD", "total": number, "items": [{"name": string, "price": number}]}.',
     photo.files[0],
     { json: true, maxTokens: 1000 }
   );
   console.log(receipt.merchant, receipt.total, receipt.items);

4. Or call it over HTTP

   From a server or a script, send the image as a base64 data URL. Python needs only the standard library:

   import base64, json, urllib.request
   
   image = base64.b64encode(open("receipt.jpg", "rb").read()).decode()
   body = {
       "key": "YOUR_ONETAG_KEY",
       "prompt": "Transcribe all the text in this image exactly. Keep the line breaks. Return only the text.",
       "input": [{"type": "image", "data_url": "data:image/jpeg;base64," + image}],
       "max_tokens": 2000,
   }
   req = urllib.request.Request(
       "https://onetagai.com/v1/ask",
       data=json.dumps(body).encode(),
       headers={"Content-Type": "application/json"},
   )
   with urllib.request.urlopen(req) as res:
       print(json.load(res)["text"])

What we measured

Both tests ran against the live API on 2026-11-10, with images we generated so the right answer was known.

These were clean, printed, high-contrast images. Handwriting, blurry photos and very small print are harder: test with your own images before you rely on it.

Free OCR APIs compared

Other providers' figures were read on their own pages on 2026-11-10; follow the link in each row for the current numbers. Free tiers change often.

Which free OCR API should you use?

• You want fields, not raw text (the total on a receipt, the name on a card), from a static page: OneTag AI. You describe the JSON and get it back in one call.
• Images must stay on the visitor's device, or you need no quota at all: Tesseract.js. It runs in the browser; the trade is a larger download and weaker results on photos.
• You need PDFs or many small scans: OCR.space. 25,000 requests a month, and it accepts PDFs up to 3 pages on the free plan.
• You need word positions, or you process documents at volume on a server: Google Cloud Vision, Azure AI Vision or Amazon Textract.

Limits to know before you rely on it

• One image per call. In the browser the script resizes it first, so a large phone photo is fine; over HTTP the image can be up to 2 MB. PDFs do not work.
• No word positions or bounding boxes, and no confidence scores.
• It is a vision model, not a classic OCR engine. It can misread a blurry digit, or tidy up a typo it should have copied. For numbers that matter, such as invoice totals, show the result to the user to confirm.
• Answers are 500 tokens long by default. Set maxTokens (max_tokens over HTTP) to 2,000 or more for a full page, up to 10,000.
• 60 requests a minute per key and 20 a minute per IP address.

Questions

Related guides

• Free AI API for websites: no backend, no credit card
• Free translation API for websites: no backend, no credit card
• Free chatbot API for websites: chat history, streaming, no backend
• Is it safe to put an AI API key in a static HTML page?

Updated 2026-11-11. Something wrong or missing? Email support@odysseyx.app.