A field by its label
Name an index field after the label printed on the form, "Patient Date of Birth", and the model finds the label and reads what's written next to it.
Use OCR for most data, and your favorite AI model for the rest
SimpleIndex 12 can hand the hard parts of a document to a language model: Claude, ChatGPT, Gemini, Perplexity, Azure OpenAI, DeepSeek, or an open model such as Qwen or Llama running on your own server. Local OCR still reads the full page and makes it searchable; the model is asked only for the fields a template can't describe. You add your own API key and pay the provider directly. We take no cut and charge nothing per page.
01
What a model reads that OCR can't
Each of these runs on the fields you choose. The rest of the page stays on the local engine.
Handwriting
Draw a zone around one handwritten box and hand just that zone to the model, while the rest of the page stays on the local engine.
Marks
Checkboxes and filled circles come back as values.
Document type
Ask the model what kind of document a page is, and route it by the answer.
Tables and line items
A table field comes back as rows, such as an invoice's line items, written out as CSV beside the other values.
Translation
Ask for the page text in English, Spanish, French and other languages, with the original kept alongside it.
The job decides what the model sees
Local OCR reads the page
Tesseract, ABBYY FineReader or a cloud OCR engine reads the full page and makes it searchable. A model can't do that part, because it doesn't report where each word sits on the page.
The easy fields are done first
Zones, patterns and barcodes fill most fields, at no cost per page.
The model gets only what's missing
It's asked for the fields still empty after local OCR, and only on the pages you allow.
Nothing is paid for twice
Fields the model filled aren't sent again, and the fields local OCR already read are never sent at all.
Your model, your key
Pick a provider and a model in the job: Claude, ChatGPT (OpenAI), Gemini, Perplexity, Azure OpenAI with your own deployment, or DeepSeek. Any server that speaks the OpenAI API works too, which covers Ollama, LM Studio and vLLM on your own hardware and hosts like Groq and Together. A cloud provider bills you by tokens, not by page.
A model on your own server costs nothing per page. It has been tested on a server with 8 GB of memory, and it can run with no internet connection at all.
Your API key is encrypted with Windows on the computer that saved it, so a job file can be copied to another machine without taking the key with it. One key can also be set for the whole computer, so every job on it uses the same account.
Three ways to send a page
Hybrid, the default, sends the model the page image and the OCR text with where each word sits. It suits forms, handwriting and checkboxes, where the model needs to see the page.
Text only sends just the OCR text and positions, no images. It costs the fewest tokens and suits typed documents, and it works with models that can't read images.
Images only lets the model read the page itself, with no local OCR. The PDF then gets no searchable text layer, because the model doesn't say where the words are.
OCR or LLM?
Most documents don't need a model at all, and the cheapest page is the one it never sees. The case against sending everything to a model is in our article OCR vs. LLM: You Don't Need AI for That!
Before you point a model at your documents
Which models can I use?
Claude, ChatGPT, Gemini, Perplexity, Azure OpenAI and DeepSeek, or any model behind an OpenAI-compatible endpoint, such as Ollama, LM Studio or vLLM on your own server. A model name can be typed in, so a new release doesn't have to wait for an update from us.
What does it cost?
The SimpleIndex license is paid once, and Simple Software charges nothing per page. A cloud model's provider bills you by tokens; a model on your own server costs nothing per page.
Can the model make my PDFs searchable?
No. Full-text search comes from the local OCR pass, which records where every word sits. The model supplies field values.
Which edition do I need?
Professional. LLM OCR runs under the Cloud OCR license, which Professional includes, so a Standard license that already carries the Cloud OCR add-on can use it too. The trial includes it.
Do our documents leave the network?
Only the pages you send to a cloud model. A model on your own server keeps everything on your hardware.
Can it return JSON or XML instead of fields?
Yes. Turn on raw model output and the response comes back exactly as your prompt asked for it, whether that is JSON, XML or a format of your own.
Where is my API key kept?
In the job, encrypted with Windows so it only works on the computer that saved it, or once for the whole computer. It is never sent to us.
Try it on your hardest documents
The free trial runs the full software on your own files.
