AI agent company AgileSoDA said on Thursday it has launched Agentic OCR, an AI document processing automation platform that can be applied without separate model training.

The company said the key is that it has shortened the work needed to add new document types, which previously took weeks to months, to the same day.

Existing AI OCR (optical character recognition) technology required collecting training data by document type and training models separately. Even a small change in a form meant repeating retraining from scratch.

AgileSoDA said Agentic OCR automatically finds the relevant information in a document using large language models (LLMs) and vision language models (VLMs) once a person in charge defines the items to extract, such as a name, contract date and amount. When a sample document is uploaded, it automatically generates a draft for classification and definitions of extraction items. No coding or separate training data is needed. The person in charge can begin processing immediately that day after review and edits.

AgileSoDA said its own verification found 98 percent accuracy in document classification and at least 95 percent accuracy in information extraction in tests covering 20 types of unstructured and structured documents.

AgileSoDA CEO Dae-woo Choi (최대우) said, "Corporate work ultimately starts with documents." He said Agentic OCR is not an information extraction tool that extends existing OCR, but "a starting point toward a document agentic platform that covers overall corporate work, with documents as the point of contact, from retrieval-augmented generation (RAG)-based chatbots and document comparison and analysis to business automation."

AgileSoDA aims to expand its pilot customers for Agentic OCR through the second quarter of this year and develop it into a "document agent platform" that develops, deploys and operates document-based agents for corporate business automation beyond simple document processing.