---
title: "PreciseMD"
description: "Turn documents into AI-ready data — set up the models, tune the parsing in Playground, schedule folder-based OCR tasks, then drive it from an agent, MCP or the CLI"
canonical_url: "https://docs.ardge.com/en/guide/ai-applications/precisemd"
---
# PreciseMD
> Turn documents into AI-ready data — set up the models, tune the parsing in Playground, schedule folder-based OCR tasks, then drive it from an agent, MCP or the CLI
## Overview
PreciseMD takes documents that an AI can't use directly and turns them into **AI-ready text**.
It reads images, PDFs and Office files, recognises the text, tables, columns and layout inside them, and writes out Markdown that keeps that structure.
Why does this step matter? Because a scanned PDF, a complex table or a multi-column document that parses badly leaves the RAG or AI agent downstream unable to find what matters. PreciseMD's job is to get the source documents into shape first — cleaner, searchable, usable by an AI.
**PDF / scan / Office → PreciseMD → AI-ready Markdown → RAG / agent / search**
Every parsing model runs in [AI Model Center](https://docs.ardge.com/en/guide/use-ai/ai-model-center) on the same ARX, so the documents never leave the device.
## Before you begin
Confirm the recommended OCR model and VLM model are running in AI Model Center.
## Step 1: Set up the model connection
Fill in the model IDs of the OCR model and the VLM model in PreciseMD's settings — see [Set up the model connection](https://docs.ardge.com/en/knowledge/precisemd/model-connection).
## Step 2: Try a real document in Playground first
Run 1–3 of your most representative documents through it and check whether:
- The text is complete
- Tables stay readable
- Multi-column layouts, charts and complex pages come through correctly
- The prompt needs adjusting
- You need vision correction or another advanced option
See [Tune prompts and compare results in Playground](https://docs.ardge.com/en/knowledge/precisemd/playground-prompt-tuning) for how.
Before you upload anything, check [what file formats and limits PreciseMD supports](https://docs.ardge.com/en/knowledge/precisemd/formats-and-limits). The caps differ between Playground and an OCR task.
## Step 3: Compare results from different settings
Use Playground's comparison to put two runs side by side and decide which output suits the AI downstream better. Settle on settings here before you automate anything.
## Step 4: Create an automatic OCR task
With the settings decided, create a folder-based OCR task:
**Source folder → OCR / parse → output folder → Markdown and related output**
Connect an SMB share, pick the input and output folders, and set the weekly schedule so new files get parsed on their own — see [Schedule an OCR task to process a folder](https://docs.ardge.com/en/knowledge/precisemd/schedule-ocr-tasks).
## Step 5: Check the run history
If the output count doesn't match what you expected, check the task history to find which file failed and why — see [Check task history and troubleshoot failed files](https://docs.ardge.com/en/knowledge/precisemd/ocr-task-history).
For what lands in the output folder, see [What files are produced in the OCR output folder?](https://docs.ardge.com/en/knowledge/precisemd/ocr-output-files)
## Step 6: Drive it from an agent, MCP or the CLI (advanced)
These are advanced integrations — most people can skip them at first.
- [Operate PreciseMD through the built-in agent](https://docs.ardge.com/en/knowledge/precisemd/built-in-agent) — describe what you want in plain language and the agent calls the matching features for you
- [Create a personal access token](https://docs.ardge.com/en/knowledge/precisemd/access-tokens) — the credential an AI agent or the CLI authenticates with, scoped by permission level and data source
- [Wire parsing into your own tools](https://docs.ardge.com/en/knowledge/precisemd/mcp-cli-integration) — hand the tools to an AI agent over MCP, or call parsing from scripts and schedules with the CLI
## When do you need vision correction?
If your documents hold complex tables, multi-column layouts, charts, inconsistent scan quality, or need layout understanding, consider turning vision correction on. It adds processing time but improves parsing quality on those documents — see [When should I enable vision correction?](https://docs.ardge.com/en/knowledge/precisemd/vision-correction)
## The question that actually matters
Don't just ask "did the OCR finish". Ask:
**Is the converted data clean enough to hand to a RAG or AI agent with confidence?**
## Good to know
- [What's the difference between an administrator and a member?](https://docs.ardge.com/en/knowledge/precisemd/admin-vs-member) — which sidebar item separates the two roles
## Sitemap
See the full [sitemap](https://docs.ardge.com/sitemap.md) for all pages.