Skip to main content

What types of documents are supported?

From clinical PDFs and Word files to audio recordings of patient encounters, BastionGPT accepts the formats healthcare teams actually work with every day, giving clinicians and staff a secure way to put existing files to work.

J
Written by Josh Spencer

AI Assistant

Description

Extension

Professional

Plus

Ultra

Plain Text File

.txt

Portable Document Format

.pdf

Microsoft Word 97–2003 Document

.doc

Microsoft Word Document (Office Open XML)

.docx

Word Macro-Enabled Document

.docm

Microsoft PowerPoint Presentation (Office Open XML)

.pptx

Microsoft Excel 97–2003 Workbook

.xls

Microsoft Excel 2007

.xlsx

Excel Macro-Enabled Workbook

.xlsm

Comma-Separated Values File

.csv

HyperText Markup Language File

.html

JPEG Image

.jpg

Portable Network Graphics Image

.png

Custom or Proprietary Formats

***

AI Scribe

Description

Extension

Professional

Plus

Ultra

MPEG Audio Layer III

.mp3

MPEG-4

.mp4

Ogg Vorbis Audio

.ogg

Waveform Audio

.wav

MPEG-4 Audio

.m4a

WebM Audio

.weba

Windows Media Audio

.wma

Free Lossless Audio Codec

.flac

Custom or Proprietary Formats

***

Document size and page limits

The tables above show which file formats you can upload. Here is how much each document can hold, and how to get great results with even the largest records.

How large can an uploaded document be?

  • Pages: up to about 1,000 pages per document, up from 500 pages earlier in 2026. This limit grows as AI models improve.

  • File size: up to 40 MB per attachment.

Capacity also varies by plan: Professional is designed for smaller everyday documents, while Professional Plus and above support the full document capacity.

See page limits explained in depth for the full details behind these numbers.

Why is there a page limit?

The limit is a deliberate quality safeguard. Beyond it, every AI model on the market today shows a "lost in the middle" effect: it understands the beginning and the end of a long document well, but details in the middle get fuzzy, and the AI can invent details, such as a medication that never appears in the chart. We test each model and set the cap at the upper end of what is clinically safe, then raise it as the technology proves itself.

How do I work with a document that is over the limit?

Two approaches work well today:

  • Split the document. Divide a 1,500-page packet into two parts, for example, and have BastionGPT review each part separately.

  • Summarize in passes. Upload about 200 pages at a time and build a running summary: "Here are the first 200 pages; give me a two-paragraph summary." Then: "Here are the next 200 pages; add anything new to the summary." This works especially well for long records where many pages repeat earlier visits.

If you need BastionGPT to write a long document rather than read one, see generating reports over 10 pages.

Can BastionGPT read scanned, faxed, or handwritten documents?

Yes. You can upload photos of paper forms, scanned charts, faxes, and documents with handwritten notes. Recognition is on par with a careful human reader, so if handwriting is hard for a person to make out, the AI may find it hard too. BastionGPT also reads structured plain-text content such as markdown, JSON, and XML.

How long are uploaded documents kept?

Uploaded documents stay available in your account permanently. They are securely erased only after your account has been idle for 30 days.

If your documents contain patient information, see uploading PHI safely.

Why can some other AI tools accept larger documents?

Tools that accept very large files usually read them with a technique called retrieval (RAG). Rather than reading the whole document, the AI selects roughly 20 half-page excerpts it predicts are most relevant and answers from those alone. That works for pinpoint questions like "When was the last visit?", but for a full-record summary the AI is blind to everything outside those excerpts. BastionGPT gives the AI your entire document, so within the page limit, nothing is skipped.

For a side-by-side look at the numbers, see word limits across AI platforms compared.

Did this answer your question?