Skip to content
No accountand neither file is uploaded

They send the agreement back and say they only changed the payment terms. Visura lines up the two texts passage by passage and lists what else changed — in this browser, without either file leaving your machine.

sample data

employment-agreement.pdf

Version 4 of 4·94 sections·sample data

Sections 00–93, in reading order

10 sections
68reused
+ 26new
94sections
72%carried forward
94 sections·normalized text·SHA-256

Want to keep the versions too?

With an account, upload each revision under the same file name: a section whose normalized text matches one Visura already summarized in an earlier version keeps that summary, and only the rest are summarized.

Fig. 1 — Illustration with sample data: a 94-section document going from version 3 to version 4. 68 of the 94 sections still match once spacing, line breaks, character variants such as curly quotes, dashes and ligatures, running headers and footers, repeated short labels, page numbers and other lines of bare digits are set aside; the 26 that changed are the ones marked. Visura does not draw this map on any page; a version's processing insights page lists how many sections were reused and how many were new, and leaves them out of its overview when the file was a scan or was only partly read.

What changed, on the record

When a version finishes, Visura compares it with the version before and, for the first 40 changed passages, records each change it can back with a quote found in the document: the section it touched, the passage before and after, and how much it matters. The app stores these records but does not show them on any page.

4 changes · v3 → v4

94 chunks
  • clarification

    Definition of confidential information expanded

    0.88
  • risk_added

    Non-compete extended to 24 months

    0.84
  • scope_change

    Duties expanded to include team management

    0.79
  • modified

    Termination notice extended from 30 to 60 days

    0.91

employment-agreement.pdf · v3 → v4 · sample

Ten kinds of change

addedremovedmodifiedpolicy_shiftrisk_addedrisk_removedassumption_addedassumption_removedclarificationscope_change

Four changes recorded between version 3 and version 4 of the sample document, pinned to the sections that changed.

Only what changed gets summarized again

The text is normalized, grouped into sections of sentences, and each section is hashed with SHA-256. Upload a revision under the same file name and sections whose hash matches an earlier version keep their summaries; only the new ones are summarized. Upload it again under the same name with text that matches the latest version exactly and no new version is made.

Section sizea run of sentences, ended by a sentence's own hash or once it reaches about 1,200 characters
HashSHA-256 · 64 hex
Carried forwardsections whose text matches an earlier version after normalizing (spacing, quote marks, dashes, hyphenation) and dropping page numbers and repeated short lines such as headers

Version lineage · 94 chunks

v3 a91f3c0783 reused · 8 new
v4 fce188a268 reused · 26 new
68reused+ 26new
= 94chunks

carried forward 72%

sections 54–73 · sample

Version 3 to version 4 of the sample document. Sections 41–60 and 83–88 changed; the other 68 kept their summaries.

Upload it, and see what ran

Upload a new version under the same file name and only sections without an earlier summary are summarized. Each run's events are recorded on its version.

Text is read in your browser first — up to 50 pages of text per PDF; fully scanned PDFs are OCR'd in English (up to 10 pages). Word (.docx), Excel, and PowerPoint (.pptx) files are read directly.

employment-agreement.pdf

v4 · 94 sections · sample

  1. Reading text

    In your browser, before the file uploads.

  2. Uploading

    The original file is stored.

  3. Checking for a repeat

    Same file name and the same text as the latest version: its summary comes back.

  4. Splitting into sections

    Normalized, split mostly at sentence ends, each section hashed with SHA-256.

  5. Matching earlier sections

    A section that matches a summarized one from an earlier version keeps its summary.

  6. Checking the daily limit

    Only new sections count. Over the limit, the run stops before a version exists.

  7. Summarizing new sections

    Sections with no earlier summary are summarized in small batches.

  8. Merging the summary

    Once every section has one, they merge into the summary; the full text is kept for chat.

run complete · totals recorded on the version

Events recorded for this version

  • upload_started
  • chunking_started
  • chunking_completed
  • hash_diff_started
  • reuse_calculated
  • llm_processing_started
  • embeddings_started
  • indexing_started
  • version_completed
  • version_failed

version_failed exists; it did not happen in this run.

The steps of a finished run, and the events the version records along the way. Sections, insights, the graph and search embeddings are generated afterwards, in the background.

What happened to each section

Every section is kept from an earlier version, summarized in this one, or still waiting for its summary. An upload that would pass the daily limit stops before any of that.

reused

#0D786C / #37D2BE

cached from a previous version

the hash matched and the summary was kept

new

#A65808 / #F2A93B

processed this version

no earlier summary matched its hash

in flight

#5B7690 / #7E9BBB

queued or processing right now

its summary has not been written yet

blocked

#BF3B2F

this upload would pass the daily processing limit

the run stopped before a version was created

“This document would take you past today's processing limit. Try again tomorrow or upload a smaller file.”

Checked before a version is created. Only new sections count toward the default limit of 2,000,000 estimated tokens per account per day.

The legend, and a state the readout above never shows: a run the daily limit stopped.

Features

Read it, share it, search it


Five ways into one document

The summary, an outline of its sections, what the run reused, what the model found, and a graph of its topics and entities. The outline, the findings and the graph are drawn from the start of the document.


Search across a workspace

Search runs across the documents shared into the workspace and returns the passages closest to your query.


Comparing is free. Keeping more is what you pay for.

Compare two versions as often as you like — it runs in your browser, so it costs nothing and neither file is uploaded. Free keeps 5 documents. Starter keeps 10 and Pro keeps 1,000; each is a one-time payment. Each new version of a document counts as one of them. Every account can download 2 summaries, on any plan.

Free

Compare and keep 5

Comparisonsunlimited
Documents kept5
Payment—

Pro

Keep up to 1,000

INR₹1,770
EUR€17
USD$20

Formats and limits

FormatsPDF, DOCX, TXT, MD, XLSX, XLS, PPTX
Max size32 MB per file · very long documents can fail to process
.doc, .pptcan't be read; convert to .docx or .pptx
PDF extractionyour browser reads up to 50 pages of text · fully scanned PDFs are OCR'd in English, up to 10 pages

Visura Chat

Ask a
document
what it says

When does the probation period end?

Visura compares two versions of a document and shows what changed, word by word, and, when both were read in full without OCR, how much stayed the same once spacing, line wrapping, typographic styles, page numbers, running headers and footers, and short repeated lines are set aside. Keep documents here and each new version is split and hashed the same way, so unchanged sections keep their summaries. Read more

For questions, billing or anything else, get in touch.

Built to read

Visura reads PDF, fully scanned PDFs (first 10 pages, English OCR), Word .docx, Excel .xlsx and .xls, PowerPoint .pptx, plain text, Markdown.
Visura — see what changed between two versions of a document