How it works

Not an AI wrapper.

A reasoning engine using mathematics to answer your patent needs.

Semantyr is not a chat window with a patent pasted in. It measures your patent data, so every answer is reliable and repeatable.

Request Demo Where your data goes

Analytics powered by a reasoning engine,
explained by AI.

Mathematical models do the analysis. AI translates and explains. Humans interpret and finalize.

The order matters

From patents to actionable insights

Every Semantyr workflow runs in this order.

Your documentsPatents, a product document, an Office Action
MapDocuments read, sorted and placed
PlanThe work set up: groups, features, rejections
AnalyzeCounts, scores, rules and checks
Use of AI (LLM)Explains the results in plain English
ReportEvery statement traces back
The analyst reviews and decides at every step

The language model never decides a score, a group or a number.

Some workflows also use AI earlier, for example to read an Office Action or to help pick product features.

Four steps, the same way everywhere

Learn them once and every workflow makes sense

1

Map

Your documents are read, sorted and placed, so the engine knows what each one is.

Your partConfirm or correct.
2

Plan

The work is set up before anything is scored: technology groups, product features, or the rejections in an Office Action.

Your partMerge, split, add, remove or refine.
3

Analyze

The reasoning engine counts, scores, ranks and checks. The same input gives the same answer.

Your partRead it, question it and dive deeper.
4

Report

The findings are explained in plain English by the AI model (LLM), with every statement traced to its source.

Your partDownload any time and share.
Not a wrapper

A chat window with a patent pasted in is a different thing

A chat window
Semantyr
Who decides
A language model predicting likely words.
A mathematical model counting, scoring and applying rules.
Ask twice
The answer can change.
The measured parts give the same answer from the same input.
Where it came from
It cannot say which patent, or why.
Every statement has a place in the structure and traces back to its patents and passages.
The language model's job
Everything.
Reading and writing, inside a fixed structure.
In every workflow

Reasoning engine and AI model (LLM)

The same split, applied to each job.

Patent Landscape
Reasoning engineTechnology groups, zones and bands come from the mathematics.
AI model (LLM)It turns them into findings your team can read in a meeting.
Portfolio Analysis
Reasoning engineForms the shared map and the mathematical associations between the two portfolios.
AI model (LLM)The SWOT, IP Edge and Executive Summary are derived from those associations and explained in English.
Risk Analysis
Reasoning engineWhich patents rank highest is decided by the mathematics.
AI model (LLM)It writes the reasons, the design-around options and the summary, using only the patent data it was given.
Prosecution Analysis
Reasoning engineExact checks run first: the Office Action broken into rejections, claims and limitations, and every claim checked.
AI model (LLM)It then judges one narrow question at a time, such as what the examiner targeted, inside the structure the engine built. The attorney decides.
What you can rely on

Numbers you can defend

~10xless analyst time
Same input,
same answerso you can compare this year with next
Every number
traces backto the patents and passages behind it

Reliable, repeatable, human-controlled analysis. Your data is never used to train any model.

Why it is hard to copy

Many tools can call an AI model. What is hard to copy is the orchestration: which techniques, in what order, proven on real patent data.

Questions we hear

Is Semantyr a wrapper around ChatGPT?+
No. A mathematical model does the grouping, counting, scoring and checking. A language model only reads and writes, inside a fixed structure. There is no open prompting.
What does the language model actually do?+
It turns measured results into plain sentences. It never decides a score, a group or a number. Where a judgment is needed, as in Prosecution Analysis, it judges one narrow question at a time, on facts already pulled out.
Can I check how an answer was reached?+
Yes. Every statement has a place in the structure, and every number lists the patents and passages behind it.
Will I get the same answer twice?+
The measured parts, yes. In Patent Landscape the same patent set returns the same groups on every run. In Risk Analysis the scoring is repeatable once the features are set. The features themselves are picked with the help of a language model, so they can vary slightly between uploads, and you review them before the ranking is written.
Does Semantyr replace the analyst or the attorney?+
No. It does the measuring so the analyst can spend time on judgment. The analyst or attorney stays in control and makes the final call.
Is my data used to train models?+
No. Customer data is never used to train any model, and workspaces are isolated from each other.

See the method on your own patents

Bring a patent set. We will show you what is measured, what is written, and where each number came from.

Request Demo