Patent Continuation
Free, nothing to activate

Download Patent Practitioner Tools

Linux binaries for the command line, plus a Mac app you double-click and a signed Windows program. Every link is the current release.

New in 0.14.0: the app reports how it is used, anonymously and never with your papers or your clients' (what is sent, and how to turn it off). For continuation claims, since 0.13.0: ask the conversation where the papers discuss something, and each passage is shown in the paper's own words with its page; or ask for a digest of what each file of the matter states, each statement above the words it rests on.

Download for Linux, x86-64

continuation-drafter-linux-amd64

If you are on arm64, take that build from the full list at the foot of the page instead.

Release notes and checksums

Five steps from download to first claims

Installing takes under a minute and setting up drafting a few minutes; a local model adds a download of several gigabytes, under an hour on a fast connection.

Download

The button above takes the build it names; the links over it choose another computer.

Open the app

On Linux: make it executable with chmod +x continuation-drafter-linux-amd64, then run it with serve, and the interface opens in your browser.

The app: double-click the downloaded .dmg, drag Patent Practitioner Tools to your Applications folder, and double-click it there. macOS asks once whether you are sure you want to open something downloaded from the internet; the button is Open. Nothing else is needed: no Terminal, no permissions to change.

It has no Dock icon on purpose, because it runs no window of its own: the interface is the browser tab it opens. To stop it, press Stop in that tab.

Try the sample

The first screen asks what you would like to work on. Under continuation claims, press Try the sample: it opens a matter holding a short published-style application for a temperature-actuated valve, with its filed claims. Opening it needs no model.

Patent Practitioner Tools v0.10.1's start screen, opened for the first time, with a card for a response to an Office Action and a card for continuation claims, each with Try the sample.
  1. Try the sample for a reply opens the public Office Action of 18/541,216, already read.
  2. Try the sample for continuation claims opens the app's own short application for a temperature-actuated valve, with no model needed to open it.
  3. Every matter stays on this machine, between sessions.
Unedited screenshot, cropped, of Patent Practitioner Tools v0.10.1, captured 2026-10-20: the start screen before any matter exists. The numbers and notes are ours; the picture is the app.
Patent Practitioner Tools v0.10.1's continuation sample, a published-style application for a temperature-actuated valve, open at its Embodiments step before any model has run.
  1. The sample matter: its parent's 3 claims, run on this machine.
  2. The sample's application, already loaded.
  3. Discover embodiments is the first step that uses a model, on this machine or through your own key.
Unedited screenshot, cropped, of Patent Practitioner Tools v0.10.1, captured 2026-10-20: the continuation sample built into the app, a published-style valve application, before any model has run. The numbers and notes are ours; the picture is the app.

Set up drafting

Finding embodiments is the first step that uses a model, so this is where you choose where it runs.

On this computer

When the documents must not leave your computer: they are processed on it and not sent anywhere. It needs 16 GB of memory or more. On 16 GB the model is gemma4:12b, holding about 12 GB, flat while it works. Install Ollama or another local server and download the model once, several gigabytes, under an hour on a fast connection. Choose a model for your machine.

With your own provider key

With your own key for OpenRouter, OpenAI or Anthropic: no download, set up in a few minutes. The median draft took 78 seconds across 131 timed runs on remote models. The provider receives the specification, so this suits published material, like the sample. Using your own provider key.

For a real matter that is a decision you make per matter. For the sample it is not a decision at all: it is the app's own published-style application, so nothing of yours is involved, and a provider key is the quickest way to see it work. Then, on the sample's Embodiments step, press Discover embodiments, and Draft first claims on the one worth pursuing.

Your own papers

Press Start continuation claims, and under Papers add the parent's specification, its claims as filed and, if you have it, the Patent Center file history. Each slot takes a PDF, a Word file or text. Scanned pages are read on your computer, by an OCR engine (tesseract) installed beside the app.

The file history can come two other ways. Import from Patent Center, beside New matter, takes the papers you downloaded for any number of applications, as a zip of separate files or one PDF, shows each application's papers with its matter, and adds nothing until you choose. With your own USPTO Open Data Portal key, Get the file history (sends only the number) on Papers gets a published application's file history from the USPTO, and its application as filed becomes the specification when the matter has none. Give a matter its client, and the matter list groups your matters by client.

Try it on a public patent first

Before any client material goes near it, run it on a granted application's public record: US 12,299,058, application 18/541,216.

  1. Download its Patent Center file history (a 13 MB PDF, exactly as Patent Center serves it), its specification and its allowed claims.
  2. In the app, press Start continuation claims, and under Papers add the specification and the allowed claims in their two slots.
  3. Add the PDF as The prosecution history. It picks out the Office Actions, the responses and the allowance and reads them on your computer, several scanned pages at a time; the download also carries the application as filed, used as the specification when the matter has none. The Prosecution record panel then shows each paper it read and where each limitation of the claims entered the record. Claims numbered 100 and up are read too, and a limitation whose range or value was narrowed is placed in the round that recites its numbers.
  4. On Embodiments, press Discover embodiments, read how each differs from the claims and the passage it cites, and press Draft first claims on the one worth pursuing. Run discovery again lists what a second pass finds beside the ones you kept; when it finds no further supported direction, it says so and offers another reading, and the ones you kept stay.
  5. On Claims, read the critique's findings under each claim; Show in claim shows the words a finding is about. Edit the first claims or revise them with an instruction of your own; revised claims are checked again, and a revision whose check did not finish is marked unchecked. Then build the full set with Expand, keeping these claims first. Save the set as Text, Word, or PDF; Next steps lists what the matter has left, a note can go on one claim or the whole set, and History of these claims records each change.

    On both lists, j and k move between items; a keeps and d dismisses an embodiment, and r marks a claim's next finding reviewed.

The same application, run through these steps, is pictured on the sample page.

Patent Practitioner Tools' Embodiments step on 18/541,216, run locally with gemma4:31b, listing the embodiments it found in the specification before anything is drafted.
  1. The matter's line: the parent's issue date from its Issue Notification, against today, the claims in the parent and the embodiments you kept.
  2. Next steps lists what the matter has left, each linking to the step where it is done.
  3. Continuation targets holds the embodiments you kept, each one's wording yours to edit.
  4. Each candidate embodiment includes a citation to the specification passage identified as supporting it, for you to verify, and a note on how it differs from the parent's claims.
  5. How the embodiment relates to the parent's prosecution record: the allowed claims' limitations it builds on and the examiner's reasons for allowance, each checked against the record, or that nothing in the record was matched to it.
Unedited screenshot of Patent Practitioner Tools v0.10.1, captured 2026-10-20: application 18/541,216, public record, run locally with gemma4:31b. The numbers and notes are ours; the picture is the app.

Before you install

What does it cost?

It is free to download and use, with no account and nothing to activate. With a local model there is no cost per use; with a provider key you pay that provider directly, with no markup from us. More on cost.

What setup does it need?

To open it, none. To read scanned pages, an OCR engine (tesseract) on the same computer. To read what an Office Action states, and to draft, a model: one on your computer, through Ollama or another server that speaks the OpenAI API, such as LM Studio or llama.cpp; or your own key at OpenRouter, OpenAI or Anthropic. Try the sample opens with no model, because its application is already loaded; a model is needed from the first step, finding embodiments. More on setup.

What stays on my computer?

With a model running on your own computer, your application documents, the text read from them, and your drafts are processed on that computer and are not sent to us or to an AI provider. A model your Ollama server would send on to another machine, such as an Ollama cloud model, is refused. Other network activity, such as an update check you ask for, is listed on /security. More on where the papers go.

What reaches a provider if I use my own key?

The text of the papers the model reads and, for a draft, what that draft is built from, sent to the provider you chose, under your key. More on what a provider receives.

Can I try it without client documents?

Yes. Try the sample opens a sample matter for either task: the public Office Action of application 18/541,216 for a reply, or a short published-style application for continuation claims. The public papers of 18/541,216 (its Patent Center file history, specification and allowed claims) can also be downloaded from /download and added yourself. More on the sample.

Does it work with Word?

Yes, both ways. Every paper can be added as a Word file, as well as a PDF or text, and struck-through text in one is read as deleted. Claims save as Text, Word or PDF wherever they are shown, and a reply downloads as Word, from Reply in Word or Report of the action in Word; notes on claims or a reply are never written into a download. More on the Word file.

Need help getting started? Write to [email protected]; it goes to a person.

What is checked, and how you can check it

Every claim here is one you can confirm yourself. None of it asks you to take our word.

Reviewing it for a firm? The IT and security page lists every network connection it makes, where it keeps its files and how each download is signed.

Run the AI on your own machine

With a model running on your own computer, your application documents, the text read from them, and your drafts are processed on that computer and are not sent to us or to an AI provider. Point the app at a cloud provider instead and it names the company that will receive it, before it sends anything. That is a decision you make per matter, and it is stated at the moment you make it.

Signed, and you can see by whom

Windows shows Geeks in the Woods as the verified publisher; the macOS app is signed with a Developer ID and notarized by Apple, which scans every build. To confirm it yourself on Windows: right-click the file, then Properties, then Digital Signatures.

Every file publishes its fingerprint and its contents

A SHA-256 for each download is published with the release in checksums.txt. It is what establishes that the file you have is the file that was built, which a signature alone does not tell you.

Each binary also ships a bill of materials listing every library inside it, with versions, down to the version of the Go runtime it was built with. That is what lets anyone answer whether a particular build contains a library named in a future advisory, without rebuilding it.

The checksums file is itself signed, and the app uses that signature to update itself: press Check for updates in Settings and a newer version installs only after the signature, the checksum and, on a Mac, Apple's signature all verify. It checks only when you ask.

Known vulnerabilities stop a release

Before any release is published, the code and every library in it are checked against published security advisories. One that is reachable from this code fails the release rather than shipping in it.

These are claims about the software, not about the quality of what it drafts.

From the command line 
chmod +x continuation-drafter-darwin-arm64 && \
  mv continuation-drafter-darwin-arm64 continuation-drafter

That is two things: make it executable, and give it a shorter name to type.

Both the app and the raw binaries are signed with a Developer ID certificate and notarized by Apple, so nothing has to be done to either file after downloading it and there is no quarantine flag to clear. If you find an older page telling you to run xattr, it predates v0.1.4.

Notarization means Apple scanned this developer's build; it does not tell you the file in front of you is that build. The check that establishes that is its SHA-256, listed with the release. Linux and Windows need neither step.

The Windows binary is signed too, with a certificate issued to Geeks in the Woods, which is the name Windows shows as the verified publisher. Windows can still warn on a fresh release, saying the file isn't commonly downloaded and preselecting Delete: that is scored separately from the signature and clears as people take each build. Read the publisher line rather than the warning text, because a build from this project names Geeks in the Woods, and the SHA-256 above is what establishes the file is that build.

Draft against the sample built into the program, with no file of your own:

./continuation-drafter draft --demo

It drafts from the same published-style valve application the app's Try the sample opens, the walkthrough the course teaches, and never counts against any allowance. A full run is printed unedited here. Name the model on the command itself if you have not saved one: DRAFT_MODEL=<model> ./continuation-drafter draft --demo

On your own work the command is draft --document your-application.txt, which reads a whole application and works out the specification and the claims itself. If it does not run, ./continuation-drafter models reports what the program resolved and what your machine can host.

The same binary serves the app's interface, with ./continuation-drafter serve. It binds to 127.0.0.1, which is your own machine and not a network address anyone else can reach, and opens your browser.

Every build, and its checksum

On a Mac, take the double-clickable app rather than a raw binary: continuation-drafter-macos.dmg. The two bare macOS builds below are for a terminal, a script or CI.

Not sure which Mac you have? Apple menu, then About This Mac. A chip named Apple M1, M2, M3 or later means Apple silicon; anything named Intel means Intel.

Release notes and checksums

One question, if you have a minute

We do not know how many practitioners will run a model on their own hardware versus pointing the app at their own provider key, and that assumption sits underneath most of how this is built. If you downloaded it, telling us which you chose, or that you chose neither and why, is genuinely useful.

[email protected]. It goes to a person. There is no list to join and nothing follows.