Patent Office Actions
For IT and security review

What stays on the computer, and what can leave it

A Patent Center file history for an application that has not published is confidential, and reading its Office Actions and drafting the reply is what Patent Practitioner Tools is used for here. This page is what a firm's IT or security reviewer needs before a practitioner opens one in it, stated for the current release, on one page that prints cleanly.

At a glance

  • Where documents are processed: on this computer, by a model on it or on the firm's own server, or by the provider of a key you choose (the papers).
  • What reaches us: none of a matter's documents. The app sends anonymous usage reports, which you can turn off (what they hold).
  • Training: no matter data reaches us to train on; a provider key brings that provider's terms (providers).
  • Retention and deletion: a matter stays in its own folder until you delete it, and the trash is emptied after the days you set (where it is kept).
  • Encryption at rest: the disk's own, where it is turned on; the program adds none (where it is kept).
  • SOC 2: no service of ours holds a matter's documents, so there is no report of ours to ask for; a provider you choose answers for its own (providers).
  • What IT can enforce: five settings, among them no remote providers and offline mode (the settings).

What happens to the papers you add

  • You supply them, or ask for them. The program never signs in to USPTO.gov or Patent Center. You add the papers to a matter in the browser interface: the Office Action and the others as their own files, a Patent Center download as one PDF or a zip of its papers, or, with a USPTO Open Data Portal key of your own, a published application's file history, which the program gets when you press Get the file history (sends only the number) on the Papers step. An unpublished application is not served by the Portal, so you download it from Patent Center yourself. What it reads from the USPTO is that file history and a cited reference, each with your key (below).
  • Sorting and text extraction happen on this computer, with no model and no network. Each paper's kind and date are read from its own words, and a whole download is sorted by its own bookmarks into Office Actions, responses, the allowance and the rest. Text comes from a Word file or text as it is, from a PDF's text layer or, for scanned pages, from the OCR engine tesseract installed on the same computer. Page images are passed to it in memory, never written to disk, and never sent to a model.
  • One step uses a model: asking what the examiner said. Each Office Action and Notice of Allowance is then read by the model you selected. With a model on this computer, their text is processed there and not sent to us or to an AI provider. With a remote provider, the text of those papers goes to that provider, which the interface names before anything is sent.
  • A draft of the reply uses the same model, and only when you ask for one. For each item of the action it sends that model the item as the examiner states it, the pending claims, the examiner's citations, and the passages of the references you have and of the specification. The outline of the action, the claim listing, the examiner's citations and the checks over your draft use no model and no network.
  • The papers you add are kept with the matter. An Office Action or other paper of the record added as a PDF, and a Patent Center file history, is kept as a PDF in the matter's own folder beside its text, so the record can link each paper and a quotation of the action can open at its page. The specification and an attached reference are kept the same way when added as PDFs, and so is a granted patent's printed copy when it is fetched by number and a cited foreign reference's copy when it is read from the file history, so their quotations open at their pages too. A paper is kept only once it is taken: a refused upload keeps nothing. A template, and anything added as a Word file or text, keep only their text. The PDF opens through the program's own server, in your browser on the same computer; no network is involved.
  • What is kept. The text of the papers it reads, the PDFs above, what it found in them, a reply of yours added as a template (kept as the matter's form), the instructions you keep for every draft, your notes on the reply (never written into the exported reply), the draft reply and proposed claims for each action, the text of any reference fetched for the matter, and the reference papers of a related application's file history if you attach one (its other papers are dropped before anything is stored) are saved inside the matter, readable only by your account, and go to the trash and its purge with the matter.
  • What it reports, and what it does not. Each rejection, its claims, its ground and references, and the examiner's words, with every quotation checked against the paper it came from. A draft comes as separate parts you adopt, edit or discard. It does not judge whether a rejection was right, what a paper means for the claims, or whether an argument will persuade.

Where a matter and its file history are kept

  • Everything lives in one folder, .continuation-drafter in your home directory. Apart from files you export yourself, nothing is written anywhere else.
  • Each matter is one file in its matters folder, holding the text of the file-history papers it read and what it found in them. Only your account can read it (files 0600 in a 0700 folder on macOS and Linux).
  • A deleted matter goes to the trash, file history and all, and the trash is emptied after 30 days by default, or the number of days you set in Settings.
  • The program adds no encryption of its own. Its files are protected by the account permissions above and by the disk's own encryption, such as FileVault on a Mac or BitLocker on Windows, where it is turned on.
  • Settings, and any key you save (a model provider's, SerpApi's for patent search, or a USPTO Open Data Portal key for references) sit in config.json in the same folder, readable only by your account on macOS and Linux (mode 0600). A key can come from the environment instead and never be written.

The complete list of network connections

Four connections can serve a reply, each made only when you ask or have turned it on: fetching a cited reference, the USPTO Open Data Portal with your own key (a cited reference, a file history, and a check for new papers), and two ways to search patents on your own account. Reading the papers once they are in makes none. This is everything the program can connect to in the current release, and the list is checked against its source on every change, so a new connection cannot ship without appearing here first.

Every network connection Patent Practitioner Tools can make
ConnectionWhenGoes to
Reading Office Actions, and drafting, through the model server you runEach step that uses a model Ollama on this computer unless you change it, or an OpenAI-compatible server you started. It stays on this computer unless you point it at another machine. Since 0.8.0 a model your Ollama server would send on to another machine, such as an Ollama cloud model, is refused, and the model list does not offer it.
Asking that server to download a modelOnly when you press Pull Your model server, which fetches the model from its own registry
Reading Office Actions, and drafting, through the provider you choseOnly if you set up a remote provider with your own key OpenRouter, OpenAI, Anthropic. Two small lookups go there too: a model's context window, and the list of models for the picker.
A check for updates, and the download of the new releaseWhen you press Check for updates or run update, and each time the app opens if you turn that on in Settings (it is off unless you do) GitHub's public releases page for this program. Nothing about you is in the request. A new version is installed only once its signature, its checksum and, on a Mac, Apple's signature verify.
License renewalOnly with a paid license, close to its expiry. No license is sold today, so this call does not happen The license server, with the key's ID, its token and the program's version, and nothing more
Anonymous usage reportingOnce you dismiss the welcome notice: while the app is open, and as it closes. Offline mode holds the reports on this computer until it is off; DO_NOT_TRACK=1 or ForceOfflineMode stops them Our own service at telemetry.obviouslynot.ai, with the reports described below and nothing from a matter
A check whether it is already runningOn launch 127.0.0.1, which is this computer
Fetching a cited reference by its number, to show its own words beside the examiner's citationOnly when you ask, or each time if you turn that on in Settings Google Patents, with the publication number in the address and nothing else, and for a granted patent its printed copy, from Google's patent images at the address that page gives
Reading from the USPTO Open Data Portal, with your key: a cited reference, a published application's file history, and a check for new papersOnly once you save your own Portal key in Settings. A US reference Google Patents does not have is read there when it is fetched as above; a cited foreign reference's copy in the application's own file history, only when you ask. A file history, only when you press Get the file history (sends only the number). The check for new papers, only if you turn on Check the USPTO for new papers in each matter each time the app opens in Settings, which is off unless you do The Portal at api.uspto.gov, with your key in a request header: for a US reference its number, in a search; for a file history, a foreign reference's copy or the check for new papers, the application's number, to list its papers. Nothing else is sent. The Portal answers a file request with a redirect to the USPTO's own file host, data.uspto.gov or data-documents.uspto.gov, which the program follows without the key; each paper of a file history is fetched that way and kept with the matter. It never signs in to USPTO.gov
Running a patent search you asked for, through SerpApiOnly with your own SerpApi key in Settings. The search is shown, with where it will go, before it is sent, unless you turn that off SerpApi, which runs it on Google Patents. The query is written from your request, so it describes the invention
A patent search in your own Google Cloud projectOnly once you set up a project, or sign in with gcloud on this computer, and confirmed the same way Google's BigQuery, over Google's public patent data, in your project. A search that would read more than a set limit of your query allowance is refused before it reads anything
Asking your own Homebrew to install OCR, on a Mac without it, to read scanned pagesOnly when you press Install, whose label shows the command it runs. On Linux and Windows the program shows the command and runs nothing Homebrew's own sources, through the Homebrew you installed, into your own account. The program does not fetch the engine itself

The program connects to nothing else, and sends no crash reports. Offline mode in Settings stops the update check, the license renewal, reference fetching from Google Patents and the Portal, a file history from the Portal, the check for new papers, patent searches and the OCR install, but not a remote provider you chose: to keep a file history's text on the computer, read it with a model on the computer, or have your IT department enforce that (below). It holds usage reports on this computer and sends them once it is off; it does not stop them.

Which certificate authorities are trusted. Since version 0.7.5, on macOS and Windows, SSL_CERT_FILE and SSL_CERT_DIR, when set, decide which certificate authorities are trusted for connections to model providers and patent sources, instead of the system's store. Most machines set neither.

Anonymous usage reports

Since version 0.14.0 the program reports how it is used, so that it can be improved from real use. Nothing is sent until you dismiss the welcome notice the first time the app opens, which says so and links to the privacy policy.

  • What a report holds: which steps ran, how long they took as a range, and whether they finished; the hour of use and a time-zone region; the operating system, the processor family and a memory range; which provider and model ran a step, from a fixed list, with its tokens and cost as ranges; your answer to the notice's optional question of how you heard about the app; and a random install number our service issues, tied to no person, license or account.
  • What it never holds: a paper, a specification, a claim, a file name, a matter, an application number, model output or an error message. Every value a report carries comes from a list fixed in the program, so text from your work has no field to go in.
  • Where it goes: our own service at telemetry.obviouslynot.ai, which stores no IP address. Our hosting provider's request logs keep the connecting address for up to 90 days, as the privacy policy says.
  • Turning it off: set DO_NOT_TRACK=1 in the environment of the account that runs the program, and it reports nothing and deletes its install number and any reports waiting. The ForceOfflineMode setting below does the same. Offline mode in Settings only holds reports until it is off.
  • Seeing it: reports wait in the telemetry folder inside .continuation-drafter until they are sent, and continuation-drafter telemetry show prints them.

Settings your IT department can enforce

Five settings can be enforced rather than trusted. The program reads them from the operating system's managed-settings store, which takes an administrator to change, and a practitioner cannot override them from the program, the environment or its settings file.

  • Windows (Group Policy or Intune): values under HKEY_LOCAL_MACHINE\SOFTWARE\Policies\ObviouslyNot\PatentPractitionerTools, the switches as REG_DWORD 0 or 1 and the server as a REG_SZ.
  • macOS (Jamf or any MDM): a configuration profile for the domain ai.obviouslynot.patent-practitioner-tools, installed by the system as /Library/Managed Preferences/ai.obviouslynot.patent-practitioner-tools.plist.
What each enforced setting does
KeyValueEffect
AllowRemoteProviders0, or false Office Actions are read, and replies drafted, only by a model on this computer or the server ModelServerURL names. Remote providers show as off, a model server on another machine is refused, and so is a patent search, because its query describes the invention. Fetching a reference, from Google Patents or the Portal, is not affected.
DisableUpdateCheck1, or true No contact with the releases page, from the button or at launch; IT deploys each version.
ForceOfflineMode1, or true Offline mode, on, and not something the practitioner can turn off. Unlike offline mode in Settings, it also stops usage reporting and deletes reports waiting to be sent (since 0.14.0).
ModelServerURLan http or https address The firm's own model server. It becomes the model host, and with remote providers off it is the only other machine a file history's text can reach.
DisablePatentLookups1, or true No cited reference is fetched, from Google Patents or the USPTO Open Data Portal, no file history is got from the Portal, no matter is checked for new papers, and no patent search is run. New in 0.7.0; the file history and the check since 0.12.0.

The practitioner sees in Settings which settings the organization has set. A value that cannot be read as written is applied at its strictest and shown there, and a settings store that exists but cannot be read applies every restriction. These locations apply from version 0.6.2. Linux has no such store: use the environment variables for the account that runs it. To stop usage reports without offline mode, set DO_NOT_TRACK=1 in the environment of the account that runs it.

If you read with a remote provider

No matter data reaches us, so none is ours to train a model on, and no service of ours holds a matter's documents, so there is no SOC 2 report of ours to ask for. A provider you choose has its own terms on training, and any audit report is that provider's to give.

Whether a provider is acceptable for a particular file history is the practitioner's judgment, and turns on facts we do not have, such as the engagement terms and the provider's own contract. What each says it keeps:

What each route keeps, as the providers document it
How it runsWhat is kept
A model on your own computerDocuments and drafts stay on the computer, as long as the model server runs on it too. The specification goes only to that server. Since 0.8.0 a model your Ollama server would send on to another machine, such as an Ollama cloud model, is refused.
Claude through OpenRouterAnthropic's sources differ. Its API documentation says prompts and outputs are not retained by default except on its Fable and Mythos models, which require 30 days; its commercial privacy article says API inputs and outputs are deleted within 30 days, and kept up to two years if flagged; OpenRouter's provider listing declares 30 days. Plan on up to 30 days. The app also asks for a prompt cache, which holds the start of each prompt, mostly the specification, in memory for up to an hour after its last use.
Claude with your own Anthropic keyNo prompt cache: the Anthropic endpoint the app uses does not support one. Otherwise as above: Anthropic's documentation and its privacy article differ, and the longer reading is up to 30 days, more if a request is flagged, unless the account has a zero data retention agreement.
OpenAI, with your own OpenAI keyThe app asks for the shortest prompt cache OpenAI offers, which depends on the model: minutes on older models, up to 24 hours on gpt-5.5, and 30 minutes after last use on gpt-5.6 and later. OpenAI says cached data may be kept encrypted in GPU-local storage.
OpenAI models through OpenRouterOpenRouter does not pass the cache setting on, so OpenAI's own default for the model applies.
Other models through OpenRouterThe model runs on whichever host OpenRouter chooses, and that host's policy decides what it keeps. One question can reach more than one host: in a test, four calls about one specification went to four different companies.

What gets installed

  • A single program, as a macOS app or a command-line binary for macOS, Windows or Linux. There is no installer to run, it needs no administrator rights, and it adds nothing to startup.
  • Nothing runs in the background. While open, it serves its interface to your browser on 127.0.0.1, which only this computer can reach, and it stops when you quit it. Since version 0.7.3 that page loads nothing from any other site.
  • Signed by its publisher. The macOS app and binaries carry an Apple Developer ID signature and Apple's notarization, and the Windows binary is Authenticode-signed, naming Geeks in the Woods as the verified publisher. In a new release's first days Windows may still call the file not commonly downloaded and preselect Delete, because it scores each release's reputation apart from its signature; check the publisher line. The download page shows how on both systems.

Checking a download yourself

  • macOS: the Apple Developer ID signature and notarization on the app and binaries.
  • Windows: Authenticode on the binary, with Geeks in the Woods as publisher.
  • Every file: its line in checksums.txt, which is itself signed and is what the update check verifies before installing anything.
  • A software bill of materials published beside every binary.

About this website

This site counts visits with Cloudflare's web analytics: page views, page timings, and coarse facts such as country and browser. It sets no cookies, does not fingerprint your device and builds no profile across visits or sites. The site and the program are separate: no file history ever reaches this site, and the program's own usage reports are described above.

Questions from a review that this page does not answer: hello@obviouslynot.ai. More on confidentiality in the FAQ.