Release notes

For what's in progress and planned next, see the Roadmap.

August 19, 2026

Improvements

  • More Reliable Knowledge Base Processing — Files whose extracted content or images previously failed to display now process reliably. Knowledge Sources added before this release need to be deleted and re-added once
  • Fewer Interrupted Sessions — Sessions that can no longer be refreshed now prompt you to sign in again and return you to the page you were on, instead of stopping on an error page
  • More Reliable Text Extraction — Documents containing repeated page headers, footers, footnotes, or page numbers now finish processing without losing extracted content
  • Consistent Project Document Handling — Project documents now use one consistent path for upload, processing, download, and deletion
  • Fewer Missing Abstracts — Fixed an issue where some PCT applications were missing their abstract after document processing

August 17, 2026

New Features

  • Knowledge Base — Add reference PDFs to a Project, review extracted content alongside the original PDF, and copy Markdown or download a portable bundle for reuse. Learn more
  • Complete Content Reads by Default — completion_required now defaults to true for patent and application content, so connected assistants continue until all requested text is read unless a bounded read is requested
  • Patent Markdown Export — Copy or download complete patent text as Markdown directly from the patent workspace
  • Knowledge Base MCP Tools — Claude and ChatGPT can now list Project sources, search completed reference PDFs, read source-attributed passages, and retrieve extracted images. View the tools

Improvements

  • One Result Per Patent Family — Prior-art searches no longer show multiple publications from the same patent family
  • Open Original Project PDFs — Examiners can open and download the original PDF for Project documents after processing
  • Reliable Large-Document Processing — Large Project documents continue processing without document-status checks timing out

August 4, 2026

New Features

  • Use Application Numbers With Project Tools — In the assistant, use a Singapore application number to identify a Project. If another Project already uses that application number, Goveda asks before creating a duplicate

Improvements

  • Create Projects Before Register Data Is Available — A valid Singapore application number can create a Project even when the official register has no matching record. Upload either PF1 or PF37 to supply its bibliographic information
  • Connect the Assistant Once — After connecting Goveda MCP, use your organization's tools immediately without signing in to the workstation first

July 31, 2026

Improvements

  • Accurate Keyword Highlighting — The patent viewer's highlighter now matches multi-word phrases, and each keyword reports its own hit count instead of a shared total

July 30, 2026

New Features

  • Single Sign-On Documentation — Setting up single sign-on for your organization is now documented end to end: connecting your identity provider, choosing whether people join automatically on first sign-in or by invitation, testing the connection, and requiring it for everyone. See Single sign-on
  • Contact — A Contact entry in the site navigation, so you can reach us from any page rather than only from the sign-in screen

July 29, 2026

New Features

  • Organizations — Goveda Pro is now organized around your workspace. Members are invited by an administrator, work together on the organization's cases, and can belong to more than one organization and switch between them. Administrators manage members from a Members page
  • Sign In With Your Work Email — Enter your work email and Goveda Pro recognises which organization you belong to and takes you to the right sign-in method. There is no organization to pick and no separate workspace address to remember
  • Enterprise Single Sign-On — Organizations can connect their own identity provider and let members sign in with their existing company account. An administrator sets this up from Settings without contacting us and tests it end to end before turning it on. Members who join this way can work on any of the organization's cases, not only the ones they created
  • Invitation-Only Access — Access to an organization is by invitation. Signing in with a personal Google account is no longer offered
  • Open Results Straight From the Assistant — Open a project, its documents, its analysis results, or a completed prior-art search directly in the workstation from a link in the assistant
  • Narrow Prior-Art Results You Already Have — Filter an existing prior-art result set, either with the same keyword expressions used on patent pages or with a plain-language relevance query, and revisit the saved set later

Improvements

  • Rebuilt Unity of Invention Page — The verdict and the independent analyses behind it now sit side by side, each analysis can be read in full, and reasoning that was previously produced but never shown is now visible. Results opened from the assistant link through to the full analysis, and a verdict flagged for review no longer reads as though the decision were already made
  • Faster Prior-Art Searches — A search you start returns completed results in a single request while automatic recommendations continue in the background. Filtering large result sets completes reliably, and repeated filtering returns faster while preserving ordering and pagination
  • Citation Links — Family literature citations include a direct link to the cited work, or a focused search when no direct identifier exists
  • Readable Account Names — Accounts show a proper name — the one your company's identity provider publishes where available — instead of a full email address
  • Clearer Sign-In Problems — When sign-in cannot be completed, the page explains what happened and offers a way out instead of leaving you stuck. Sessions no longer drop shortly after signing in, and opening several pages at once no longer signs you out
  • Assistant Document Uploads — Uploading documents through the assistant's document panel no longer fails, and the document-type list is complete for every file
  • Shorter Unity Reports — Unity of Invention reports end after the substantive conclusion and omit obsolete procedural guidance

July 20, 2026

New Features

  • Public Patent Viewer — Anyone can look up a patent by publication or application number and read its record without signing in: core bibliographic data, legal events, citations, family, drawings, and the PDF. Records open in a consistent searchable workstation with split views and improved family, legal, and citation navigation
  • Keyword Highlighter — Highlight several keywords at once in different colours while reading a patent, each with its own hit count and its own lane on a position map you can use to jump between matches. Keywords match whole words by default and support truncation, proximity within a block, and Chinese, Japanese, and Korean text
  • Three-Level Unity Verdict — Lack of Unity now reports pass, needs-review, or objection instead of a single yes/no. Borderline cases surface the key question and a suggested next step rather than being forced to one side, in the workstation, in the assistant, and in the API response
  • In-Product Feedback — When you hit an unsupported capability, an inaccurate result, or an unrecoverable failure, you are offered a ready-to-submit feedback draft instead of having to report it separately

Improvements

  • Drawings on Their Own Page — Extracted drawings open on a dedicated page with the selected drawing preserved, and leaving fullscreen no longer leaves an invisible layer blocking the page
  • Reliable Structured Content — Patent pages display tables, formulas, chemical structures, and embedded images reliably across single and dual views
  • Cleaner Paragraph References — Descriptions no longer show placeholder paragraph markers when the source document has no citable paragraph number
  • Patent Links Open in the Workstation — Publication numbers in results, exports, and Lack of Unity conclusions now open the patent record in Goveda Pro rather than the public website
  • More Dependable Unity Analysis — Lack of Unity analysis no longer fails on certain applications, and invention groups are listed starting from the lowest claim number
  • Consistent API Key Connections — Connecting with an API key now works the same way across browsers, including where an earlier sign-in to the same account was still active

July 11, 2026

New Features

  • Create a Project from an Application Number — Enter a Singapore application number and the project's bibliographic data is retrieved from the official register on the spot, with no PF1/PF37 upload. Choose the source when you create the project; file upload remains the default
  • Official Bibliographic Data — Applicants, IPC classifications, deemed filing date, and application type (Normal / PCT national phase / Divisional) now come from the Singapore patent register. Applicants and IPC were previously unavailable, and unpublished applications now carry complete bibliographic data
  • Search Within a Patent — Ask which paragraph of a prior art document discloses a feature, instead of reading the whole document. Hits come back ranked, with their paragraph numbers
  • Fetch Specific Paragraphs — Paste paragraph numbers back to retrieve those paragraphs in full, so a citation can go straight into a written opinion as "D1 [0034] discloses…"
  • Native Paragraph Numbers for Non-US Patents — Paragraph numbers now match the printed PDF for Japanese, Korean, German, and French documents. Numbers shown with a p prefix are positional and may not match the printed document
  • Family and Citation Scope — Choose between the strict same-priority family and the extended family on patent, family, and citation lookups
  • Tool-Call Audit Trail — Every tool call, including failed authentication, is recorded for security and compliance. Parameters are redacted and result content is never stored

Improvements

  • Divisional Search Accuracy — More accurate prior art search results for divisional applications
  • Refined Non-Patent Literature Counts — Search-report pointers and empty placeholder rows are no longer counted as non-patent literature
  • Family Lists in Citation Order — Family members are ordered the way an examiner would cite them (US A-publications first, since they carry paragraph numbers), so the first entry is the one best suited to cite
  • Clickable Patent Links — Publication numbers across tool results, CSV exports, and Lack of Unity conclusions now link to their patent page
  • Summaries for Very Long Applications — Applications with very large descriptions now produce a summary instead of failing
  • Bibliographic Accuracy — Filing dates, invention titles, priority claims, and PCT numbers are taken from the register rather than read off a scanned form

Action required for MCP users. Two citation tools were renamed — get_office_citations → get_application_family_citations and get_office_npl_citations → get_application_family_npl_citations. The old names no longer exist. Reconnect or use "Refresh tools list" in your client to pick up the new names, and update any prompt that hard-codes the old ones. The Lack of Unity field has_uoi is now has_lou.

Family and citation lookups now default to a narrower scope — the strict same-priority family, and a patent's own backward citations. Lists will be shorter than before. To widen the scope when you need full coverage, pass scope="extended" to the backward citation tools (get_patent_backward_patent_citations, get_patent_backward_npl_citations, get_application_family_citations) or family_scope="extended" to the patent and family lookups. Forward citations (get_patent_forward_citations) already default to the full family.

A patent's description is now a list of paragraphs, not a single string. Each paragraph carries its number. If you read description from the MCP response in your own code, it now comes back as a list — update any script that assumed a string.

July 4, 2026

New Features

  • Office NPL Citations — Non-patent literature cited by examiners at other offices, deduplicated and cleaned of search-report boilerplate
  • Lack of Unity Analysis Card — Interactive card showing the LOU verdict and invention groups, with one-click copy of Section 4 text
  • Classification Lookup — Find authoritative CPC/IPC codes from a keyword, symbol, or plain-language description
  • Project Lifecycle Tools — Mark projects complete, reopen, or delete them without leaving the chat

Improvements

  • Full-Text Family Representatives — Office citations now surface the best full-text family member (US A1 > WO > EP > …), no manual family look-up needed
  • Read Full Documents — Long texts continue past the 8,000-character cap via an adjustable budget, cursors, and per-section "Load more"
  • Full Family & Legal Status — Family and legal-status lists paginate, so even 500+-member families are fully reachable
  • Compact Publication Numbers — Outputs now use WO2015179075A1 (Google Patents style); both formats accepted on input
  • Lack of Unity Alignment — "UoI" renamed to "LOU" and Section 4 defaults to first-invention examination, per IPOS Examination Guidelines
  • Smarter Search Filters — The application's effective date is the default search ceiling; new classification (IPC/CPC) and patent-status filters
  • In-Widget Patent Cards — Click a search hit to open the full patent card in place, without losing the results list
  • Amended Documents Respected — Content reading, in-application search, and sequence analysis consistently use amended claims and description
  • Organization-Bound Credentials — OAuth connections and API keys are pinned to their organization; existing MCP connections need a one-time reconnect