Changelog

Large-scale research can be quite token-hungry and costly. AI subscriptions like your ChatGPT and Claude subs are heavily subsidized, meaning that you get a lot more tokens compared to the money you spend. You can now use your own subscription on Patalyze Desktop to save money.

• Claude and ChatGPT subscriptions are supported.
• Credentials are only stored locally on the device.
• AI subscriptions are not supported on the web, only in Patalyze Desktop.

Learn more about using your own subscription here.

Local storage mode keeps your research out of the cloud. Turn it on under Data controls in settings and research data, skills, connectors and memories are only stored in your browser or, in Patalyze Desktop, on your device.

• Research agents run locally too. Only model requests leave the device, going straight to your provider.
• Real-time collaboration and cloud agents are disabled while it is on.
• A research can move either way: sync a local one to the cloud, or download a cloud one back.
• Organization admins can set a policy for every member, from forbidding local storage mode to enforcing it.

Learn more about local storage mode here.

Patalyze Desktop is available for macOS, Windows and Linux.

• Supported platforms are macOS 13.3 or later on Apple silicon and Intel, Windows 10 or later on 64-bit and ARM64, and a 64-bit AppImage for Linux.
• Agents can run on a connected API key or an existing Claude or ChatGPT plan, sent from the device straight to the provider.
• Local storage mode keeps a research on the device, with storage reserved on disk.
• Cloud sync enables real-time collaboration with your teammates and agents.

Download it here.

Research agents now run on a provider of your choice. You can either choose a model hosted by Patalyze or provide your own API key.

• Eight providers are hosted by Patalyze and served from the EU: Claude, OpenAI, Gemini, Kimi, GLM, Qwen, DeepSeek and MiniMax.
• An Anthropic, OpenAI or Gemini key can be connected instead, optionally pointed at a custom endpoint.
• Keys stay in the browser, or in the operating system keychain in the desktop app, and never reach Patalyze servers.
• Organization admins choose which hosted providers members can use, and whether custom providers are allowed at all.

Learn more about model providers here.

Skills extend the domain knowledge of research agents by giving the agent instructions on how to solve a specific problem. Patalyze ships in-built skills for the most important use cases, and custom ones can be written from scratch or imported from directories or other tools.

• A skill is a markdown instruction file plus reference files.
• Skills are compatible with any skill from Claude or any other AI app.
• Pinning a skill brings it into the agent window so you can quickly activate it when spawning an agent.

Learn more about skills here.

Added research agents. An agent takes a task or a goal and works for minutes or hours: searching patents, browsing for products, driving a browser, building claim charts, and writing the results into tables, notes and dashboards.

• On heavy workloads an agent can spawn subagents. The number of concurrently working agents is limited by your plan.
• Agents proactively ask when a decision is needed, and email once a report is ready or they need input from you.
• Attach any MCP server as a data source, with Semantic Scholar, BioContextAI, Wolfram and DrugBank among those preconfigured.

Learn more about research agents here.

Added two MCP (Model Context Protocol) servers so AI assistants can work with Patalyze directly. Claude and ChatGPT are supported, as is any other MCP client.

• Global Patents is read-only: Tools for searching and reading the worldwide patent corpus and the CPC/IPC classification tree. It never touches a research database or stores any data.
• Patalyze is read-and-write: Tools covering a research database end to end (patent search and import, product discovery and browsing, claim-by-claim mappings, tables, dashboards, and notes).

Learn how to connect and use MCP here.

AI and data usage is now priced in euros and itemized per action.

• AI work (mappings, browsing) is billed at the model's token cost plus a surcharge; patent searches have a fixed price each.
• Each plan includes usage that is reset monthly. After included usage is exhausted, the organization admin can purchase top-ups valid for 1 year. Find more information on the pricing page.
• Each research database has a usage costs panel: an itemized list with a running total, what each charge was for, and the token breakdown behind AI work. Settings show usage across the whole organization.

Added dashboards as a new page type in research databases, alongside tables and notes. Dashboards show aggregated views of the data and redraw live as it changes.

• Chart types: single-number stat, bar, radar, funnel, line, area, pie, donut, ring, sankey, and sunburst.
• Charts count patents, products, or the mappings between them, grouped by any dimension: date, applicant, classification, category, or custom attribute.
• Patent and product charts have their own search-style filters; clicking a bar or slice opens the records behind it.
• Charts can be inserted into notes from the slash menu; they stay live and fill the screen in presentation mode.

Learn more about dashboard and charts here.

Added PDF and Word export for single records. Excel and CSV exports still cover a full research database or selected rows.

• Patents export as a PDF or Word report, with mapped claim charts appended.
• Mappings export on their own as a PDF or Word claim chart: each claim element with its color-coded verdict and reasoning, referencing the product's files and features.
• Products export as a Word document: description, features, properties (including custom attributes), and the mapped patents ordered by match score.

Added citations in notes, claim mappings, and product descriptions, linking text to its source.

• Citable sources: patent claims and claim elements, specification paragraphs, figure regions, rows of other claim mappings, and product description sections. Inside a mapping, a product's images, PDFs, and features can also be cited.
• Insert with "/" in the editor: the source opens in the side panel, and clicking the exact spot sets the citation.
• Citations render as superscript numbers like [1], with a matching numbered list beside the block; clicking a number opens the source.

Research agents can now browse the web with a hosted browser session. This allows the agent to do deep scans of each subpage of a website, take screenshots and download files. The research agent can use the results to build a detailed product documentation with screenshots and deep links to specific paragraphs on a website.

The research agent will automatically start a web browser session in followig scenarios:

• Ask the agent directly to use a browser
• During Infringement Detection to find more information about each individual product
• During Freedom-to-Operate to find more information about your product when your specification is too narrow

During browsing the agent window transforms into a live view of the current agent activities.

Added notes: rich-text pages inside a research database, alongside its patents and products. Every analysis workflow drafts one as an executive summary.

• Editor with headings, lists, tables, and a table of contents; blocks are inserted from a "/" menu.
• Supported content: images, PDFs, 3D models, chemical structures, embedded patent and product tables, and charts that count live records and redraw as data changes.
• Publication numbers and product names become links with hover previews; claims, figure regions, and mapping rows can be cited with margin footnotes.
• Executive summaries open with a verdict and list the highest-risk patents and the claim elements that read on the product.

Learn more about notes here.

Added product search: finds commercial products on the web that match a description and imports them into a research database. Runs in the background.

• Search from a plain-language feature description, narrowed by category, manufacturer, distributor, and territory, or seeded from a patent's first independent claim.
• Results include only real products, services, and software (no blog posts, news, or research papers).
• A live panel shows candidates grouped into added, filtered out, and pending, with the reasoning behind each verdict.
• Every kept product is browsed, distilled into a feature tree, and mapped claim by claim against every patent in the database.

Patent search now matches by meaning: describing an invention in plain words finds patents about the same idea, even when worded differently.

• Search with a sentence or paragraph; results are ranked by semantic closeness.
• Search from a document: upload a PDF or an image, or combine text and files in a single filter.
• Works together with the precise filters: classifications, priority-date ranges, assignees, inventors, territory, legal status, and kind code.
• Every active filter shows a live match count; results preview in a table, sortable by relevance, before saving into a research database.

Every patent in a research database now carries a legal status, derived from official patent-office legal records worldwide and grouped into three categories.

• In-force: granted, maintained, and still in effect, including patents under opposition or challenge.
• Pending: filed, under examination, or marked for grant or refusal, and not yet granted.
• Dead: no longer in force (withdrawn, abandoned, rejected, expired, lapsed, or revoked).
• Searches and tables filter by status, including or excluding each, with a live match count.
• The exact status shows as a color-coded chip on each patent (green, blue, or red), down to every family member.

Products can now be created from a written description alone, or extended with one, in a rich text editor. The description is analyzed into features and mapped against the database's patents.

• Text formatting: headings, lists, tables, quotes, and dividers.
• Inline visuals: images, sketched diagrams, 3D models, and chemical structures from SMILES strings.
• Images placed in the description join the gallery and are analyzed alongside uploaded files.

Replaced all icons with one consistent set across menus, toolbars, tables, panels, and dialogs. Purely visual: nothing changes position or behavior.

• Menus and toolbars: muted single-weight outline icons.
• Data fields (patent fields, product attributes, column types, filters): two-tone filled icons.
• Patent statuses: a distinct two-tone icon and color each for in-force, pending, and dead.
• Destructive actions: red icon and label.

Redesigned the patent and product pages as a single scrolling column with a side panel.

• Main column: a patent's claims, figures, legal events, and citations; a product's description, features, images, and files.
• Inside a research database, a patent page also shows its claims scored and ranked against the database's products.
• Side panel: key properties (editable for products) and quick actions (open a patent's PDF or description, browse a product's site, or hand either to ChatGPT or Claude).
• Sources opened from the text (a figure, PDF, claim, or claim-to-product match) open in the panel, with previous/next navigation through results.

Added global search, opened from the sidebar search icon or by pressing F. Results appear live while typing, grouped by category.

• Finds research databases and products by name, and patents by publication number.
• Queries that look like a patent number also match patents from the worldwide corpus not yet in any database.
• Opens any settings page directly (profile, billing, API key, and more).
• Quick actions: start an offensive or defensive research, a patent or product search, or invite a teammate.
• Keyboard navigation: arrow keys to move, Enter to open, Escape to close.

The library (home of every research database in an organization) is now customizable. Layout and column choices are remembered.

• Two layouts: Explorer (compact table) and Grid (cards).
• Any column can be shown or hidden; in Explorer, columns can be reordered and sorted by products, patents, tags, or date.
• Folders can be created and nested; databases are organized by dragging them in and out.
• Research databases can be filtered by tag, people, access, or date; sharing can be changed directly from the library.

Added an "Add patent" row to the patents table for adding patents directly by publication number.

• Accepts a single number or a pasted list, one number per line.
• Input is normalized: spacing, punctuation, and capitalization are cleaned up, so "ep 3324419 b1" and "EP3324419B1" match the same patent.
• Unrecognized numbers are flagged by name and skipped, duplicates are left untouched, and valid numbers go through.
• The number appears immediately; title, claims, and dates load in the background.

Added Excel and CSV export for patents and products, from selected rows up to a whole research database.

• Selected rows: pick rows in a table and export from the selection bar.
• Whole database: export everything from the top-right menu, not just the rows on screen.
• Copy puts the selected publication numbers or product names on the clipboard, one per line.
• Exports keep the configured columns, including custom attributes formatted to their type and patent-to-product match scores; Excel opens as a styled workbook named after the database.

We've launched Patalyze! This is just the starting point and we will extend and improve the product over the next years. Have fun!