Service Specification

This Service Specification governs the use of the Services. Capitalized terms not defined here will have the meanings in the Terms and Conditions, Usage Terms, Desktop Terms or other agreement between Organization and Patalyze governing the use of the Services (“Agreement”).

In the event of a conflict, Section 14.6 of the Terms and Conditions applies. The amounts and rates published by Patalyze for the Services, including the Model Rates and Request Fees published on the pricing page, are the current values. The mechanism for changing them is set by Sections 9.6 and 9.7 of the Terms and Conditions.

1. Access to the Services#

The Services are accessible by Users (a) via a website using their access credentials, (b) via Patalyze’s application programming interface (“API”) using API credentials (keys) issued to the Organization, (c) via the Patalyze desktop application (the “Desktop App”) installed on a User’s device, using their access credentials, and (d) via third-party AI assistants and agents that Organization or a User connects to Patalyze’s MCP servers with the User’s authorisation. Organization Content that such an assistant retrieves is transmitted to that third party at the User’s direction and is governed by its terms (Section 8.1.6 of the Terms and Conditions). All Services described below are available via the website, the Desktop App and the API, unless otherwise stated. Use of the Desktop App is additionally subject to the Desktop Terms. Organization is responsible for keeping website and API credentials secure and for all activity carried out under them, in accordance with Section 8 of the Terms and Conditions.

Organization shall not be allowed to access and control Services via the website through automated procedures, for example by scraping, circumvention of access controls, and use exceeding the published rate limits or fair-use limits. API access allows authorized automated access. Use of the API within those limits is permitted.

2. The Services#

2.1 Research Agent#

The “Research Agent” is the AI-based Service by which Patalyze processes data from the web and from Databases, together with Organization Content, and generates Output within Databases. A Research Agent is initiated by a User from the website, the API or the Desktop App, and is executed within the access rights of that User and no further. Once initiated, it may proceed without further instruction until it is completed, until the User cancels it, or until the AI & Data Usage Balance is exhausted.

Information obtained from the web is compiled from third-party sources not under Patalyze’s control. It is provided on an “as is” and “as available” basis, may be incomplete, inaccurate, outdated or duplicated, and Patalyze does not warrant its coverage, accuracy, timeliness or completeness (Section 11 of the Terms and Conditions). Organization is solely responsible for its use of such information, and in particular for ensuring that such use complies with applicable law and does not infringe the rights of third parties (Section 8.1.1 of the Terms and Conditions).

The Research Agent may delegate parts of a task to subagents, which run with the same access rights. To store information about a User’s preferences and context a Research Agent can write memories. A memory belongs to the User who created it and is visible only to that User. Memories are stored on Patalyze’s servers, except in local storage mode (Section 2.7), where they are stored on the device. The same applies to the skills a User writes and to the connectors a User adds. At a User’s direction the Research Agent can send emails to the User or to other Users of the Organization. A User can connect external tools (connectors) and Third-Party Applications to the Research Agent. Data the agent passes to them is governed by the terms of the respective provider (Section 8.1.6 of the Terms and Conditions). Output, including mappings, scores and assessments, is not legal advice (Section 3.1.7 of the Terms and Conditions).

2.2 Patent Search#

The Data API allows Users to query Patalyze’s Datasets by submitting a search query and to retrieve corresponding Records. A “Patent Search” is a request against the Data API that queries or fetches Records from the patents Dataset, together with their classifications. Such Records may be used directly or imported into a Database for use with the Research Agent (Section 2.1).

2.3 Product Search#

The Product Search Service searches public sources for products matching search criteria defined by the User and adds each product it identifies to the Database as a Record. The Input is the search criteria. The Output is the set of products added, with the source from which each was taken.

Products and the information recorded for them are taken from public third-party sources that Patalyze does not control, and whether a product matches the search criteria is determined automatically. They are provided on an “as is” and “as available” basis: the set of products added may omit products that match the criteria and include products that do not, the same product may be added more than once from different sources, and a product may have changed or been discontinued since its source was published. Patalyze does not warrant the coverage, accuracy, timeliness or completeness of the products added or of the sources recorded for them (Section 11 of the Terms and Conditions). Organization is solely responsible for its use of the products added and of the sources recorded for them, and in particular for ensuring that such use complies with applicable law and does not infringe the rights of third parties (Section 8.1.1 of the Terms and Conditions).

2.4 Web Search#

The Web Search Service searches the public web for pages matching a User-defined search query, optionally narrowed by source category, domains and publication dates, at a standard or deep search depth. The Input is the search query and its filters. The Output is a ranked list of results with page titles, links, metadata and text excerpts (“Web Search Results”).

Web Search Results are compiled from third-party web sources that Patalyze does not control. They are provided on an “as is” and “as available” basis: the ranking is generated automatically and is neither an assessment nor an endorsement of a result, titles, metadata and publication dates are those reported by the source and may be wrong, excerpts are extracts which may be incomplete or read differently outside the page they were taken from, and a page may have changed or ceased to be available since it was retrieved. Patalyze does not warrant the coverage, accuracy, timeliness or completeness of any result (Section 11 of the Terms and Conditions). The content of the pages linked to remains that of their publishers. Organization is solely responsible for its use of Web Search Results and of those pages, and in particular for ensuring that such use complies with applicable law and does not infringe the rights of third parties (Section 8.1.1 of the Terms and Conditions).

2.5 Browser Sessions#

Where reading a web page requires a browser, the Services open a hosted browser session and read the page on the Organization’s behalf. A session is opened for up to fifteen minutes and may serve more than one request. On the Desktop App, pages are read with a browser installed on the device where one is available. If the Desktop App cannot start or connect to a browser on the device, the request is served by a hosted session instead, and further requests may be directed to that hosted session for up to fifteen minutes without a renewed attempt on the device. In local storage mode (Section 2.7) no hosted session is opened.

Pages are read from third-party web sources that Patalyze does not control, as they stand at the time of reading. What is read is provided on an “as is” and “as available” basis: a site may block or rate-limit an automated browser, may require sign-in or consent before its content is shown, and may serve different content to different visitors or change between one reading and the next, so that a page is read only in part or cannot be read at all. Patalyze does not warrant that any page can be read, or the accuracy, timeliness or completeness of what is read (Section 11 of the Terms and Conditions). Organization determines which pages are read on its behalf and is solely responsible for that reading and for its use of the content read, and in particular for ensuring that both comply with applicable law and with the terms of the site concerned and do not infringe the rights of third parties (Section 8.1.1 of the Terms and Conditions).

2.6 Document Processing#

Documents and images submitted as Input, and images the Services capture from a web page, are processed to extract their structured content so that the other Services can use it. Where the text of a document cannot be read directly, in particular where the document is scanned or image-based or where its text encoding is damaged, the document is processed by an optical character recognition service.

2.7 Local storage mode#

Organization or a User can opt-in to store a Database on the User’s device instead of on Patalyze’s servers (“local storage mode”). While local storage mode is active, the memories the Research Agent writes, the skills a User writes together with the files bundled with them, and the connectors a User adds together with their access credentials (Section 2.1) are also stored on the device and are not stored on Patalyze's servers. A Research Agent reads only what is stored on the device while local storage mode is active. The following Services continue to involve Patalyze’s servers and the providers named in the Privacy Policy, and the Organization Content contained in the respective request is transmitted accordingly:

• A request to a model that Patalyze provides is transmitted through Patalyze to the provider hosting the model, and the response is returned the same way. Patalyze stores neither the request nor the response, only the usage record (Section 3.1).
• Patent Search transmits the text and any file used as a search criterion to Patalyze’s Data API. Patalyze stores neither the request nor the response, only the usage record (Section 3.2).
• Product Search and Web Search transmit the search criteria and the query (Sections 2.3 and 2.4). Patalyze stores neither the request nor the response, only the usage record (Section 3.2).
• On the website, Browser Sessions (Section 2.5) are hosted sessions. On the Desktop App a browser on the device is used and no hosted session is opened in local storage mode.
• Emails the Research Agent sends are transmitted through Patalyze.

Content that is transmitted to a model provider that Organization supplies (Section 3.1) does not pass through Patalyze. Patalyze cannot access, restore or recover a locally stored Database or a memory stored on the device, and the availability commitment in Section 4.1 of the Terms and Conditions does not apply to locally stored content. Browser storage may be cleared by the browser or the operating system when it is not reserved. Organization is responsible for the security of the device, including access control, device encryption and backup.

3. Pricing#

Consumption-based Services are funded from the AI & Data Usage Balance held on the Organization’s account. For each request, Patalyze deducts the applicable charge, where one arises, after the request has been processed:

• AI Usage (the Research Agent): the AI Usage Charge calculated under Section 3.1.
• Data Usage (patent searches, product searches, web searches, hosted browser sessions and document processing): the Request Fee for that request under Section 3.2.

3.1 AI Usage#

The Research Agent (Section 2.1) is charged as AI Usage. Each AI Request it makes to a model that Patalyze provides is charged at the AI Usage Charge. Requests already made to a model in the course of a run are charged even where the run does not complete, including where the User cancels it. Where the Research Agent runs on a User’s device, cancelling the run does not stop a request that has already been sent to the model, and that request is charged in full. A Research Agent run may also make Data API Requests (for example a patent search) and other Data Usage requests; those are charged under Section 3.2, not as AI Usage.

The “AI Usage Charge” is calculated per request from the Model Rates of the model used: (Input Tokens other than Cached Input Tokens × the model’s then-current input rate) + (Cached Input Tokens × the model’s then-current cache read rate) + (Output Tokens × the model’s then-current output rate). Cached Input Tokens are Input Tokens served from the model’s context cache. Where no cache read rate is published for a model, they are billed at its input rate. Input Tokens written to the model’s context cache are not Cached Input Tokens and are billed at the model’s cache write rate. Output Tokens include all tokens generated by the model, including reasoning/“thinking” tokens, which are billed at the model’s output rate.

The model rate table on the pricing page publishes the input, cache read, cache write and output rate for every model Patalyze provides. These are the Model Rates. They are Patalyze’s own rates, stated in euro net per 1,000,000 tokens, and are set by Patalyze at its reasonable discretion (§ 315 BGB), in particular having regard to the rates published by the provider hosting the relevant model.

Organization may configure the Services to use a model provider with credentials or a subscription that Organization supplies, instead of a model that Patalyze provides. Where it does:

• No AI Usage Charge arises for those requests, because the model is not one that Patalyze provides and has no Model Rate. The Request Fees under Section 3.2 are unaffected and continue to apply to any Data Usage such a request gives rise to.
• The requests, and the Organization Content contained in them, are transmitted from the User’s browser or device directly to that provider and do not pass through Patalyze’s infrastructure. They are governed by Organization’s own agreement with that provider, and Patalyze is not a processor or sub-processor in respect of them.
• Statements Patalyze makes about where the models it provides are hosted, and about the providers hosting them not using Organization Content to train models, do not apply to such requests.
• Where the configured provider cannot be used, for example because its key was removed or it was disabled by Organization’s policy, the Services use a model that Patalyze provides instead. The model in use is shown in the model selector, and such requests are charged at the AI Usage Charge.
• Patent Search, Web Search, Product Search, Browser Sessions and Document Processing on Patalyze’s servers continue to be performed by Patalyze and its providers regardless of the model provider configured (Section 2.7 lists what stays on the device in local storage mode).

3.2 Data Usage#

Data Usage is charged at a fixed fee per request or unit of use, published in advance.

The fees are published in the data rate table on the pricing page, which states, for each request type, the unit charged and the fee for that unit. These are the Request Fees. They are Patalyze’s own fees, stated in euro net. Each request or unit of use is charged at the Request Fee published for it.

What counts as one such request or unit of use is as follows:

• Patent Search: each query or fetch executed against the patents Dataset, or against the classification data, through the Data API is one request. A search the Research Agent or a connected AI assistant runs against the patents Dataset is one request.
• Product Search: each product the search identifies which matches the criteria is one unit of use.
• Web Search: each search query executed against the Web Search Service is one request, charged according to whether it is executed at standard or deep search depth.
• Browser Sessions: each browser session opened is one unit of use, charged once when the session is opened. A session which becomes unavailable within the fifteen minutes for which it is opened is replaced, and the replacement is charged as a new session. Sessions on browsers installed on the device in the Desktop App are not charged.
• Document Processing: each page of a document processed by optical character recognition on Patalyze’s servers, and each image, is one unit of use. A large image that is processed in parts counts one unit per part. Documents whose text the Services read directly on the User’s device at upload, and documents processed on the device in local storage mode, are not charged.

4. Processing, fair use and suspension#

Patalyze may queue, batch, slow or limit the throughput of requests per Organization, including via the rate limits in Section 5. Processing is not guaranteed to be sequential or concurrent. The Service may process requests more slowly, or temporarily suspend access on a case-by-case basis, where the size of the Input (e.g., number of Records or file size of uploaded documentation) significantly exceeds normal and reasonable use and thereby affects availability of the Services to other Organizations.

5. Limitations#

• The number of concurrent Research Agents per organization is limited by the subscription plan, as stated on the pricing page. Research Agents started beyond that limit are queued and started as earlier runs finish.
• The Data API and Patalyze’s MCP servers are subject to a rate limit of 100 requests per minute. Other interfaces apply their own technical limits. Patalyze may throttle or temporarily suspend access that exceeds these limits or that affects availability for other Organizations (Section 4).
• Extraction of a substantial part of a Dataset for use outside the Services is restricted under Section 8.1.1(l) of the Terms and Conditions, including via the Data API.