Clerq Patent Research

Read progress and selected results from a patentability triage workflow. With no sections this returns only compact workflow and step status. Request `features`, `prior_art`, `analyses`, or `non_obviousness` only when that material is needed. Prior-art and non-obviousness lists are capped by `max_results`; detailed analyses are returned only for the supplied patent document numbers. A requested section not yet produced appears in `unavailable_sections` rather than as invented or stale data. Treat `next_action` as authoritative: poll only for `poll`, wait until `next_poll_at`, and stop for `use_results`, `ask_user`, `stop_failed`, or `stop_stalled`. The API owns the clock and rate-limits rapid polling, so the language model must not estimate elapsed time. A terminal section state of `failed` or `not_produced` will not become available through further polling. Args: workflow_id: Identifier returned by `start_patentability_triage`. sections: Optional result sections to load in addition to status. max_results: Maximum entries per bounded result list, from 1 through 20. analysis_document_numbers: Up to 10 selected patents whose reasoning is needed. Returns: Current progress and only the selected result sections currently available. Raises: ToolError: If the input is invalid, the workflow is unavailable, or reading fails.

get_patentability_triage

Find patents that disclose subject matter similar to a given natural language description. Given a natural language description of an invention, problem or technical approach — a sentence or a few paragraphs no longer than 512 tokens — this tool returns a ranked list of patents families that deal with similar subject matter. The match is semantic, not keyword-based or lexicographic. Boolean operators, quoted phrases, wildcards and field prefixes are not interpreted and only dilute the query, so describe the subject matter rather than constructing a boolean keyword search string. Longer, more detailed descriptions perform better than short keyword-like ones. Results are grouped by patent family, so each entry is one invention rather than one publication. Each result includes its patent number, title and abstract. use the patent lookup tool separately if more fields are needed. Optional filters narrow which patents can match. They do not replace the semantic query: keep the subject matter in `query`, and use filters for metadata such as dates, patent offices, CPC/IPC classifications, parties, legal status, or document numbers. Values within one list are alternatives. Different filter fields all have to match. On a party, `not` true excludes that name and `not` false includes it. A `*_date_to` value is exclusive and must be after the matching `*_date_from`. Args: query: The subject matter to search for, described in natural language up to 512 tokens in length. max_results: Number of patent families to return, from 1 to 25. filters: Optional metadata limits. Omit them to search the whole index. Returns: The requested number of patent families, most similar first. Raises: ToolError: If the query is unusable, or the API did not answer with results.

natural_language_patent_search

Retrieve selected data for up to 25 patents. Supply a list of up to 25 patent identifiers. Identifiers can be patent publication, application, or grant numbers. For example, values like 'US20160266243A1', 'EP4016123B1', and 'US14/643,719' are all accepted. Select only the fields needed for the user's current task. Claims and description sections contain full text and can be very large, so request them only when their text will be used. Citation fields contain compact citation identifiers and phases rather than nested patent records. Common choices are `title` and `abstract` for a summary, `claims` for claim analysis, dates and `legal_status` for status questions, parties for ownership questions, and citations or classifications for prior-art investigation. Args: identifiers: One to 25 publication, application or grant identifiers, including a jurisdiction where applicable. Duplicates are ignored. fields: One or more fields needed for the current task. Returns: Ordered resolved patents with their canonical document numbers and requested data, plus identifiers that could not be resolved. Raises: ToolError: If the input is invalid or the API call fails.

get_patents

Find patents that disclose similar subject matter to a given patent. Supply a publication, application or grant number of a published patent. For example, values like 'US20160266243A1', 'EP4016123B1', and 'US14/643,719' are all accepted. This tool will then return a list of patents, grouped by patent familiy, that deal with similar subject matter to the given patent. Results are ordered from most to least relevant. Each result contains its patent number, title and abstract; use the patent lookup tool separately if more fields are needed. Optional filters narrow which patents can match. They do not replace the patent number: keep the source document in `patent_number`, and use filters for metadata such as dates, offices, classifications, parties, legal status, or document numbers. Values within one list are alternatives. Different filter fields all have to match. On a party, `not` true excludes that name and `not` false includes it. A `*_date_to` value is exclusive and must be after the matching `*_date_from`. Args: patent_number: A patent identifier that can be either a patent publication, application, or grant number. max_results: Number of similar patent families to return, from 1 to 25. filters: Optional metadata limits. Omit them to search the whole index. Returns: The requested number of similar patent families, most similar first. Raises: ToolError: If the number is invalid, cannot be resolved, the filters are unusable, or the search fails.

find_similar_patents

Start an automated patentability triage for an invention disclosure. This is a billable operation that consumes the account's configured triage credit charge. It creates a durable Clerq workflow and returns immediately; claim/feature generation, patent searching, result analysis and non-obviousness evaluation continue asynchronously and may take several minutes or longer. After a successful start, do not call `get_patentability_triage` in the same model turn. Tell the user that the workflow is running, provide its workflow ID, and return control to them. A later turn can check progress and retrieve only the result sections needed, but never before the returned `next_poll_at` and only when `next_action` is `poll`. This version accepts disclosure text, not file uploads. Known prior art is optional and is normalized by Clerq before the workflow starts. Args: title: Short name by which the user can recognize this workflow. invention_disclosure: Technical description or feature list to assess. priority_date: Optional earliest relevant date, in ISO `YYYY-MM-DD` form. known_prior_art_patents: Up to 20 patent identifiers already known to the user. Returns: The workflow and run identifiers needed to inspect asynchronous progress. Raises: ToolError: If the input is invalid, credits are unavailable, or kickoff fails.

start_patentability_triage