Patent Bear

The full text of one open-access journal article from PubMed Central, by PMCID (from a search_patents scope:npl hit), DOI or PMID: title, journal, volume, pages, publication date, authors, license, abstract and the body as plain text (paged with from_char), figure captions and the reference count. Use it to read and quote a paper the search surfaced, for prior art or for the science behind a claim. Only the PMC open-access set has full text (about 5.5 million); computer-science papers from OpenAlex have an abstract and oa_url on the search hit and nothing here. Every account.

get_article

Read the stored Overview / Claim Summary analyses for a patent from the account's Library (whoever authored them), with model provenance and timestamps.

get_patent_analysis

Has this US patent been litigated or challenged? Takes a US grant; a publication or application number resolves to its grant when one issued (resolved_from says so), and when nothing has issued the answer is found:false with the application's number and a pointer to get_file_history for its status — not an error. One call returns district-court cases (1963→present complaint-level dockets: filing date, parties, PACER cause code — no docket events, so stays, motions, counterclaims and current status need PACER), FJC outcome buckets for terminated cases (settled / verdict / dismissed / ongoing, and for each terminated case the clerk's coding of how it ended, which side the judgment favoured, the nature of the judgment and the amount awarded in thousands; no_outcome_record = no outcome row matched, the case may be open), PTAB proceedings (IPR/PGR/CBM) with status, dates and petitioner, and the written decisions in the opinion corpus — those listing the patent in suit and appeals that name one of its PTAB trials (matched_by says which). patent_outcomes answers "has this patent been held invalid, unpatentable, ineligible, infringed or construed, and on what ground" directly: every decision in the corpus that recorded an outcome for this patent, newest first, with the result, ground (s101, s102, s103, s112, inf, cc, unenf), claims, and the decision's link. They are a language model's read of each opinion, per decision and not the patent's current status (a finding can be reversed on appeal): open the decision with get_opinion and confirm before stating it. Also returns certificates: reexamination (ex parte / inter partes) and AIA-trial certificates from Patent Bear's certificate corpus — which claims were cancelled, confirmed, amended, or added, and when. Reexamination proceedings themselves are not indexed, only their certificates. Public litigation data.

get_legal_activity

List the notes in a matter's notebook — the long-form workspace on the matter, shared with everyone who can see it. Returns each note's title, when it changed, note_id, and a link that opens it. Pass include_content to read the notes themselves, or note_id for one note. Call it before writing so you revise the right note instead of starting a second one on the same subject.

list_notes

List the account's analysis runs across every analysis (Prior Art, Charting, Citations, Disclosure, Priority, Claim Construction, and more) — including runs started in the browser. Newest first. Filter with analysis and/or patent. Each row carries status and a deep link to the results. Answers "show me my last prior art run".

list_research_runs

List the claim charts in the user's Patent Bear Library — name, target patent, matter, origin (Prior Art analysis, Charting analysis, Disclosure analysis, hand-built), last edit, and a link that opens each chart. Filter with patent and/or matter (id or name). Then call get_claim_chart for the full element-by-element content. Requires a plan that includes the Library; otherwise the call returns 403.

list_claim_charts

List saved documents across accessible matters: definitions sections, claim charts and finished drafts (IDS, Definitions section, Double patenting chart). Optional matter id and type filters. Returns document ids, updated times and links (editors for definitions and charts, the run page plus download_link for drafts); the list lives in the Library at /library/documents. Use get_document to read a document before revising it.

list_documents

List the account's drafts (Definitions sections, IDSs, double patenting charts), newest first, with status and links. Filter with kind and/or patent. Answers "show me my last IDS".

list_drafts

List the user's highlights (with notes), claim terms, and claim constructions for one patent — or, with decision_id, the highlights on one court decision — each with a link that opens the reader on that passage. Notes come grouped by named set (note_sets, with counts and who created each: the user or an AI agent by app name); pass set_name to read one set in its order on that patent (set.sort = document | date | custom; order it with save_highlights order / sort). A set spans documents: set_name ALONE lists every patent and decision it has notes on, each with a link, plus a library_link that opens them all — the one link to hand back for a brief. Use it before saving to avoid duplicates, or when the user asks what they have marked up.

list_annotations

List the account's Library matters (client-matter folders) with ids, names, and item counts. Call before save_patent when the user names a matter, to resolve or confirm it.

list_matters

Create a Library matter (client-matter folder) by name. Idempotent — if a matter with the same name exists, returns it instead of duplicating. Requires a plan that includes the Library; otherwise the call returns 403.

create_matter

Which patents are listed for an FDA-approved drug, and when they expire. FDA Orange Book patent listings: which patents an NDA holder has listed against an approved drug, with the use codes, the expiry the holder submitted, pediatric extension and delisting flags. Look up by patent ("is this patent Orange-Book listed, for what"), drug (trade name or active ingredient, "Eliquis", "apixaban"), applicant (NDA holder) or an NDA application number. From the monthly edition named in the response. Exclusivities (NCE, ODE, pediatric) are a separate table and are not returned. Every account.

get_orange_book

Human-readable guide to everything Patent Bear can do through this connector — each tool, what it's for, and example prompts a user can say. Call it when the user asks what Patent Bear can do, how to phrase a request, or when you're unsure which tool fits. Pass topic to get one section: search, records, figures, ownership, citing, citations, family, legal, file_history, definitions, similar, office_actions, labs, reading_results, corpora, library, plans, or version.

help

How patents define a term, or how one patent defines its own terms: the definitions as written in the specifications, quoted with the patent, applicant, art area and date, and how many patents define it, so the answer rests on real drafting rather than general knowledge. The patent dictionary: 94.8 million explicit definitions mined from US patents and published applications, 1976 to today — the sentences where a specification actually defines a term ("as used herein", "refers to", "means"), plus every abbreviation a patent spells out. Three questions. patent = how THIS patent defines its terms (add term to ask about one). term = how the whole corpus has defined that term, newest first, with filters: assignee (substring; exact for one entity), cpc (the field), from / to (issue date), kind (grant | pub), definition_kind, limit (max 200). assignee + top_terms = which terms that applicant defines. Use it when drafting a definitions section, checking whether a term already has an accepted meaning in the art, showing how an assignee or an era used a word, or testing a construction against how others defined it. Discovery: assignee + top_terms lists the terms that applicant defines most often (count, distinct forms, patents, first/last seen) — the first call of a definitions draft, because it finds the 40–60 terms an application carries rather than the few someone typed. Batch: pass terms (a list, up to 40) and/or patents (up to 20) to look up a whole claim set's terms — and the definitions already inside related patents — in ONE call, each term searched in singular and plural and merged. Use batch when drafting a definitions section for an application: the definitions_workflow prompt carries the full recipe, and the finished draft belongs in the matter notebook (write_note). Single-term matching is on the normalized term, so singular and plural are different entries — search both, or use terms for the merge. The rows carry the definition sentence but NOT its location: never cite a column or line from this tool; run the sentence through find_in_patent for a verified cite. Every account.

get_term_definitions

Show a patent's drawings and what each figure depicts. The drawings of a US or EP grant or publication. Without sheet it lists the drawing sheets and returns figure_descriptions, the record's own "FIG. 1 is a block diagram of…" paragraphs; with sheet or sheets (up to 4 per call) it also returns those sheets as images you can look at. Use it when the user asks what a figure shows, when a claim element is best understood from a drawing, or to check a reference numeral against the text. A sheet is a page and can hold several figures, so pick sheets from figure_descriptions and the labels printed on the sheet; describe a drawing only from an image you received, and cite it by figure label and reference numeral. figures_status says where things stand: ready; pending = an older US grant whose drawings are being extracted from the issued PDF (tell the user they are being prepared, follow figures_note for when to call again and when to stop, and never report that the patent has no figures; pdf_link is the issued PDF, which always has the drawing pages); unavailable; none. Every account.

get_figures

A patent's family: its US continuations, divisionals and parents, and its foreign counterparts (EP, JP, CN, KR, WO). Patent family. scope:'us' (default) = the US continuation/divisional tree — parents, children, siblings with application numbers, from the grant index. scope:'global' = the worldwide family from USPTO Global Dossier (the application page's Family Tree): every member application by office (US, EP, JP, CN, KR, WO...), its kind, its publications, its own filing_date at that office (publications[].date is the publication date — never read it as a filing date, and never assume a counterpart shares the priority filing date), and the priority claims linking them, plus a link to the interactive tree. Use for continuity questions, foreign counterparts, and double-patenting or priority-chain context.

get_family

Look up a US or EP patent or publication by number: title, owner, dates, status, claims and full text. Fetch the full text of a US or EP grant or publication by id (e.g. US12574394B2, EP1674995A2; a bare grant number like US11470243 or 11470243 resolves too), pulled from the Patent Bear record store. EP records return biblio (publication, application and priority dates, applicants, inventors, IPC/CPC, designated states), abstract, claims with their dependencies and the description with each paragraph's printed number ([0073], in paragraph_numbers); the US-only blocks below (term, certificates, column:line) do not apply to them, and an EP document is cited by claim number or paragraph. Returns title, biblio (issue date, application number and filing date, earliest priority date, pre-grant publication, assignee, inventors, examiner and art unit, CPC, continuity chain, and application_status = the Office's current status of the application — pending, abandoned, allowed, patented as — with restriction and terminal-disclaimer events, for publications too), term (patent term adjustment with the A/B/C breakdown, an expiration estimate, terminal-disclaimer status from our file-wrapper mirror and the front page, any disclaimer documents on file, and double-patenting rejections given or received from our office-action corpus, and maintenance_fees = whether the grant is still in force by fees: paid through 4 / 8 / 12 years, expired for non-payment, reinstated — grants only). For who holds it now, get_assignments., abstract, full claims (with numbers), and description sections — verbatim text suitable for quoting. Read dates from biblio; never infer an issue date from a file-history event. For issued grants every description paragraph carries its column:line range (col_lines, parallel to paragraphs; format:'text' prints it inline as [¶12 · 4:12-19]) — the same cites the Patent Bear reader shows, so you can read the whole specification and cite as you go. Claims are the OPERATIVE version: after any reexamination / AIA-trial certificate, cancelled claims stay in place with status 'cancelled', amended claims show the certified text with the as-issued wording in original_text (status 'amended'), and substitute claims are appended (status 'added'); claim_version says which certificates applied. A statutory disclaimer under 35 U.S.C. § 253 marks its claims 'disclaimed'. claims_status lists the numbers, the disclaimer documents and any certificate of correction on file. Never chart, construe, or assert a claim marked cancelled or disclaimed, and quote amended claims from the certified text. Use format:'json' for structured fields, 'text' for a single flat document with inline cites, or 'both'. Use fields to limit to abstract, claims, and/or description. Articles are not available here; use the abstract returned by search_patents instead. For the exact lines of ONE sentence (rather than its paragraph's range), call find_in_patent with quote = that sentence.

get_patent_record

Prosecution timeline for a US grant or publication, with application_status = the Office's current status (pending, abandoned, allowed, patented) and its restriction and terminal-disclaimer events. The file wrapper grouped into dated events (office actions, amendments, notices of allowance), each marked applicant/USPTO and flagged when it's a rejection or allowance decision. Each document carries document_id and text_available; read a document's text with get_file_history_document. Reexamination papers live in their own wrapper: pass application = the certificate's source_number (from term_summary.certificates or get_legal_activity) to read a reexamination file. Reissue patents are cited as USRE050016E1 (six digits). Use for prosecution-history questions, estoppel context, and "what happened during examination". Document PDFs are viewable at the returned source_url. Also returns term_summary: issue date, PTA days, expiration estimate, terminal-disclaimer status, double-patenting rejections, and any reexamination / AIA-trial certificates (the post-grant claim outcome) — read those from here rather than reconstructing them from wrapper dates. The response carries as_of, when this wrapper was last read; when the answer depends on whether something newer has been filed, give the user that date and ask whether they want it refreshed (then call again with refresh:true) rather than assuming the timeline is current.

get_file_history

Read one court decision in full by decision_id (from search_litigation or get_legal_activity): the opinion text page by page as parsed from the court's PDF, the reporter citation when published (cite as "<citation>, slip op. at <page>"), footnotes, the parties, judge, patents in suit, disposition, and links: the Patent Bear reader, pdf_link (the court's own PDF), and the govinfo record. The first call also returns the structured read of the decision: holding and summary (model-written guideposts, never quotable), winner, rulings (one per legal issue, each with the court's own quoted paragraphs, their page numbers and a reader link that highlights them), patent_outcomes (per patent: result, ground, claims), and constructions (each construed claim term with the paragraph where the court construed it). The quoted paragraphs are the court's text located by pointer, not model text; the labels around them are model-assigned. Pass analysis: false to skip it, or analysis: true to get it again on a later page. Pass include_pdf: true to receive the PDF itself alongside the text, so you can read the original and confirm every quote against it before it goes into a filing; keep it off for routine reading (long opinions are large). Long opinions page: follow next_page with from_page. Before you cite a pinpoint, pass quote (and the page you think it is on): quote_location returns the page the words are actually on, corrects an off-by-one, and warns when the stored text dropped a line-break hyphen ("patenteligible") so you print the quote as the PDF has it.

get_opinion

Read the current saved document, including wording, selected terms and source quotations for definitions, the structured claim chart, or a draft run's JSON and download link. Pass type and document_id from list_documents or a document link. Before revising definitions, read this version and pass its updated_at to save_definitions with note_id to avoid overwriting another person's edits. Preserve source quotations and selection flags.

get_document

Read a draft made with draft_definitions, draft_ids or draft_double_patenting_chart (or on the site): status, the Word download_link, the draft page link and the JSON data behind the document. Use it to re-open or re-download an existing draft; this read does not use research credits.

get_draft

The TEXT of one file-wrapper document — an office action, an amendment, the applicant's remarks, a claims listing, an examiner's reference list — sectioned. Take document_id from get_file_history (documents with text_available:true). Text comes from the Office's rendition when Patent Bear holds it, otherwise from OCR of the PDF: a first read of a scanned document returns status:processing — tell the user, then call again with wait:true. Retry at most twice; a document the servers could not read returns an error with retry_after_seconds — stop, give the user pdf_link, and answer from the metadata (for a terminal disclaimer, get_file_history's term_summary.terminal_disclaimer has the filed/approved events without the text). Use it before characterizing prosecution: which statute a rejection rests on and over what art, what the applicant amended and what it argued to distinguish the art, what the examiner said in allowing. Long documents page by section (follow next_section). No page numbers in the text — use pdf_link for a page cite.

get_file_history_document

Who owns a US patent, publication or application of record, and its chain of title. USPTO recorded assignments: the reel/frame documents filed with the Assignment Recordation Branch. Two modes. patent (a US grant, publication or application number) = that document's recorded history newest first, with latest_recorded_assignee = the newest ownership-type conveyance (assignment, merger, change of name); records filed against the application before grant are included. party = every record naming a company or person as assignee (role assignee, default), assignor, or either — "what has X acquired since 2024", with from / to on the recorded date and offset paging. Recording is voluntary and a record is what the parties filed, not a title opinion: say "the latest recorded assignee is X (assignment recorded <date>)", never "X owns it". ownership_transfer false = a security interest, license, release or correction. Use it for who holds a patent, chain of title, portfolio acquisitions and encumbrances; the front-page assignee on get_patent_record is the assignee at grant only. US only (EP transfers are on the EPO Register). Every account.

get_assignments

Start a Patent Bear analysis, exactly as if run from the Patent Bear site — the run appears in the account's Library and completes on Patent Bear's servers. Availability and required research credits depend on the account and analysis. Prior Art, Disclosure, Citations and Charting take a patent; Sequence and Chemistry take a patent or a sequence / structure; Clearance takes a product description and Invention an invention disclosure. Every run is TWO calls because credits may be used. The first call, without intake_confirmed, returns the intake and starts nothing: for Prior Art, the operative claim set (after any reexamination or AIA-trial certificate: cancelled claims excluded, amended and substitute claims carrying the certificate's text), each claim's status and text, the required credits, and the constructions already saved on the patent; for the other analyses, what will run (patent, claims, references, family members, the extracted elements for clearance and invention, the sequence or structure) and the required credits. Put the intake to the user, then call again with intake_confirmed:true and their choices; when the intake carries confirm_with (clearance, invention), echo those fields on the second call. The server does not choose the claims; a failed run is not retried without the user. If the account was recently working in a matter (e.g. after save_patent), the run inherits that matter automatically. Returns the run id, the credits held, and a Patent Bear link where the results appear. TIMING: Prior Art and Invention runs take 15-20 minutes (thorough mode longer); Charting and Clearance several minutes; Sequence, Chemistry, Citations and Disclosure a few minutes. After starting, give the user the link immediately — progress and results render there live — and check get_research_run only when the user asks or after ~10 minutes. Do not busy-poll. A 'concurrency' error means that analysis already has 3 runs in flight for this account — wait for one to finish, don't retry in a loop. Prior Art follow-ups: when a finished run left limitations unmapped, get_research_run's next_steps carries a gap_menu. Put that menu to the user, and once they pick, start the follow-up here with search_mode 'gap', parent_run_id, and gap_limitations (their picks, by label) — or search_mode 'broad' with parent_run_id to widen the same search. Either can also carry search_queries: your own 2-6 word search strings for vocabulary the claim does not use — the follow-up searches everything the parent did plus those. The same two-call confirmation applies: the first returns the targeting and required credits, and the second with intake_confirmed:true starts it. Available analyses — prior_art: invalidity prior-art search with element-by-element claim mapping. Args: patent; optional claims ("independent" default; "all" or claim numbers may require additional credits — the intake states the requirement), thorough (deeper, slower search), run_name. disclosure: §112 written-description/enablement support analysis. Args: patent (issued grants only); optional claims ("independent" default, "all", or claim numbers). The intake states the required credits. sequence: finds patent DNA, RNA and protein sequences by homology. Args: patent (searches that patent's own sequence listing; optional seq_ids = SEQ ID NOs to limit it) OR sequence (a raw or FASTA sequence; optional seq_type prot|nuc, cutoff_date YYYY-MM-DD); optional max_hits (default 50). The intake reports account availability and any required credits. chemistry: finds patents by chemical structure. Args: patent (searches that patent's own structures; optional structure_ids like C001) OR smiles (a SMILES string); optional search_type substructure|similarity (default substructure), tanimoto 0.3-1 for similarity, max_hits (default 50, up to 100). The intake reports account availability and any required credits. citations: forward and backward citation analysis of a patent's whole granted family (US continuity, EP and CN/JP/TW counterparts). Args: patent; optional include_ep (default true), include_asia (default true). The intake reports account availability and any required credits. charting: charts named references against a patent's claims element by element, with verbatim evidence and column:line cites. Args: patent; references = 1-25 patent or publication numbers; optional claims (claim numbers like "1,3,5", up to 10; default = independent claims). The intake states the required credits. clearance: screens a product or feature description against active US patents for potentially blocking granted claims (a watch list, not an FTO opinion). Args: product_description (40+ characters; it is processed by Patent Bear's model provider like on the site); optional title. The first call extracts the product's searchable elements and returns them with the required credits; the second call echoes confirm_with (result_id, target_id, title, elements — edited if the user wishes). invention: prior-art search from a plain-text invention disclosure, before any claims exist. Args: disclosure (40+ characters; it is processed by Patent Bear's model provider like on the site); optional cutoff_date YYYY-MM-DD (art published before it; default today), title. The first call extracts the invention's elements and returns them with the required credits; the second call echoes confirm_with (result_id, target_id, title, elements, cpc). First run on an account: the account holder must have accepted the Research terms once (authorizing this connector does that; older connections get a one-click link on the site) — a run refused for that reason says so and carries accept_link.

run_research

Save a court decision to the user's Patent Bear Library — the reader's Save button, so the decision is in their Library under Opinions with its caption cached, optionally with a note. Idempotent: saving again refreshes it (existed:true). Use it as part of any hand-back: every decision you relied on for a brief, chart or memo should be saved (save_opinion) AND its quoted passages highlighted (save_opinion_highlights). Pass decision_id from search_litigation or get_legal_activity. Requires a plan that includes the Library.

save_opinion

Save a drafted definitions section into a matter's notebook as a structured note, and get a Word file where every definition paragraph carries a comment bubble with its provenance: the verbatim source sentences, the patents (linked), how many of the applicant's definitions took that form, and any divergence flag. draft_definitions builds a Definitions section from the dictionary after a confirmation step and returns the same Word file; use save_definitions when the assistant drafted the section itself and the user wants it filed with provenance. Pass matter (id or exact name), title, optional application / assignee / preamble, and entries: one per term with term, definition (the proposed text), sources [{ patent, sentence (verbatim from get_term_definitions), count, of, date }], and optionally divergence, original: true (no prior definition existed), note. Every entry needs sources or original: true. Returns link (the note) and docx_link (the Word download, opened signed-in) — give the user both. Requires a plan that includes the Library.

save_definitions

Save a claim-term construction to the user's Constructions tracker (term → what it means), scoped to a patent and optionally a matter — the same record the reader's "Add construction" creates, listed at /library/constructions and importable into Research analyses. Fields: term (required), construction (required — the proposed meaning), claims (numbers), basis (intrinsic/extrinsic support, quote it with col:line cites), type (e.g. plain meaning, means-plus-function), status (proposed default / agreed / court / disputed), matter (name or id). Requires a plan that includes the Library. Never invent a construction: ground it in the specification, claims, and file history the tools return.

save_claim_construction

Save a patent to the account's Patent Bear Library, optionally filing it into a matter (by matter name or matter_id) and attaching a note. Answers "save US10998418B2 to matter Z". If the matter doesn't exist, the save still succeeds and the response says so — offer create_matter, don't silently create one. Requires a plan that includes the Library; otherwise the call returns 403.

save_patent

List the account's saved patents, newest first — optionally only those filed in one matter (by name or matter_id). Each row has the patent id, cached title/assignee, note, and a Patent Bear link.

list_saved_patents

Find US patent applications, pending, granted or abandoned, by applicant, inventor, examiner, art unit or CPC. US applications by what the Office's own record says: applicant, inventor, examiner, art unit, CPC, title words, filing window and status (pending | granted | abandoned | allowed | rejected | appeal). Answers "Apple's pending applications filed since 2025", "what has this examiner allowed", "abandoned applications in art unit 3689". Rows carry the application number, USPTO status text, filing date, applicant, first inventor, publication and patent numbers when they exist, examiner, art unit, entity status, CPC and a link; unpublished applications appear with title and status only. This reads the application record (13 million, refreshed daily), not the full-text index: use it when the question is about status or prosecution, and search_patents / search_patents_bulk when it is about the technology. applicant and inventor match the first-named party by word. Newest filing first; page with offset. Every account.

search_applications

Find patents that disclose or claim a chemical structure. For a compound named in words (a drug name, an IUPAC name), convert it to SMILES and search here: this finds the documents that draw it or name it differently, which a text search misses. Pass smiles, or patent (US or EP) to list the structures its claims cite, then patent + structure_id (a C##### drawing id or SCHEMBL id from that list) to search one. With a patent the search keeps only art published before the patent's earliest priority date; pass before (YYYY-MM-DD) to set another cutoff. mode: similarity (default; tanimoto 0.3 to 1, default 0.7), exact, or substructure (the query as part of larger molecules, closest size first; matched_compounds is a floor when matched_compounds_note says so). Reading a hit: similarity is the Tanimoto score; query_share is the query's heavy atoms as a % of the matched compound's (100 = the query is the whole molecule, small = a small part of a bigger one); locations is where the compound appears (claims, abstract, title, description, image, drawing); claimed_in_reference and reference_claims say whether a claim of the reference cites that drawing (checked for the first 15 hits; absent = not checked). With a cutoff, when nothing predates it the message says how many matching compounds exist ("N matching compounds were found, none published before D"): report that, never a bare no-results. Coverage: EP and WO drawn structures published after 2024-11-13 are text-mined only (image extraction stopped then). offices limits results to US, EP and/or WO documents. WO hits link to the WIPO Patentscope record. Pro and Firm accounts search without limit. Free and Plus accounts share one lifetime allowance of 10 Sequence and Chemistry searches with the site; listing a patent's claimed structures does not count. One call is one request.

search_structures

Search 13 million examiner office actions (non-final and final rejections, July 2007 to the current week) by the examiner's own words and by rejection type. Answers: how has this examiner or art unit applied § 101 to claims like these; what does a final rejection in tech center 3600 citing "prima facie" obviousness look like; every action on this application or patent; how often 112 rejections appear in an art unit, by year. query = words (all must appear) or a "quoted phrase" from the action's text; filters: rejection_type (101 | 102 | 103 | 112 | DoublePatent | Other), tech_center, art_unit, examiner_id, document_code (CTNF non-final | CTFR final), year, application, patent. Rejection types are read from the action text and joined with the USPTO's own tags (which miss most § 101, § 102 and § 112 rejections); rejection_types_uspto keeps the USPTO's alone. Hits carry an excerpt around the match, stated_rejections (each rejection as the action states it: statute such as 102(a)(1) or 112(b), and the claims it names), examiner / art unit / tech center, action_link (the action on Patent Bear), and facets by rejection type, tech center and year; include_text: true returns the full action. Newest first, limit up to 50, offset to page. Examiner side only: the applicant's response is in the file wrapper — get_file_history on the application pairs them. Quote the examiner verbatim. Every account.

search_office_actions

Find patents and published applications whose sequence listings contain a nucleotide or protein sequence, and whether their claims cite it. Pass sequence (residues or FASTA), or patent (US or EP) to list the SEQ ID NOs its claims cite, then patent + seq_id to search one of them. With a patent the search keeps only art published before the patent's earliest priority date (the whole priority chain, provisionals included); pass before (YYYY-MM-DD) to set another cutoff. Results are grouped by patent family (US, EP and WO members joined through their priority chains); each family's best hit is checked against the reference's own listing and claims. Reading a hit: verification exact = identical; contained = the query sits verbatim inside a longer reference sequence (contained_at, target_length); differs = aligned but not identical (identity is % over the aligned span); unchecked = not verified, never report it as a match; unavailable = the reference's listing could not be read, so identity is the search engine's alignment, unconfirmed. A small target_coverage means the query is a small piece of a much larger molecule: say so rather than calling it a 100% match. listing_size of 100 or more suggests a laundry list that names many sequences without singling this one out. claimed_in_reference and reference_claims say whether a claim of the reference cites that SEQ ID NO (US and EP; WO claims are not checked). kmer_coverage (DNA, unchecked hits only) is a screening share, not identity. Short queries (nucleotide under 30, protein under 20 residues) count only verbatim containment, and no result is not a clear result; the response's warnings say so. Coverage: US and EP sequence listings, with WO family members; EP listings about 94% complete since 2010 (the rest are mostly Euro-PCT filings whose listing was never published in text); DNA has no gapped alignment yet. offices limits results to US, EP and/or WO documents. WO hits link to the WIPO Patentscope record. Pro and Firm accounts search without limit. Free and Plus accounts share one lifetime allowance of 10 Sequence and Chemistry searches with the site; listing a patent's claimed sequences does not count. One call is one request.

search_sequences

Search Patent Bear patents, published applications, and technical literature. HOW TO SEARCH: query = the technology, in the words a specification would use ("three-dimensional environment", not "AR"); owner = the company or person (matched on assignee AND applicant — a published application usually names only the applicant, so a company name in the query or in search_fields misses most pending filings); cpc = a classification prefix to widen a topic search past its vocabulary; inventor = a name; date_from / date_to for the window. Every query word must appear in the same record, so ask the same thing two or three ways ("augmented reality", then "mixed reality", then the CPC) and say how each was searched. num_hits is the true total; facets.by_cpc and by_assignee show where the hits sit. For a sweep or a count, search_patents_bulk. US records by default; office:'ep' searches the European corpus (8.1 million EP grants and publications in full text, 1980 to today) and office:'all' searches US and EP together — use them whenever the user asks about European patents, foreign or worldwide art, or gives an EP number. An EP number as the query (EP1674995A2) scopes itself to EP. For exact US records, query can be a full identifier like US12574394B2 or US20030000656A1; the service also accepts numeric patent/publication numbers. Use source_type:'patent' or version:'B2' for issued grants, and source_type:'publication' or version:'A1' for published applications when the query is numeric only. Use publication_date, publication_date_from/publication_date_to, or date_from/date_to for publication-date windows, and filed_from/filed_to for filing dates. Words are matched as written, not stemmed: detect* matches detect, detects, detection and detecting; put a phrase in quotes ("cleaner head"). Returned source_url values point to Patent Bear records. One call is one request against the monthly quota whether it returns 5 hits or 25, so ask for 15-25 on the first call (max_hits) rather than paging in small steps. Results come a page at a time: num_hits is the TOTAL that matched, returned is this page, and next_offset is the offset for the next page. Always tell the user the total and that more are available, and offer the next page rather than presenting one page as everything; more_note gives the exact call, and says when a set is too large to page and should be narrowed. sort chooses the order: relevance (default, keyword relevance), newest or oldest by publication date (a US grant's publication date is its issue date) — pick newest for "latest filings", oldest for "earliest disclosure". For a landscape or any request for more than a few pages, use search_patents_bulk (hundreds of compact rows per call). Each hit includes title and abstract (verbatim text suitable for quoting) plus metadata; articles also include journal and authors. Snippet fragments from the claims and specification are included when they match. For full claim and description text, call get_patent_record with the hit id (US and EP ids both open); get_figures returns the drawings.

search_patents

The same search as search_patents (same query, owner, inventor, cpc, dates and office arguments — put the company in owner, not the query), returned as a long compact list instead of a page of cards: for landscapes, portfolio sweeps, "everything on X since 2015", counting and grouping — any time the user wants the field rather than the top few. One call returns up to 1,000 rows. detail sets the row and the ceiling: ids (each row one short string, the id then the publication date; up to 1,000 rows), brief (adds title, owner and top CPC codes; up to 150; the default), snippets (adds the best matching passage; up to 100). sort = relevance (default), newest or oldest by publication date. office = us (default), ep or all. Same query syntax, date window, search_fields, source_type and version arguments as search_patents. Every response gives total_hits (everything that matched), returned, next_offset and bulk_note: always tell the user the total, and offer the next block or a narrower search rather than presenting one block as the whole field. facets (counts by assignee, CPC, kind and year over the WHOLE matched set) come with the first block, so you can describe a set of thousands without fetching it. Metering: one request per 25 rows returned (100 rows = 4 requests, 1,000 = 40), charged on what actually comes back; requests_charged says how many. Say so before a large pull when the remaining quota is low. Rows are a working list: read the ones that matter with get_patent_record and find_in_patent before quoting or characterizing them.

search_patents_bulk

Find the patents whose claims read closest to a patent's independent claims, or to any text — a claim, one element, a sentence describing the idea. Nearest-neighbour search over 47.6 million claim embeddings (bge-small-en-v1.5), collapsed to the patent whose best claim matches, with title, assignee, date and a link for each hit. Pass patent (US grant or publication; its own claims are excluded) or text (text wins if both). record = grant | pub narrows to one kind; limit up to 50 (default 20). No date cutoff and no CPC filter — it finds neighbours to read, not prior art: for art against a claim with a priority date and element-by-element evaluation, run the Prior Art analysis (run_research prior_art). Every account.

find_similar_patents

Write a structured Overview or Claim Summary for a patent into the account's Patent Bear Library — it renders on the patent's library page exactly like a Patent Bear-generated analysis, attributed to the connector. AUTHOR the content yourself from the patent's real record (get_patent_record / find_in_patent) — verbatim grounding only, never memory. content must be a JSON object matching the kind's schema (see the kind property). Overwrites any existing analysis of that kind — call get_patent_analysis first if unsure. Requires a plan that includes the Library.

save_patent_analysis

Write a note into a matter's notebook — a case summary, a per-patent prosecution and litigation history, a research memo, a call note. This is where long-form work belongs: highlights live on the patent, a note lives on the matter, and the user opens it at /library/notebook. Pass matter (from list_matters or create_matter), a title, and content in markdown: headings, bullet and numbered lists, bold, links. To revise a note, pass its note_id from list_notes — with append:true to add a section, without it to replace the note. Write real prose with the cites the other tools returned, not a bare outline. Requires a plan that includes the Library.

write_note