FindIp

Retrieve patent text through the Deep Research-compatible `search` and `fetch` interface. **Input:** Set `id` to a publication or registration number returned by `search`. For a document split across multiple responses, use the `next_id` returned in `metadata` to retrieve the next part. **Response:** Returns `id`, `title`, `text`, `url` (patent-detail link), and `metadata`. Documents of 175,000 UTF-8 bytes or fewer are returned in full. Longer documents are divided into parts of at most 175,000 bytes; `metadata` reports the byte count, current part, total parts, and `next_id`. Metadata also includes applicant names, filing, publication and grant dates, IPC codes, jurisdiction, and kind code.

fetch

Search FindIP's patent index through the Deep Research-compatible `search` and `fetch` interface. **Input:** Set `query` to a natural-language description of a technical topic, problem, or solution. **Response:** Returns `results` with up to 50 relevant patents from KR, US, CN, JP, and EP. Each result contains `id` (publication number), `title`, and `url` (patent-detail link). Use an `id` with `fetch` to retrieve the full patent document.

search

Resolve an applicant name to harmonized applicant candidates for patent search or trend analysis. **Input:** Set `q` to the applicant name. `match_mode` controls whether `q` matches any part (`contains`), the beginning (`prefix`), or the whole indexed applicant name (`exact`). `top_n` limits the number of returned candidates. **Response:** Returns `{query, match_mode, candidates, top1_dominance, recommendation, total_pubs}`. Each `candidates[]` item contains an `entity_id` representative-name string and `pub_count` and `share` ranking signals. Candidates are ordered by `pub_count`. `top1_dominance` is the first candidate's `share`, and `total_pubs` is the sum of the returned candidates' `pub_count`. `recommendation` is `auto` when the first candidate's share is at least 0.9; otherwise it is `ask`. These count and share values are not exact patent-portfolio statistics. **Next step:** Use the relevant returned `entity_id` strings as one `applicant_entity_ids` list in `findip_search_patents` or `findip_analyze_trends`. For `auto`, use the first candidate. For `ask`, select from the user's context and ask only if the intended applicant remains unclear. An empty `candidates` list means no matching applicant name was found.

findip_find_applicant

Report FindIP patent-publication coverage for KR, US, CN, JP, and EP. **Input:** None. **Use when:** Verify whether a jurisdiction or publication-date range is covered by the index. **Response:** `{countries, total, updated_at}`. Each `countries[code]` contains `{count, from, to, last_ingested_at}`: publication count, coverage start year, latest publication date, and the last successful source-ingestion time. `total` is the combined count, and `updated_at` is the coverage refresh time.

findip_get_coverage

Normalize user-provided patent numbers using the country formats in `patent_ids`, then retrieve patent text and metadata. Infer the country and number type from the number format and user context; ask only if the country cannot be determined. If the number type is unclear, query all applicable types and return deduplicated results. Each patent includes `title_link`, a ready-made markdown link to the patent-detail page. Reproduce it verbatim when presenting a patent; never present a title without its link. Found patents return `{title_link, patent_id, url, metadata, content, drawings?, representative_figure?, sequence_count?, sequences?, sequences_truncated?}`. `patent_id` is the resolved publication or registration number, and `content` is markdown. Drawing references do not contain image bytes, and sequence results are limited to 50 per patent. Missing patents return `{patent_id, error: "not_found"}`. `legal_status` indicates the publication stage, not current enforceability.

findip_get_patent_details

Retrieve one patent drawing by publication or registration number and drawing filename. **Input:** Use the `patent_id` and exact drawing filename returned by patent details. **Response:** `base64` returns `[TextContent, ImageContent]` for inline display. `url` returns `{url, publication_number, filename}`.

findip_get_patent_drawing

특허 통합 검색 — 텍스트 시맨틱·생물학적 서열(sequence)·도면 이미지(image_url) 유사도로 5개국(KR·US·CN·JP·EP) 특허 검색. **Purpose** — Unified patent retrieval: text semantic search, drawing-similarity search via `image_url`, and biological sequence-similarity search via `sequence`. For text search, describe the *technical concept* in one sentence; keyword bags work poorly. Query language is independent of result language (multilingual). **Image (drawing-similarity) search** — pass `image_url` to search by how patent DRAWINGS look. `image_url` alone → results ordered by `image_score`, with a `similar_drawings` gallery (renderable `drawing_url`s). `image_url` + `query` → hybrid: visually similar patents are then semantically searched/re-ranked by the text; each result carries `matched_drawing` (which figure matched, how close). **Sequence (bio-sequence) search** — pass `sequence` (raw DNA/RNA/ protein string) to find patents whose sequences resemble it. `sequence` alone → patents with matching sequences. `sequence` + `query` → combine a sequence with a topic. **When the user's message embeds a sequence, SPLIT it: the raw sequence goes in `sequence`, the topic words in `query` — never mix the two.** **Use when** — exploratory topic search, competitive landscaping, prior-art style retrieval. **NOT for** — known publication ID (use `findip_get_patent_details`), pure counts/trends (use `findip_analyze_trends`), image bytes (use `findip_get_patent_drawing`). **Multi-country** — pass all jurisdictions in ONE call (`countries=["KR","US"]`), not separate calls; splitting degrades result quality. **Applicant — MANDATORY `findip_find_applicant` first.** If the user names a company (any language), call `findip_find_applicant(q=name)` → pass `applicant_entity_ids=[…]` (list, ONE call, include near-duplicates like `'SAMSUNG ELECTRONICS CO LTD'` + `'삼성전자주식회사'` + `'삼성전자 주식회사'`). Empty candidates → not indexed (no fallback). **Returns** — `{total, results[]}` where each result has `patent_id`, `title`, `country`, `kind_code`, `legal_status` (`GRANTED`|`PUBLISHED` — dynamic post-grant status NOT reflected), `applicants[]`, `applicants_harmonized[]`, `inventors[]`, `inventors_harmonized[]`, `application_number`, `family_id`, `filing_date`, `publication_date`, `grant_date`, `priority_date`, `ipc_codes[]`, `similarity_score` (loose), `rerank_score` (sort-only, never threshold), `representative_figure` (filename or null), `chunks[]` (matched passages ordered by rerank_score), `url` (public detail page — link each hit to this). **Broad / multi-concept queries drift off-topic.** A query mixing two concepts ("wireless-charging + IR temperature sensor") often latches onto the more common one and drops the other; a short broad term ("injection molding") scatters across unrelated fields. If results look off-topic or the `ipc_codes` across results are scattered/unrelated, narrow with the `ipc_codes` parameter — pick codes from what the results actually show (or call `findip_analyze_trends` for its `top_ipc` rows), NOT from memory. For a genuinely broad query, ask the user which sub-field they mean in **plain technical language (not IPC codes)**, then re-search. **Rendering** — representative figures are shown by DEFAULT. When the client renders an HTML artifact/canvas (Claude=artifact, ChatGPT=canvas), build a gallery/table with each hit's figure base64-embedded; otherwise an inline markdown table (columns: `patent_id` · `title` · `country` · `applicants_harmonized` · `filing_date` · `kind_code` · `legal_status` · `rerank_score`) and still list every hit. Always prepend a 1–3 sentence plain-text summary. When noting why a hit is relevant, cite its matched `chunks[].claim_numbers` / `paragraph_numbers` (e.g. "matched claim 3", "paragraph 174") so the grounding is explicit. **Link each `patent_id`/title to its public detail page** — use each result's `url` field directly (= `https://findip.ai/p/{publication_id}`, server rule 3); in HTML add `target="_blank" rel="noopener"` so it opens in a separate tab. Never link to third-party sites. For inline content use `findip_get_patent_section` / `findip_get_patent_details`. **Drawings** — always include each hit's `representative_figure`. Build the figures into the HTML result base64-embedded via code (see `findip_get_patent_drawing`). Never paste base64 yourself; let code write it to the file. **Related** — `findip_find_applicant` (before company filter), `findip_get_patent_section` / `findip_get_patent_details` (drill in), `findip_analyze_trends` (aggregates), `findip_get_coverage` (date sanity).

findip_search_patents

Retrieve one area of a patent document by patent or application number. **Input:** Normalize the number by jurisdiction and number format, then set `publication_id`. Use `id_type="publication"` for publication or registration numbers and `id_type="application"` for application numbers. If the number type cannot be determined, try both. Select the area with `section`. Use `nums` only to select claim numbers, description paragraph numbers, or `SEQ ID NO` values. **Response:** Returns the requested area as a markdown string. Drawing results contain exact filenames and captions, not image content; use a returned filename with `findip_get_patent_drawing` to retrieve the image. Family results list every equivalent publication, and only members shown with a patent-detail link have retrievable text — a section request for one without a link returns an empty result. Sequence results include sequence counts, metadata, and sequence text.

findip_get_patent_section

Patent statistics across KR, US, CN, JP, and EP. Omit `query` for CENSUS, which returns exact publication counts for applicant, inventor, IPC/CPC, date, or non-`all` document-stage filters. Country alone is not sufficient. Provide `query` for TREND, which returns vector-similarity estimates for a technical topic. TREND totals are not exhaustive, and country counts do not indicate index coverage. On the first page, `results` starts with statistics rows (`by_year`, `by_country`, `top_applicant`, `top_ipc`), followed by patent previews. TREND with applicant, inventor, IPC/CPC, date, or document-stage filters also includes exact `census_*` rows before the TREND rows. Present the statistics visually with suitable tables or charts when possible. Each patent preview includes `title_link`, a ready-made markdown link to the patent-detail page. Reproduce it verbatim when presenting a patent; never present a title without its link. `legal_status` indicates the publication stage, not current enforceability.

findip_analyze_trends