Highlyt

Create multiple text-based highlights in one batch (max 10 per request). Efficient for chapter analysis — create highlights in batches of up to 10. If you have more than 10, split into multiple calls. Highlight format: {page_number: int, text: str, comment?: str, color?: str} Colors: amber (#E5A84B) core ideas, sage (#6BBF8A) action items, sky (#5BA3D9) questions.

Create a new highlight in a document. Uses react-pdf-highlighter-extended format: - position: ScaledPosition with boundingRect, rects, pageNumber - content: {text: "..."} or {image: "..."} - highlight_type: "text" or "area" - comment: User's comment Colors should be hex codes (e.g., #FFEB3B for yellow).

Create a link between two highlights to form an insight thread. Supports both same-document and cross-document linking. IMPORTANT — parent/child semantics: - parent_highlight_id = the highlight you are viewing (the "context" highlight) - child_highlight_id = the highlight you are linking TO (the "related" highlight) - link_type describes what the CHILD does to the PARENT: - supports: "child supports parent" (child provides evidence for parent) - contradicts: "child contradicts parent" (child challenges parent) - expands: "child expands parent" (child elaborates on parent) - questions: "child questions parent" (child raises questions about parent) - related: general connection For cross-document links, set target_document_id to the document containing the child highlight. If omitted, both highlights are assumed to be in source_document_id. Example: If highlight A says "GDP grew 5%" and you want to link highlight B "Exports increased 12%" as supporting evidence, then: parent_highlight_id = A (the claim), child_highlight_id = B (the evidence), link_type = "supports" (B supports A).

Create a text-based highlight on a specific page. Use this after reading page content with highlyt_get_page_text. The highlight will appear in the sidebar and be searchable/exportable. Copy the exact text from the page content for the 'text' field.

Delete a cluster summary from the knowledge graph. This removes the summary record but does not affect the highlights or their links.

Delete a highlight permanently. This action cannot be undone.

Delete a link between two highlights. This removes the connection but does not delete the highlights themselves.

Export document highlights for AI context with pagination. Returns up to 'limit' highlights (default 20, max 50) starting from 'offset'. Use has_more and next_offset to paginate through all highlights. Includes document metadata, color meanings, and insight threads relevant to the returned highlights.

Export reviewed references as BibTeX, RIS, or canonical CSL-JSON. This always excludes drafts. Never cite a draft as a verified source.

Export a portable researcher-authored Markdown brief with safe receipts.

List all cross-document highlight links for the current user. Returns links where source_document_id != target_document_id, with metadata about both source and target documents. Useful for exploring connections across your entire library.

Get detailed information about a document. Returns full metadata including book context, reading progress, and organizational information.

Retrieve images (charts, graphs, figures, diagrams, photos) and their surrounding text context from an EPUB or web clip document. WHEN TO USE THIS TOOL: - User asks about a chart, graph, figure, diagram, or image in a document - User wants visual data or illustrations from a book/article - User references "Figure X.X" or any visual element Returns each image's base64 data URI, alt text, dimensions, chapter location, and the paragraph text before/after it so you can describe what the image shows. Use the optional query parameter to filter images by surrounding text context. For PDFs, images are embedded in the rendered pages — use highlyt_get_page_text instead.

Get all links connected to a specific highlight. Returns parent links (where this highlight is referenced by others) and child links (where this highlight references others). Useful for exploring how a specific idea connects to other highlights.

Retrieve the text content of specific pages from a document. Use this to read document content before creating highlights. Returns up to 5 pages per request. If the requested range exceeds 5 pages, use has_more and next_page_start to paginate through remaining pages.

Get one reviewed research reference. Never cite a draft without explicit permission.

Read a researcher-authored brief with verified, reviewed source receipts. Use this as context only. Cite only the included receipts. The tool does not reveal a draft reference or proposed evidence as support for a claim.

List all cluster summaries for a document's knowledge graph. Returns each summary with its highlight_ids array — the highlights that were in the cluster when the summary was last generated. DETECTING STALE SUMMARIES: To check if summaries need updating after adding new links: 1. Call highlyt_list_highlight_links to get all current links 2. Compute connected components (group highlights by connectivity) 3. Call this tool to get stored summaries 4. For each summary, check if its highlight_ids is a SUBSET of a current component - If highlight_ids == component_ids → summary is current - If highlight_ids ⊂ component_ids → summary is STALE (cluster grew) - Set difference (component_ids - highlight_ids) = newly added highlights 5. Regenerate stale summaries with highlyt_upsert_cluster_summary, passing the FULL current component highlight_ids

List PDF documents in your Highlyt library. Returns documents sorted by last update time with pagination. Filter by shelf_id to see documents in a specific collection.

List highlight links (insight threads) in a document with pagination. Returns up to 'limit' links (default 50, max 100) starting from 'offset'. Use has_more and next_offset to paginate through all links. Each link has a type: 'related', 'supports', 'contradicts', 'expands', or 'questions'.

List highlights in a document with pagination. Returns up to 'limit' highlights (default 20, max 50) starting from 'offset'. Use has_more and next_offset to paginate through all highlights. Each highlight includes the color meaning (e.g., "Key concepts", "Questions").

List reviewed references, optionally in one project. Drafts are excluded unless include_drafts is explicitly true. Never cite a draft as a verified source.

List the caller's research projects for a bounded AI research task.

Full-text search across your entire library — documents, highlights, and page content. This searches the actual text content of your PDFs (every page), highlight text/comments, and document metadata (title, author, keywords). Results are ranked by relevance with text snippets showing where matches were found. Supports natural language queries and websearch syntax: - Simple: "machine learning" - Exact phrase: '"neural network"' - OR: "neural OR deep" - Exclude: "learning -supervised" Each result includes match_type ('content', 'highlight', or 'metadata'), the page number where the match was found, and a snippet with **bold** markers around matching terms.

Search highlights across your library. Finds highlights containing the query text. Each result includes the color meaning based on the document's definitions. Optionally limit to a specific document.

Update a highlight's color or comment. Only provided fields will be updated.

Update a highlight link's relationship type or note. Only provided fields will be updated.

Create or update a summary for a connected component cluster in the knowledge graph. A cluster is a group of linked highlights that form a connected component. IMPORTANT: - highlight_ids must be the COMPLETE, CURRENT set of highlight UUIDs in the cluster (not just newly added ones). The array is sorted internally for matching. - If a summary already exists for this document_id + highlight_ids combination, it will be UPDATED (upsert behavior). - When updating a stale summary (cluster grew due to new links), pass the FULL updated highlight_ids array so the stored record reflects the current cluster. - The summary should describe thematic connections across ALL highlights in the cluster, including any newly added ones. Typical workflow: 1. Read highlights with highlyt_list_highlights 2. Read links with highlyt_list_highlight_links 3. Compute connected components from links 4. For each component, generate a summary covering all highlights 5. Call this tool with the component's highlight_ids and generated summary