@absolutejs/rag 0.14.0 → 0.15.1

package/CHANGELOG.md

@@ -6,6 +6,18 @@ This file is generated by `absolute-changelog` from the entries in
 `changelog/`. Edit an entry, not this file — and add new ones under
 `changelog/unreleased/`.
 
+## 0.15.1 — 2026-09-20
+
+### Fixed
+
+- **Preserve all sections of public websites instead of extracting one article fragment, and retain labeled image evidence in source-bound documents.** (`readRAGWebpage`, `readRAGWebsite`)
+
+## 0.15.0 — 2026-09-18
+
+### Added
+
+- **Expose bounded website documents bound to source IDs and compact inline citation links for native MCP evidence presentation.** (`readRAGWebsite`)
+
 ## 0.14.0 — 2026-09-18
 
 ### Added

package/README.md

@@ -191,3 +191,9 @@ evidence text. Failed pages and unvisited links are not citable source entries.
 `citationRequirements` asks consumers to place links beside supported claims,
 avoid unsupported exclusions or ownership relationships, and keep verified brand
 changes brief. Host applications still control their generated final answers.
+
+`kind: "website_evidence"` identifies a website research result. `documents`
+binds each bounded text excerpt to a source ID, while source entries provide
+compact numbered `inlineCitation` links. This lets adapters present source-local
+evidence without dumping raw retrieval metadata into a model's response context.
+The combined document text respects `maxChars`; truncated excerpts are marked.

package/changelog.json

@@ -2,6 +2,33 @@
 	"contract": 1,
 	"name": "@absolutejs/rag",
 	"releases": [
+		{
+			"changes": [
+				{
+					"kind": "fixed",
+					"summary": "Preserve all sections of public websites instead of extracting one article fragment, and retain labeled image evidence in source-bound documents.",
+					"symbols": [
+						"readRAGWebpage",
+						"readRAGWebsite"
+					]
+				}
+			],
+			"date": "2026-09-20",
+			"version": "0.15.1"
+		},
+		{
+			"changes": [
+				{
+					"kind": "added",
+					"summary": "Expose bounded website documents bound to source IDs and compact inline citation links for native MCP evidence presentation.",
+					"symbols": [
+						"readRAGWebsite"
+					]
+				}
+			],
+			"date": "2026-09-18",
+			"version": "0.15.0"
+		},
 		{
 			"changes": [
 				{

package/dist/src/web/text.d.ts

@@ -0,0 +1,2 @@
+/** Website research needs the whole page, not the highest-scoring article fragment. */
+export declare const websiteHtmlText: (html: string) => string;

package/dist/src/web/website.d.ts

@@ -59,10 +59,18 @@ export declare const readRAGWebsite: (options: ReadWebpageOptions & {
     redirects: import("./transport").WebRedirect[];
     evidence: import("./evidence").WebEvidence;
     limitations: string[];
+    kind: "website_evidence";
+    documents: {
+        sourceId: string;
+        text: string;
+        truncated: boolean;
+    }[];
     sources: {
+        id: string;
         title: string;
         url: string;
         citation: string;
+        inlineCitation: string;
     }[];
     citationRequirements: string;
 }>;

package/dist/web/index.js

@@ -11256,6 +11256,34 @@ var hasIncompleteAppContent = (html) => {
   return text.length < 100 || /^(?:loading|please wait)(?:[.\s]|$)/iu.test(text);
 };
 
+// src/web/text.ts
+var websiteHtmlText = (html) => {
+  const { document } = parseHTML(html);
+  document.querySelectorAll("script,style,template,noscript,svg,[hidden],[aria-hidden=true]").forEach((element) => element.remove());
+  const root = document.body?.textContent?.trim() ? document.body : document.documentElement;
+  const parts = [];
+  const blocks = /^(?:ADDRESS|ARTICLE|ASIDE|BLOCKQUOTE|BR|DIV|DL|DT|DD|FIGCAPTION|FIGURE|FOOTER|H[1-6]|HEADER|HR|LI|MAIN|NAV|OL|P|SECTION|TABLE|TD|TH|TR|UL)$/u;
+  const walk = (node2) => {
+    if (node2.nodeType === 3) {
+      parts.push((node2.textContent ?? "").replace(/\s+/gu, " "));
+      return;
+    }
+    const block = blocks.test(node2.nodeName);
+    if (block)
+      parts.push(`
+`);
+    node2.childNodes.forEach(walk);
+    if (block)
+      parts.push(`
+`);
+  };
+  if (root)
+    walk(root);
+  return parts.join("").split(`
+`).map((line) => line.trim()).filter(Boolean).join(`
+`);
+};
+
 // src/ingestion/ingestion.ts
 import { readdir, readFile } from "fs/promises";
 import { basename, extname, join, relative, resolve } from "path";
@@ -17786,11 +17814,41 @@ var readRAGWebsite = async (options) => {
       });
     }
   }
-  const sources = pages.filter((page) => page.status !== "error" && page.text).map((page) => {
+  const sources = pages.filter((page) => page.status !== "error" && page.text).map((page, index) => {
     const title = (page.title || new URL(page.finalUrl).hostname).replace(/\s+/gu, " ").trim();
     const label = title.replace(/[\\[\]]/gu, "\\$&");
     const href = page.finalUrl.replace(/[<>]/gu, encodeURIComponent);
-    return { title, url: page.finalUrl, citation: `[${label}](<${href}>)` };
+    return {
+      id: `source-${index + 1}`,
+      title,
+      url: page.finalUrl,
+      citation: `[${label}](<${href}>)`,
+      inlineCitation: `[${index + 1}](<${href}>)`
+    };
+  });
+  let remainingDocumentChars = maxChars;
+  const documents = sources.flatMap((source) => {
+    const page = pages.find((item) => item.finalUrl === source.url && item.text);
+    if (!page || remainingDocumentChars <= 0)
+      return [];
+    const imageLabels = page.evidence.images.filter((image) => image.label).map((image) => `${image.context ? `${image.context}: ` : ""}${image.label}`);
+    const fullText = [
+      page.text,
+      ...imageLabels.length ? [
+        "Image labels supplied by the website (not independently verified customer relationships):",
+        ...imageLabels
+      ] : []
+    ].join(`
+`);
+    const text = fullText.slice(0, remainingDocumentChars);
+    remainingDocumentChars -= text.length;
+    return [
+      {
+        sourceId: source.id,
+        text,
+        truncated: page.truncated || text.length < fullText.length
+      }
+    ];
   });
   const evidenceText = pages.filter((page) => page.status !== "error").map((page) => `SOURCE: ${page.finalUrl}
 CITATION: ${sources.find((source) => source.url === page.finalUrl)?.citation ?? page.finalUrl}
@@ -17802,6 +17860,8 @@ ${page.text}`).join(`
   const incomplete = pages.some((page) => page.status !== "ok") || signal.aborted;
   const unvisited = queue.filter((link) => !visited.has(link.url.split("#")[0]));
   return {
+    kind: "website_evidence",
+    documents,
     sources,
     citationRequirements: "The final answer must contain clickable Markdown source links, not just source names or bare domain mentions. Reuse sources[].citation beside the factual paragraph, list or table row it supports. Cite the actual supporting page, not the homepage for everything. Every factual section needs its supporting links. Do not invent sources or use unread links as evidence. Before responding, check that company facts, service descriptions, customer examples and numerical claims have supporting links; retrieve or omit unsupported claims. A list of large customers does not prove smaller customers are excluded. Preserve the brands actually named in the evidence; do not append parent companies, ownership relationships, current-account status, market rankings or rebrand history from memory. Label reasoned inferences as such. A verified brand change needs at most one short, cited sentence unless the user asks for its history.",
     ...first,
@@ -17833,7 +17893,7 @@ ${page.text}`).join(`
 // src/web/index.ts
 var messageFor = (error) => error instanceof Error ? error.message : "Website retrieval failed.";
 var codeFor = (error) => error instanceof WebReadError ? error.code : error instanceof Error && /^(?:AbortError|TimeoutError)$/u.test(error.name) ? "timeout" : "fetch_failed";
-var cleanHtml = (html) => prepareRAGDocument({ text: html, format: "html" }).normalizedText;
+var cleanHtml = websiteHtmlText;
 var titleOf = (html) => {
   const title = /<title\b[^>]*>([\s\S]*?)<\/title>/iu.exec(html)?.[1];
   return title ? prepareRAGDocument({ text: title, format: "html" }).normalizedText : null;
@@ -18002,5 +18062,5 @@ export {
   validatePublicWebUrl
 };
 
-//# debugId=C770BC61DAD3A78A64756E2164756E21
+//# debugId=64305FA75E0E709664756E2164756E21
 //# sourceMappingURL=index.js.map