@eeacms/volto-eea-chatbot 1.0.11 → 1.0.13

package/CHANGELOG.md

@@ -4,7 +4,32 @@ All notable changes to this project will be documented in this file. Dates are d
 
 Generated by [`auto-changelog`](https://github.com/CookPete/auto-changelog).
 
-### [1.0.11](https://github.com/eea/volto-eea-chatbot/compare/1.0.10...1.0.11) - 6 February 2026
+### [1.0.13](https://github.com/eea/volto-eea-chatbot/compare/1.0.12...1.0.13) - 4 March 2026
+
+#### :house: Internal changes
+
+- style: Automated code fix [eea-jenkins - [`8a8c3c4`](https://github.com/eea/volto-eea-chatbot/commit/8a8c3c4172a4f669661378cf1b5a3569d85609e6)]
+- chore: [JENKINSFILE] add package version in sonarqube [valentinab25 - [`535d986`](https://github.com/eea/volto-eea-chatbot/commit/535d986b7adc77743a668bc4ac63f835eef58df3)]
+
+#### :hammer_and_wrench: Others
+
+- update [Miu Razvan - [`04e6c3f`](https://github.com/eea/volto-eea-chatbot/commit/04e6c3f776bb9920b89a014cc97f8e5dbb284a48)]
+- update [Miu Razvan - [`f9e9beb`](https://github.com/eea/volto-eea-chatbot/commit/f9e9beb0676b215a50226a0db8c5be7c540f26ff)]
+- Forward client ip to onyx/llmgw requests, ref #298095 [Miu Razvan - [`e90a672`](https://github.com/eea/volto-eea-chatbot/commit/e90a672273f2220d2cced4ad53c4b3ed3f295691)]
+### [1.0.12](https://github.com/eea/volto-eea-chatbot/compare/1.0.11...1.0.12) - 23 February 2026
+
+#### :house: Internal changes
+
+- style: Automated code fix [eea-jenkins - [`fdcd884`](https://github.com/eea/volto-eea-chatbot/commit/fdcd8848fd4c3f990ca5ea021f407487aebd6010)]
+- chore: [JENKINSFILE] use sonarqube branches [EEA Jenkins - [`3d428d7`](https://github.com/eea/volto-eea-chatbot/commit/3d428d72f32c3d05452b0961c76f5db1c416e05c)]
+
+#### :hammer_and_wrench: Others
+
+- fix tests [Miu Razvan - [`1b08a47`](https://github.com/eea/volto-eea-chatbot/commit/1b08a4760f0f16c0a3c2a5be295196f510b8e4f0)]
+- fix tests [Miu Razvan - [`0e1c2a2`](https://github.com/eea/volto-eea-chatbot/commit/0e1c2a23cf17d0ead25132c126892778a74954e0)]
+- update [Miu Razvan - [`b9d3066`](https://github.com/eea/volto-eea-chatbot/commit/b9d306623b9034eb08fc77781a48e4ca59146e54)]
+- Filter non-verifiable sentences [Miu Razvan - [`988a4c7`](https://github.com/eea/volto-eea-chatbot/commit/988a4c71293fda99ffaf02d9750e137332d2a182)]
+### [1.0.11](https://github.com/eea/volto-eea-chatbot/compare/1.0.10...1.0.11) - 7 February 2026
 
 #### :house: Internal changes
 
@@ -19,7 +44,6 @@ Generated by [`auto-changelog`](https://github.com/CookPete/auto-changelog).
 - fix eslint [Miu Razvan - [`25b2eba`](https://github.com/eea/volto-eea-chatbot/commit/25b2ebaf1ba7eb186285e8fa1e51e625167a7fd7)]
 - improve coverage to 80% [Miu Razvan - [`7bed26f`](https://github.com/eea/volto-eea-chatbot/commit/7bed26fd8ee9622755f43c5bc625e7a60c5cc09c)]
 - update jest snapshots [Miu Razvan - [`e9e0732`](https://github.com/eea/volto-eea-chatbot/commit/e9e073286fd41fc8f22a5eba248a6cc34eda57d0)]
-- Based on tests improve ui + add playwright utility [Miu Razvan - [`62fda9a`](https://github.com/eea/volto-eea-chatbot/commit/62fda9a572e6255db95ff1a40db3a565c7c9476d)]
 ### [1.0.10](https://github.com/eea/volto-eea-chatbot/compare/1.0.9...1.0.10) - 27 January 2026
 
 #### :hammer_and_wrench: Others

package/README.md

@@ -3,16 +3,16 @@
 [![Releases](https://img.shields.io/github/v/release/eea/volto-eea-chatbot)](https://github.com/eea/volto-eea-chatbot/releases)
 
 [![Pipeline](https://ci.eionet.europa.eu/buildStatus/icon?job=volto-addons%2Fvolto-eea-chatbot%2Fmaster&subject=master)](https://ci.eionet.europa.eu/view/Github/job/volto-addons/job/volto-eea-chatbot/job/master/display/redirect)
-[![Lines of Code](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot-master&metric=ncloc)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot-master)
-[![Coverage](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot-master&metric=coverage)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot-master)
-[![Bugs](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot-master&metric=bugs)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot-master)
-[![Duplicated Lines (%)](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot-master&metric=duplicated_lines_density)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot-master)
+[![Lines of Code](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot&metric=ncloc)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot)
+[![Coverage](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot&metric=coverage)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot)
+[![Bugs](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot&metric=bugs)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot)
+[![Duplicated Lines (%)](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot&metric=duplicated_lines_density)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot)
 
 [![Pipeline](https://ci.eionet.europa.eu/buildStatus/icon?job=volto-addons%2Fvolto-eea-chatbot%2Fdevelop&subject=develop)](https://ci.eionet.europa.eu/view/Github/job/volto-addons/job/volto-eea-chatbot/job/develop/display/redirect)
-[![Lines of Code](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot-develop&metric=ncloc)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot-develop)
-[![Coverage](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot-develop&metric=coverage)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot-develop)
-[![Bugs](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot-develop&metric=bugs)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot-develop)
-[![Duplicated Lines (%)](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot-develop&metric=duplicated_lines_density)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot-develop)
+[![Lines of Code](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot&branch=develop&metric=ncloc)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot&branch=develop)
+[![Coverage](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot&branch=develop&metric=coverage)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot&branch=develop)
+[![Bugs](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot&branch=develop&metric=bugs)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot&branch=develop)
+[![Duplicated Lines (%)](https://sonarqube.eea.europa.eu/api/project_badges/measure?project=volto-eea-chatbot&branch=develop&metric=duplicated_lines_density)](https://sonarqube.eea.europa.eu/dashboard?id=volto-eea-chatbot&branch=develop)
 
 [Volto](https://github.com/plone/volto) add-on that integrates an AI-powered chatbot with a customizable interface and advanced settings to tailor its behavior and enhance user interactions.
 

package/jest-addon.config.js

@@ -430,7 +430,7 @@ module.exports = {
       '<rootDir>/node_modules/@plone/volto/jest-addons-loader.js',
   },
   transformIgnorePatterns: [
-    '/node_modules/(?!(@plone|@root|@package|@eeacms)/).*/',
+    '/node_modules/(?!(@plone|@root|@package|@eeacms|compromise|efrt|grad-school|suffix-thumb)/).*/',
   ],
   transform: {
     '^.+\\.js(x)?$': 'babel-jest',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@eeacms/volto-eea-chatbot",
-  "version": "1.0.11",
+  "version": "1.0.13",
   "description": "@eeacms/volto-eea-chatbot: Volto add-on",
   "main": "src/index.js",
   "author": "European Environment Agency: IDM2 A-Team",
@@ -43,6 +43,7 @@
     "@eeacms/volto-matomo": "*",
     "@microsoft/fetch-event-source": "2.0.1",
     "@plone-collective/volto-sentry": "*",
+    "compromise": "14.14.5",
     "fast-json-patch": "3.1.1",
     "highlight.js": "11.10.0",
     "luxon": "3.5.0",

package/src/ChatBlock/chat/AIMessage.tsx

@@ -67,28 +67,30 @@ function addQualityMarkersPlugin() {
 }
 
 export function addHalloumiContext(doc: any, text: string) {
-  const updatedDate = doc.updated_at
-    ? new Date(doc.updated_at).toLocaleString('en-GB', {
-        year: 'numeric',
-        month: 'long',
-        day: '2-digit',
-        hour: '2-digit',
-        minute: '2-digit',
-      })
-    : '';
-
-  const docIndex = doc.index ? `DOCUMENT ${doc.index}: ` : '';
-  const sources: any = { web: 'Website', file: 'File' };
-
-  const sourceType = doc.source_type
-    ? sources[doc.source_type] || capitalize(doc.source_type)
-    : '';
-
-  const header = `${docIndex}${doc.semantic_identifier}${
-    sourceType ? `\nSource: ${sourceType}` : ''
-  }${updatedDate ? `\nUpdated: ${updatedDate}` : ''}`;
-
-  return `${header}\n${text}`;
+  // TODO: CLEAN UP
+  // const updatedDate = doc.updated_at
+  //   ? new Date(doc.updated_at).toLocaleString('en-GB', {
+  //       year: 'numeric',
+  //       month: 'long',
+  //       day: '2-digit',
+  //       hour: '2-digit',
+  //       minute: '2-digit',
+  //     })
+  //   : '';
+
+  // const docIndex = doc.index ? `DOCUMENT ${doc.index}: ` : '';
+  // const sources: any = { web: 'Website', file: 'File' };
+
+  // const sourceType = doc.source_type
+  //   ? sources[doc.source_type] || capitalize(doc.source_type)
+  //   : '';
+
+  // const header = `${docIndex}${doc.semantic_identifier}${
+  //   sourceType ? `\nSource: ${sourceType}` : ''
+  // }${updatedDate ? `\nUpdated: ${updatedDate}` : ''}`;
+
+  // return `${header}\n${text}`;
+  return text.replace(/\u00A0/g, ' ');
 }
 
 function mapToolDocumentsToText(message: any) {
@@ -142,11 +144,13 @@ function getContextSources(
       );
 }
 
-function getScoreDetails(claims: any, qualityCheckStages: any) {
+function getScoreDetails(rawClaims: any, qualityCheckStages: any) {
+  const claims = rawClaims.filter((claim: any) => !claim.skipped);
   const score = (
     (claims.length > 0
-      ? claims.reduce((acc: any, { score }: any) => acc + score, 0) /
-        claims.length
+      ? claims
+          .filter((claim: any) => !claim.skipped)
+          .reduce((acc: any, { score }: any) => acc + score, 0) / claims.length
       : 1) * 100
   ).toFixed(0);
 
@@ -165,6 +169,7 @@ function getScoreDetails(claims: any, qualityCheckStages: any) {
 
 export function AIMessage({
   message,
+  prevMessage,
   isLoading,
   libs,
   onChoice,
@@ -283,6 +288,7 @@ export function AIMessage({
   );
 
   const claims = markers?.claims || [];
+  const emptyClaims = markers?.empty || false;
   const { score, scoreStage, scoreColor, isFirstScoreStage } = getScoreDetails(
     claims,
     qualityCheckStages,
@@ -434,6 +440,7 @@ export function AIMessage({
           }}
           showVerifyClaimsButton={showVerifyClaimsButton}
           retryHalloumi={retryHalloumi}
+          emptyClaims={emptyClaims}
         />
       )}
 

package/src/ChatBlock/chat/ChatWindow.tsx

@@ -185,6 +185,7 @@ function ChatWindow({
                 <React.Fragment>
                   <ChatMessage
                     key={message.messageId}
+                    prevMessage={messages[index - 1]}
                     message={message}
                     isLoading={isStreaming}
                     isDeepResearchEnabled={isDeepResearchEnabled}

package/src/ChatBlock/components/HalloumiFeedback.jsx

@@ -72,8 +72,10 @@ const HalloumiFeedback = ({
   showVerifyClaimsButton,
   sources,
   retryHalloumi,
+  emptyClaims,
 }) => {
-  const noClaimsScore = markers?.claims[0]?.score === null;
+  const claims = (markers?.claims || []).filter((claim) => !claim.skipped);
+  const noClaimsScore = claims[0]?.score === null;
   const messageBySource =
     'Please allow a few minutes for claim verification when many references are involved.';
 
@@ -98,7 +100,7 @@ const HalloumiFeedback = ({
 
       {noClaimsScore && (
         <>
-          <Message color="red">{markers?.claims?.[0].rationale}</Message>
+          <Message color="red">{claims[0].rationale}</Message>
           <Button onClick={retryHalloumi} className="icon">
             <SVGIcon name={RotateIcon} /> Retry Fact-check AI answer
           </Button>
@@ -110,12 +112,14 @@ const HalloumiFeedback = ({
           color={scoreColor}
           className={cx(
             'claim-message',
-            getSupportedBgColor(score / 100, 'claim'),
+            emptyClaims
+              ? 'claim-empty claim-gray-500'
+              : getSupportedBgColor(score / 100, 'claim'),
           )}
           icon
         >
           <MessageContent>
-            {printSlate(halloumiMessage, `${score}%`)}
+            {emptyClaims || printSlate(halloumiMessage, `${score}%`)}
           </MessageContent>
         </Message>
       )}

package/src/ChatBlock/components/markdown/ClaimModal.jsx

@@ -30,7 +30,7 @@ const trimNonAlphanumeric = (str) =>
   stripMarkdown(str).replace(/(?:^[^a-zA-Z0-9]+)|(?:[^a-zA-Z0-9]+$)/g, '');
 
 export function ClaimModal({ claim, markers, text, citedSources }) {
-  const highlightText = trimNonAlphanumeric(text?.[0] || '');
+  const highlightText = trimNonAlphanumeric(text || '');
 
   return (
     <Modal

package/src/ChatBlock/components/markdown/ClaimSegments.jsx

@@ -10,9 +10,8 @@ const VISIBLE_SEGMENTS = 50; // Number of citations to show by default
 
 export function ClaimSegments({ segmentIds, segments, citedSources }) {
   const joinedSources = citedSources.reduce((acc, source) => {
-    source.startIndex = acc.length ? acc.length + 1 : 0;
-    const sep = acc ? '\n' : '';
-    return acc + sep + source.halloumiContext; // + '\n---\n';
+    source.startIndex = acc.length;
+    return acc + source.halloumiContext;
   }, '');
 
   const snippets = (segmentIds || [])

package/src/ChatBlock/components/markdown/RenderClaimView.jsx

@@ -19,7 +19,7 @@ export const RenderClaimView = (props) => {
 
   sortedSegments.forEach((segment) => {
     const segmentStart = segment.startOffset - sourceStartIndex;
-    const segmentEnd = segment.endOffset - sourceStartIndex;
+    const segmentEnd = segment.endOffset - sourceStartIndex + 1;
 
     // Add the text part before the current segment
     if (segmentStart > lastIndex) {

package/src/ChatBlock/components/markdown/index.js

@@ -33,31 +33,57 @@ export function components(message, markers, citedSources) {
       return <td {...rest}>{processedChildren}</td>;
     },
     span: (props) => {
-      const { node, ...rest } = props;
+      const { node, children } = props;
       const child = node.children[0];
-      let claim;
 
       // identifies if the current text belongs to a claim
       if (child.type === 'text' && child.position && markers) {
+        const text = child.value || '';
         const start = child.position.start.offset;
         const end = child.position.end.offset;
-        claim = markers.claims?.find(
+        const claims = markers.claims?.filter(
           (claim) =>
-            (start >= claim.startOffset && end <= claim.endOffset) ||
-            (claim.startOffset >= start && end <= claim.endOffset),
+            claim.score !== null &&
+            ((start >= claim.startOffset && end <= claim.endOffset) ||
+              (start <= claim.endOffset && end >= claim.endOffset) ||
+              (start <= claim.startOffset && end >= claim.startOffset)),
         );
+
+        if (claims && claims.length > 0) {
+          let relStart = 0;
+          const claimsSegments = claims.map((claim) => ({
+            claim,
+            start: Math.max(0, claim.startOffset - start),
+            end: Math.min(text.length, claim.endOffset - start),
+          }));
+          const segments = claimsSegments.reduce((acc, segment) => {
+            if (relStart < segment.start) {
+              acc.push(child.value.substring(relStart, segment.start));
+            }
+            const claimText = child.value.substring(segment.start, segment.end);
+            acc.push(
+              <ClaimModal
+                claim={segment.claim}
+                markers={markers}
+                text={claimText}
+                citedSources={citedSources}
+              />,
+            );
+            relStart = segment.end;
+            return acc;
+          }, []);
+
+          if (relStart < text.length) {
+            segments.push(text.substring(relStart));
+          }
+
+          return segments;
+        }
+
+        return text;
       }
 
-      return !claim || claim?.score === null ? (
-        rest.children || []
-      ) : (
-        <ClaimModal
-          claim={claim}
-          markers={markers}
-          text={rest.children}
-          citedSources={citedSources}
-        />
-      );
+      return children || [];
     },
     a: (props) => {
       const { node, children, href, ...rest } = props;

package/src/ChatBlock/hooks/useChatController.ts

@@ -22,21 +22,6 @@ function extractRelatedQuestions(str: string): RelatedQuestion[] {
     throw new Error('Related questions were not generated properly');
   }
 
-  const regex = /\[[\s\S]*?\]/;
-  const match = str.match(regex);
-
-  if (match) {
-    try {
-      return JSON.parse(match[0]);
-    } catch {
-      // Fallback to line-by-line parsing
-      return str
-        .split('\n')
-        .filter((line) => line.trim())
-        .map((question) => ({ question }));
-    }
-  }
-
   return str
     .split('\n')
     .filter((line) => line.trim())

package/src/ChatBlock/hooks/useQualityMarkers.js

@@ -54,17 +54,6 @@ export function useQualityMarkers(
         return;
       }
 
-      // // console.log('Halloumi sources:', sources.length, sources);
-      // if (sources.length > 40) {
-      //   // eslint-disable-next-line no-console
-      //   console.warn(
-      //     `Warning: Too many sources (${sources.length}). Skipping quality control.`,
-      //   );
-      //
-      //   setHalloumiResponse(empty(message, TOOLARGE_RATIONALE));
-      //   return;
-      // }
-
       setIsLoading(true);
 
       try {

package/src/ChatBlock/style.less

@@ -580,6 +580,53 @@ mark {
       background-color: @grey;
     }
   }
+
+  // Inline code
+  code {
+    padding: 0.15em 0.4em;
+    border-radius: 3px;
+    background-color: #eef0f2;
+    color: @grey;
+    font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace;
+    font-size: 0.875em;
+    word-break: break-word;
+  }
+
+  // Code blocks
+  pre {
+    padding: 1em 1.25em;
+    border-radius: @borderRadius;
+    margin: 0 0 @marginBottom;
+    background-color: #1e2a35;
+    color: #e2e8f0;
+    font-size: 0.875rem;
+    line-height: 1.65;
+    overflow-x: auto;
+    scrollbar-color: rgba(255, 255, 255, 0.2) transparent;
+    scrollbar-width: thin;
+
+    &::-webkit-scrollbar {
+      height: 6px;
+    }
+
+    &::-webkit-scrollbar-track {
+      background: transparent;
+    }
+
+    &::-webkit-scrollbar-thumb {
+      border-radius: 3px;
+      background-color: rgba(255, 255, 255, 0.2);
+    }
+
+    // Reset inline code inside blocks
+    code {
+      padding: 0;
+      background-color: transparent;
+      color: inherit;
+      font-size: inherit;
+      word-break: normal;
+    }
+  }
 }
 
 .circle {

package/src/ChatBlock/types/interfaces.ts

@@ -121,6 +121,7 @@ export interface ToolCallMetadata {
 
 export interface ChatMessageProps {
   message: Message;
+  prevMessage?: Message;
   isLoading: boolean;
   isDeepResearchEnabled?: boolean;
   libs?: any;

package/src/halloumi/filtering.js

@@ -0,0 +1,149 @@
+import debug from 'debug';
+import fetch from 'node-fetch';
+
+const log = debug('halloumi');
+
+const LLMGW_URL = process.env.LLMGW_URL;
+const LLMGW_API_KEY = process.env.LLMGW_TOKEN;
+const MIN_CONTEXT_SENTENCES_FOR_FILTERING = 75;
+
+const filterModel = {
+  name: 'Inhouse-LLM/gpt-oss-120b',
+  apiUrl: `${LLMGW_URL}/chat/completions`,
+  apiKey: LLMGW_API_KEY,
+};
+
+export async function callLLM(apiUrl, apiKey, requestBody, { ip } = {}) {
+  const headers = {
+    'Content-Type': 'application/json',
+    accept: 'application/json',
+  };
+  if (apiKey) {
+    headers['Authorization'] = `Bearer ${apiKey}`;
+  }
+  if (ip) {
+    headers['X-Forwarded-For'] = ip;
+  }
+
+  const response = await fetch(apiUrl, {
+    method: 'POST',
+    headers,
+    body: JSON.stringify(requestBody),
+  });
+
+  return response.json();
+}
+
+function buildClaimFilterPrompt(sentences) {
+  const numberedSentences = sentences
+    .map((s, i) => `${i + 1}. "${s.trim()}"`)
+    .join('\n');
+
+  return `Given the following numbered sentences, identify which ones are NOT verifiable factual claims.
+
+Exclude sentences that are:
+- Greetings, preambles, or transitional phrases
+- Opinions, vague statements, or subjective assessments
+- Introductory sentences that set up a list or table
+- Sentences that merely restate or paraphrase the question
+- Conversational closers (e.g., "let me know if you need more")
+- Broad summaries that don't assert a specific fact
+- Disclaimers or meta-commentary about the response itself
+- Sentences that do not provide enough information to verify
+- Sentences that only mention a concept without providing specific information
+
+Respond with ONLY a comma-separated list of the sentence numbers to exclude. If none should be excluded, respond with "NONE".
+
+Sentences:
+${numberedSentences}`;
+}
+
+function buildContextFilterPrompt(contextSentences, claimSentences) {
+  const numberedContext = contextSentences
+    .map((s, i) => `${i + 1}. "${s.trim()}"`)
+    .join('\n');
+  const numberedClaims = claimSentences
+    .map((s, i) => `${i + 1}. "${s.trim()}"`)
+    .join('\n');
+
+  return `Given the following context sentences and claims, identify which context sentences are NOT relevant to verifying ANY of the claims.
+
+Context sentences:
+${numberedContext}
+
+Claims to verify:
+${numberedClaims}
+
+Respond with ONLY a comma-separated list of the context sentence numbers that are NOT relevant. If all are relevant, respond with "NONE".`;
+}
+
+/**
+ * Parses a comma-separated list of indices from an LLM response.
+ */
+export function parseExcludeIndices(content, maxIndex) {
+  const excludeIndices = new Set();
+  if (content.trim().toUpperCase() === 'NONE') {
+    return excludeIndices;
+  }
+  const matches = content.match(/\d+/g) || [];
+  for (const match of matches) {
+    const idx = parseInt(match, 10);
+    if (idx >= 1 && idx <= maxIndex) {
+      excludeIndices.add(idx);
+    }
+  }
+  return excludeIndices;
+}
+
+async function callFilterModel(prompt, { ip } = {}) {
+  const data = {
+    messages: [{ role: 'user', content: prompt }],
+    temperature: 0.0,
+    model: filterModel.name,
+  };
+  const jsonData = await callLLM(filterModel.apiUrl, filterModel.apiKey, data, {
+    ip,
+  });
+  return jsonData.choices?.[0]?.message?.content || '';
+}
+
+export async function excludeClaimSentences(sentences, { ip } = {}) {
+  if (sentences.length === 0) {
+    return new Set();
+  }
+
+  try {
+    const prompt = buildClaimFilterPrompt(sentences);
+    const content = await callFilterModel(prompt, { ip });
+    const excludedIndices = parseExcludeIndices(content, sentences.length);
+    log('Claim filter response', excludedIndices.size);
+    return excludedIndices;
+  } catch (error) {
+    log('Claim filter failed, skipping', error);
+    return new Set();
+  }
+}
+
+export async function excludeContextSentences(
+  contextSentences,
+  claimSentences,
+  { ip } = {},
+) {
+  if (contextSentences.length <= MIN_CONTEXT_SENTENCES_FOR_FILTERING) {
+    return new Set();
+  }
+
+  try {
+    const prompt = buildContextFilterPrompt(contextSentences, claimSentences);
+    const content = await callFilterModel(prompt, { ip });
+    const excludedIndices = parseExcludeIndices(
+      content,
+      contextSentences.length,
+    );
+    log('Context filter response', excludedIndices.size);
+    return excludedIndices;
+  } catch (error) {
+    log('Context filter failed, skipping', error);
+    return new Set();
+  }
+}

package/src/halloumi/filtering.test.js

@@ -0,0 +1,44 @@
+import { parseExcludeIndices } from './filtering';
+
+describe('parseExcludeIndices', () => {
+  it('parses single indices', () => {
+    const result = parseExcludeIndices('1,3,5', 6);
+    expect(result).toEqual(new Set([1, 3, 5]));
+  });
+
+  it('returns empty set for NONE', () => {
+    const result = parseExcludeIndices('NONE', 10);
+    expect(result).toEqual(new Set());
+  });
+
+  it('returns empty set for none (lowercase)', () => {
+    const result = parseExcludeIndices('  none  ', 10);
+    expect(result).toEqual(new Set());
+  });
+
+  it('ignores indices below 1', () => {
+    const result = parseExcludeIndices('0, 1, 3', 5);
+    expect(result).toEqual(new Set([1, 3]));
+  });
+
+  it('ignores indices above maxIndex', () => {
+    const result = parseExcludeIndices('1, 3, 99', 5);
+    expect(result).toEqual(new Set([1, 3]));
+  });
+
+  it('handles whitespace variations', () => {
+    const result = parseExcludeIndices('  1 ,  5 , 7  ', 10);
+    expect(result).toEqual(new Set([1, 5, 7]));
+  });
+
+  it('extracts numbers even from unexpected formats', () => {
+    // Parser uses match(/\d+/g) so it extracts all numbers
+    const result = parseExcludeIndices('1-3, 5', 10);
+    expect(result).toEqual(new Set([1, 3, 5]));
+  });
+
+  it('returns empty set for empty string with no numbers', () => {
+    const result = parseExcludeIndices('no numbers here', 10);
+    expect(result).toEqual(new Set());
+  });
+});