1 files changed, 103 insertions, 0 deletions
diff --git a/src/client/views/nodes/chatbot/tools/WebsiteInfoScraperTool.ts b/src/client/views/nodes/chatbot/tools/WebsiteInfoScraperTool.ts
new file mode 100644
index 000000000..19ccd0b36
--- /dev/null
+++ b/src/client/views/nodes/chatbot/tools/WebsiteInfoScraperTool.ts
@@ -0,0 +1,103 @@
+import { v4 as uuidv4 } from 'uuid';
+import { Networking } from '../../../../Network';
+import { BaseTool } from './BaseTool';
+import { Observation } from '../types/types';
+import { ParametersType, ToolInfo } from '../types/tool_types';
+
+const websiteInfoScraperToolParams = [
+    {
+        name: 'urls',
+        type: 'string[]',
+        description: 'The URLs of the websites to scrape',
+        required: true,
+        max_inputs: 3,
+    },
+] as const;
+
+type WebsiteInfoScraperToolParamsType = typeof websiteInfoScraperToolParams;
+
+const websiteInfoScraperToolInfo: ToolInfo<WebsiteInfoScraperToolParamsType> = {
+    name: 'websiteInfoScraper',
+    description: 'Scrape detailed information from specific websites relevant to the user query. Returns the text content of the webpages for further analysis and grounding.',
+    citationRules: `
+      Your task is to provide a comprehensive response to the user's prompt using the content scraped from relevant websites. Ensure you follow these guidelines for structuring your response:
+
+      1. Grounded Text Tag Structure:
+         - Wrap all text derived from the scraped website(s) in <grounded_text> tags.
+         - **Do not include non-sourced information** in <grounded_text> tags.
+         - Use a single <grounded_text> tag for content derived from a single website. If citing multiple websites, create new <grounded_text> tags for each.
+         - Ensure each <grounded_text> tag has a citation index corresponding to the scraped URL.
+
+      2. Citation Tag Structure:
+         - Create a <citation> tag for each distinct piece of information used from the website(s).
+         - Each <citation> tag must reference a URL chunk using the chunk_id attribute.
+         - For URL-based citations, leave the citation content empty, but reference the chunk_id and type as 'url'.
+
+      3. Structural Integrity Checks:
+         - Ensure all opening and closing tags are matched properly.
+         - Verify that all citation_index attributes in <grounded_text> tags correspond to valid citations.
+         - Do not over-cite—cite only the most relevant parts of the websites.
+
+      Example Usage:
+
+      <answer>
+          <grounded_text citation_index="1">
+          Based on data from the World Bank, economic growth has stabilized in recent years, following a surge in investments.
+          </grounded_text>
+          <grounded_text citation_index="2">
+          According to information retrieved from the International Monetary Fund, the inflation rate has been gradually decreasing since 2020.
+          </grounded_text>
+
+          <citations>
+              <citation index="1" chunk_id="1234" type="url"></citation>
+              <citation index="2" chunk_id="5678" type="url"></citation>
+          </citations>
+
+          <follow_up_questions>
+              <question>What are the long-term economic impacts of increased investments on GDP?</question>
+              <question>How might inflation trends affect future monetary policy?</question>
+              <question>Are there additional factors that could influence economic growth beyond investments and inflation?</question>
+          </follow_up_questions>
+      </answer>
+
+      ***NOTE***: Ensure that the response is structured correctly and adheres to the guidelines provided. Also, if needed/possible, cite multiple websites to provide a comprehensive response.
+      `,
+    parameterRules: websiteInfoScraperToolParams,
+};
+
+export class WebsiteInfoScraperTool extends BaseTool<WebsiteInfoScraperToolParamsType> {
+    private _addLinkedUrlDoc: (url: string, id: string) => void;
+
+    constructor(addLinkedUrlDoc: (url: string, id: string) => void) {
+        super(websiteInfoScraperToolInfo);
+        this._addLinkedUrlDoc = addLinkedUrlDoc;
+    }
+
+    async execute(args: ParametersType<WebsiteInfoScraperToolParamsType>): Promise<Observation[]> {
+        const urls = args.urls;
+
+        // Create an array of promises, each one handling a website scrape for a URL
+        const scrapingPromises = urls.map(async url => {
+            try {
+                const { website_plain_text } = await Networking.PostToServer('/scrapeWebsite', { url });
+                const id = uuidv4();
+                this._addLinkedUrlDoc(url, id);
+                return {
+                    type: 'text',
+                    text: `<chunk chunk_id="${id}" chunk_type="url">\n${website_plain_text}\n</chunk>`,
+                } as Observation;
+            } catch (error) {
+                console.log(error);
+                return {
+                    type: 'text',
+                    text: `An error occurred while scraping the website: ${url}`,
+                } as Observation;
+            }
+        });
+
+        // Wait for all scraping promises to resolve
+        const results = await Promise.all(scrapingPromises);
+
+        return results;
+    }
+}