graph TD %%{init: {'theme': 'mc','layout': 'elk'}}%% SearXng-ntpo0[Web Search SearXng] style SearXng-ntpo0 stroke:#a170ff DeepseekModel-yd7iq[Deepseek] style DeepseekModel-yd7iq stroke:#a170ff CreateData-dfb3f[Create Data] style CreateData-dfb3f stroke:#a170ff Switch-56w06[Switch] style Switch-56w06 stroke:#a170ff WebScraper-p0rr5[Web Scraper] style WebScraper-p0rr5 stroke:#a170ff DeepseekModel-4gnpf[Deepseek2] style DeepseekModel-4gnpf stroke:#a170ff CreateData-p25ng[Create Data2] style CreateData-p25ng stroke:#a170ff TextInput-3uo52[<div><img alt="logo" src="/_astro/type.Dy26vmDy.svg" style="height: 20px !important;width: 20px !important"/></div>Cantidad de citios ] style TextInput-3uo52 stroke:#a170ff TextInput-wcn04[<div><img alt="logo" src="/_astro/type.Dy26vmDy.svg" style="height: 20px !important;width: 20px !important"/></div>Query] style TextInput-wcn04 stroke:#a170ff CreateData-taieq[Create Data3] style CreateData-taieq stroke:#a170ff Switch-44sfm[Switch2] style Switch-44sfm stroke:#a170ff AdvancedAgent-plvkg[Agent] style AdvancedAgent-plvkg stroke:#a170ff GSheetCellComponent-usi3o[Sheet Cells ] style GSheetCellComponent-usi3o stroke:#a170ff DeepseekModel-c77dx[Deepseek3] style DeepseekModel-c77dx stroke:#a170ff SearXng-ntpo0 -.- DeepseekModel-yd7iq linkStyle 0 stroke:#a170ff DeepseekModel-yd7iq -.- CreateData-dfb3f linkStyle 1 stroke:#a170ff CreateData-dfb3f -.- Switch-56w06 linkStyle 2 stroke:#a170ff Switch-56w06 -.- WebScraper-p0rr5 linkStyle 3 stroke:#a170ff WebScraper-p0rr5 -.- DeepseekModel-4gnpf linkStyle 4 stroke:#a170ff CreateData-p25ng -.- SearXng-ntpo0 linkStyle 5 stroke:#a170ff TextInput-3uo52 -.- CreateData-p25ng linkStyle 6 stroke:#a170ff TextInput-wcn04 -.- CreateData-p25ng linkStyle 7 stroke:#a170ff DeepseekModel-4gnpf -.- CreateData-taieq linkStyle 8 stroke:#a170ff CreateData-taieq -.- Switch-44sfm linkStyle 9 stroke:#a170ff Switch-44sfm -.- AdvancedAgent-plvkg linkStyle 10 stroke:#a170ff GSheetCellComponent-usi3o -.- AdvancedAgent-plvkg linkStyle 11 stroke:#a170ff DeepseekModel-c77dx -.- AdvancedAgent-plvkg linkStyle 12 stroke:#a170ff

Business Contact Capture Flow Documentation

🧩 Overview

The workflow automates the discovery and collection of contact information for businesses in a specified niche. It performs a web search, filters official sites using an LLM classifier, scrapes the chosen pages, extracts structured contact details, and writes verified records to a Google Sheets spreadsheet. This end‑to‑end process streamlines lead generation and reduces manual data gathering.

⚙️ Main Features

Accepts a niche query and desired number of results from the user.
Executes a web search to retrieve candidate business sites.
Classifies each site title as official or not with a LLM.
Scrapes the content of verified sites to Markdown.
Extracts business name, contact e‑mail, phone, description, and URL using a structured‑output LLM.
Routes only valid records through a conditional switch.
Automatically appends the extracted information to a Google Sheets spreadsheet.
Uses an advanced agent to orchestrate tool usage and manage memory and iteration limits.

🔄 Workflow Steps

Component Name	Role in the Workflow	Key Inputs	Key Outputs
Text Input	Collects the desired number of search results.	None (default value “10”)	Text value
Text Input	Collects the niche query (e.g., “restaurant in Madrid Spain”).	None	Text value
Create Data	Builds a data record combining the number of sites and the query.	Number of sites, Query	Data object with fields for search
Web Search (SearXng)	Performs a web search based on the query.	Data record with query	List of search results (titles, URLs)
Deepseek LLM (Site Filter)	Classifies each title as official (“Sí”) or not (“No”).	List of search results	Data with classification label and URL
Create Data	Constructs records containing the label, site name, and URL.	Classification results	Data items with fields etiqueta, nombre_del_citio, url
Switch (Filter Official Sites)	Routes only items labeled “Sí” to the next step.	Data items	Filtered data for official sites
Web Scraper	Scrapes the HTML of each official site into Markdown.	URL	Scraped Markdown text
Deepseek LLM (Data Extractor)	Parses Markdown to extract structured business details.	Scraped Markdown	Data with nombre, correo, telefono, descripcion, url
Create Data (Prepare for Google Sheets)	Builds a row ready for the spreadsheet.	Extracted business fields	Data with etiqueta, nombre_del_citio, url
Switch (Filter Non‑“No” entries)	Excludes any records marked “No”.	Data rows	Valid rows for spreadsheet
Advanced Agent	Orchestrates tool usage and final insertion into Google Sheets.	Valid rows	Agent response
Google Sheets Cells	Adds a new row to the specified spreadsheet.	Row data	Confirmation of insertion

🧠 Notes

The LLM used for classification and extraction must receive the exact system messages; any deviation may cause the component to output “No” and halt the pipeline.
The switch components rely on the exact words “Sí” and “No”; any change in case or punctuation will prevent routing.
The web scraper is limited to a 10‑second timeout per request; sites that take longer will fail silently.
Google Sheets integration requires valid credentials and write access to the target spreadsheet; failures will surface as an exception in the sheet component.
All batch components run up to five parallel executions by default, which can be tuned via the parallel executions setting.
If no sites pass the official‑site filter, the workflow stops before the agent stage.
The advanced agent employs a 35‑iteration cap and a trim memory strategy to maintain context without exceeding resource limits.
The final agent response includes the agent’s state only when the Include State in response flag is set to true; otherwise only the result of the sheet operation is returned.