Step 2: Map Incoming Inputs

Normalize the request payload into consistent fields for the AI agents and summarizers.

1. In Assign Input Fields, set userPrompt to {{ $json.body.source }}.
2. Set cellReference to {{ $json.body.prompt }}.
3. Set ouputLanguage to Hebrew (or update to your preferred language).

Step 3: Set Up Query Refinement and Search

Use AI to refine the query, run a web search, and parse the best link.

1. Configure Refine Query Agent with the prompt text User prompt: {{ $json.userPrompt }}
Prompt's referral: {{ $json.cellReference }}.
2. Ensure GPT-4.1 Mini Core is connected as the language model for Refine Query Agent.
   Credential Required: Connect your openAiApi credentials in GPT-4.1 Mini Core.
3. Configure Bright Data Search Bot with its defined search prompt (keep the JSON-only output requirement intact).
4. Connect GPT-4o Model Core as the language model for Bright Data Search Bot.
   Credential Required: Connect your openAiApi credentials in GPT-4o Model Core.
5. Attach Bright Data MCP Tool as the tool for Bright Data Search Bot and set endpointUrl to https://mcp.brightdata.com/mcp?token=[CONFIGURE_YOUR_TOKEN]&pro=1.
6. Attach Link JSON Parser as the output parser for Bright Data Search Bot with schema { "link": "" }.

Step 4: Configure Scraping and Detail Extraction

Scrape the selected source and extract only relevant content based on the user’s query.

1. In Bright Data Scrape Request, set URL to https://api.brightdata.com/request and keep Method as POST.
2. Set body parameters to include: zone mcp_unlocker, url {{ $json.output.link }}, format json, method GET, country il, and data_format markdown.
3. Set the header Authorization to Bearer [CONFIGURE_YOUR_TOKEN] and replace with your Bright Data token.
4. Configure Extract Relevant Details with the input text ## Input
### The user's original request:
{{ $('Assign Input Fields').item.json.cellReference }} - {{ $('Assign Input Fields').item.json.userPrompt }}
### Full content scanned from a website:
{{ $json.body }}.
5. Ensure Mini GPT Model A is connected as the language model for Extract Relevant Details.
   Credential Required: Connect your openAiApi credentials in Mini GPT Model A.
6. Attach Summary JSON Parser as the output parser for Extract Relevant Details with schema { "summary": "" }.

Step 5: Generate Final Summary and Configure Outputs

Compose the final response and send it back to the requester while logging the output.

1. Configure Compose Summary Output with the prompt scraping summary information: {{ $json.output.summary }}
the actual user request/question: {{ $('Assign Input Fields').item.json.cellReference }} - {{ $('Assign Input Fields').item.json.userPrompt }}.
2. Ensure Mini GPT Model B is connected as the language model for Compose Summary Output.
   Credential Required: Connect your openAiApi credentials in Mini GPT Model B.
3. Attach Final Summary Parser as the output parser for Compose Summary Output using schema { "summary": "Intel was founded in 1968." }.
4. Configure Return Webhook Reply to respond with Respond With set to text and Response Body set to {{ $json.output.summary }}.
5. Configure Append Log Records to write input_prompt as {{ $('Assign Input Fields').item.json.userPrompt }} - {{ $('Assign Input Fields').item.json.cellReference }} and output as {{ $json.output.summary }} into the Search Logs data table (replace [YOUR_ID] with your table ID).
6. Confirm the parallel execution: Compose Summary Output outputs to both Return Webhook Reply and Append Log Records in parallel.

Step 6: Test & Activate Your Workflow

Validate the end-to-end flow from webhook input to summary output, then activate for production use.

1. Click Execute Workflow and send a POST request to the Incoming Webhook Trigger URL with a JSON body containing source and prompt.
2. Confirm Bright Data Search Bot returns a JSON link and Bright Data Scrape Request receives content.
3. Verify Return Webhook Reply responds with a concise summary text and Append Log Records writes a new row in your data table.
4. Once validated, toggle the workflow to Active so it can receive production webhook requests.