The Sales audio Call Analyzer Agent extracts insights from sales conversations using audio transcription, web scraping, and LLM-powered analysis — delivering structured, actionable feedback to sales teams.

​
Developer Guide :

​
Agent Modes & Services

ServiceInput TypeDescription
Audio ServiceUploaded .mp3, .wav, etc.Upload a recorded sales call and analyze via LLM.
Phantom ServiceFathom/Zoom URLExtract transcript using a headless browser (Playwright).
Transcript ServicePlain text transcriptAnalyze a pre-written transcript directly.
All services support optional supplemental data via:
  • Uploaded documents (.docx, .pdf, .txt)
  • Website URL

​
Flow 1: Audio Service

** Purpose:** Analyze uploaded sales call recordings. 
1.  Initialize LLM
    ↓
2.  Upload audio to Gemini Files API
    → Validate file size and duration
    ↓
3.  Transcribe audio using Gemini
    ↓
4.  [Optional] Supplemental Input:
    ├─ If URL is provided → Scrape content
    └─ If document is uploaded → Extract text
    ↓
5.  Initialize history & repository
    ↓
6.  Add LLM prompt templates for sales call analysis
    ↓
7.  Call LLM with:
    → Transcript + supplemental content
    ↓
8.  Delete uploaded audio file (cleanup)
    ↓
9.  Calculate token usage and cost
    ↓
10. Store final LLM output (⚠️ not in `additional_kwargs`)

​
Flow 2: Phantom Service

Purpose: Analyze Fathom/Zoom URLs using Playwright. 
1.  Initialize LLM
    ↓
2.  Use Playwright to extract transcript from the URL
    ↓
3.  [Optional] Supplemental Input:
    ├─ If another URL is provided → Scrape content
    └─ If document is uploaded → Extract text
    ↓
4.  Initialize history & repository
    ↓
5.  Add LLM prompt templates
    ↓
6.  Call LLM with:
    → Extracted transcript + supplemental content
    ↓
7.  Delete temporary files if any
    ↓
8.  Calculate token usage and cost
    ↓
9.  Store final response (⚠️ not in `additional_kwargs`)

​
Flow 3: Transcript Service

Purpose: Analyze a pre-provided plain-text transcript. 
1.  Initialize LLM
    ↓
2.  [Optional] Supplemental Input:
    ├─ If a URL is provided → Scrape content
    └─ If a document is uploaded → Extract text
    ↓
3.  Initialize history & repository
    ↓
4.  Add LLM prompt templates
    ↓
5.  Call LLM with:
    → Provided transcript + supplemental content
    ↓
6.  Delete temporary file (if applicable)
    ↓
7.  Calculate token usage and cost
    ↓
8.  Store final response (⚠️ not in `additional_kwargs`)

​
Shared Components

ComponentDescription
Gemini Files APIUpload and transcribe audio using Gemini’s speech-to-text model
PlaywrightHeadless browser used to scrape and extract embedded transcript data
ScraperFetches readable text from any public webpage
Document ReaderParses .docx, .pdf, .txt, etc., and extracts clean plain text
LLM (GPT/Gemini)Analyzes call content for intent, tone, objections, and insights
Prompt TemplatesStructured prompts to guide sales-focused LLM output
Token Cost TrackerComputes total LLM tokens used and cost estimate
Response StorePersists final analysis for audit/logging (excludes additional_kwargs)

​
Next.js — SALES_CALL_ANALYZER Integration

User Interaction Flow

1.  User selects "Sales Call Analyzer" agent
    ↓
2.  Form Input:
    ├─ Step 1: Provide any of the following:
    │   ├─ Call transcript (text)
    │   ├─ Zoom/Fathom URL (auto-fetch transcript)
    │   └─ Audio upload (.mp3, .wav)
    └─ Step 2: [Optional]
        ├─ Website URL
        └─ Upload document (e.g., proposal, email)

3.  Backend Call:
    → Uses `proagent code` with user inputs
    → Calls `getSalesCallResponse()`

4.  Streaming Response:
    → Displays real-time analysis output in chat
    → Final summary includes:
        ✔️ Sales insights
        ✔️ Objections & resolution
        ✔️ Talk ratio, sentiment, and key actions