Receives Vapi end-of-call webhooks for your borrower-support assistant, extracts the call transcript, and stores it — agent turns and borrower turns kept separate. You pull the stored transcripts into a Google Colab notebook and run sentiment analysis with the OpenAI API.

Try the app

Point your Vapi assistant's Server webhook URL at this endpoint. It handles the end-of-call-report event (where Vapi wraps the call in message.call) and stores the split transcript.

{
  "ok": true,
  "callId": "borrow-789",
  "transcript": {
    "agent": "I understand this is stressful...\nYes, we can discuss a deferment plan...",
    "borrower": "I'm really struggling financially...\nCan I defer for a few months?...",
    "full": "borrower: ...\nagent: ...\nborrower: ...\nagent: ..."
  }
}

Returns every stored call, newest first, as JSON — this is what the Colab notebook fetches.

{
  "calls": [
    {
      "id": "borrow-789",
      "received_at": "2026-09-03T22:01:56.000Z",
      "agent_transcript": "...",
      "borrower_transcript": "...",
      "full_transcript": "..."
    }
  ]
}

Only these fields are stored — no call duration, phone number, or other metadata.

1. In the Vapi dashboard, open your borrower-support assistant → Server → Webhooks.
2. Add the webhook URL: https://vapi-transcript-webhook.val.run/
3. Subscribe to the End-of-call report event.

Paste this into a Colab cell. Fill in your OpenAI key where noted. It fetches all transcripts and runs separate sentiment analysis on the agent's text and the borrower's text.

# ---------- 1. SETUP ----------
# Paste your OpenAI API key here (or use Google Colab secrets).
OPENAI_API_KEY = "sk-..."          # <-- put YOUR key here

WEBHOOK_URL = "https://vapi-transcript-webhook.val.run/"

import requests
from openai import OpenAI

client = OpenAI(api_key=OPENAI_API_KEY)

# ---------- 2. FETCH ALL TRANSCRIPTS ----------
data = requests.get(WEBHOOK_URL).json()
calls = data["calls"]
print(f"Found {len(calls)} stored call(s).\n")

# ---------- 3. SENTIMENT ANALYSIS ----------
def analyze(text: str) -> str:
    """Return a sentiment label + one-line summary for a piece of text."""
    if not text.strip():
        return "(no text)"
    resp = client.chat.completions.create(
        model="gpt-4o-mini",
        messages=[{
            "role": "user",
            "content": (
                "Classify the sentiment as positive, negative, or neutral and give a "
                "one-line summary. Respond in this exact format:\n"
                "Sentiment: <label>\nSummary: <one line>\n\n"
                f"Text:\n{text}"
            ),
        }],
    )
    return resp.choices[0].message.content

for call in calls:
    print("=" * 50)
    print("Call ID:", call["id"])
    print("Received:", call["received_at"])
    print("\n--- AGENT sentiment ---")
    print(analyze(call["agent_transcript"]))
    print("\n--- BORROWER sentiment ---")
    print(analyze(call["borrower_transcript"]))
    print()

# ---------- 4. (Optional) TABULAR SUMMARY ----------
# Uncomment to build a quick table of borrower sentiment per call.
# import pandas as pd
# summary = []
# for call in calls:
#     resp = client.chat.completions.create(
#         model="gpt-4o-mini",
#         messages=[{
#             "role": "user",
#             "content": (
#                 "Answer with exactly one word: positive, negative, or neutral.\n\n"
#                 f"Borrower text:\n{call['borrower_transcript']}"
#             ),
#         }],
#     )
#     label = resp.choices[0].message.content.strip().lower()
#     summary.append({"call_id": call["id"], "borrower_sentiment": label})
# print(pd.DataFrame(summary))

To store your key as a Colab secret instead of in the code:

from google.colab import userdata
OPENAI_API_KEY = userdata.get("OPENAI_API_KEY")