DEV Community: Yasser

How to Connect Vapi to Currai Without Writing Integration Code

Yasser — Mon, 07 Sep 2026 22:17:14 +0000

Building a voice AI agent is only the beginning. Once customers start using it, you need to understand what they are asking, where conversations fail, and whether the assistant follows your product and compliance rules.

The Currai Vapi Connector sends completed Vapi calls to Currai through an authenticated webhook. You do not need to install a Currai SDK, modify your live call flow, or share your Vapi private API key.

After each completed call, Currai can turn the conversation into:

A readable User Story
Detected user intents
Quality and policy violations
Unsuccessful conversation turns
Call-level events and supporting evidence
Trends and alerts across multiple calls

This guide walks through the complete connection process and explains how to verify that real calls are reaching Currai correctly.

How the Vapi connector works

Vapi continues to operate the live voice assistant. It manages the call, transcription, model execution, speech generation, and provider operations.

When a call ends, Vapi sends an authenticated end-of-call-report to a unique webhook endpoint generated by Currai.

The connection follows this flow:

Currai generates a webhook URL and a one-time Bearer secret.
You save the secret as a custom credential in Vapi.
You add the Currai webhook URL to your Vapi assistant.
Vapi sends an end-of-call-report when a call finishes.
Currai validates the report and reconstructs the conversation.
Currai analyzes the call for user needs, failures, intents, and violations.

The connector begins importing calls completed after setup. It does not automatically import historical Vapi calls.

What you need

Before starting, confirm that you have:

A Currai workspace
Owner or administrator access to the workspace
A Vapi account
A Vapi assistant that can complete a test call
Permission to edit the assistant's Server URL and credentials

Keep the Currai and Vapi dashboards open during setup.

Step 1: Create the connection in Currai

Settings → Connectors

Find the Vapi connector and select Connect Vapi.

Currai generates:

A unique webhook endpoint
A one-time Bearer secret

The endpoint begins with:

https://www.currai.app/api/connectors/v1/vapi/

Copy both values before closing the setup window.

Protect the Bearer secret

The secret authenticates webhook requests coming from Vapi. Treat it like a password.

Currai stores only a hash of the secret and cannot display the original value again. If you lose it, use Rotate secret and update the corresponding credential in Vapi.

Rotating the secret invalidates the previous value immediately.

Step 2: Create a Bearer Token credential in Vapi

Open the Vapi dashboard and navigate to:

Integrations → Server Configuration

Create a custom credential with these settings:

Setting	Value
Authentication type	`Bearer Token`
Name	`Currai`
Header	`Authorization`
Token	The secret generated by Currai
Include Bearer Prefix	Enabled

Paste only the secret value. Do not add Bearer before it because Vapi adds the prefix automatically.

The outgoing header will follow this format:

Authorization: Bearer YOUR_CURRAI_SECRET

This value is generated by Currai. It is not a Vapi public or private API key.

Step 3: Add the webhook to your Vapi assistant

Open the Vapi assistant you want to monitor and navigate to:

Advanced → Webhook Server

Complete the following steps:

Paste the complete Currai endpoint into Server URL.
Select the Currai Bearer Token credential under authorization.
Set serverMessages to only end-of-call-report when event selection is available.
Save the configuration.
Publish the assistant.

Do not shorten the webhook endpoint or remove its final connection identifier. Every Currai connection has a unique URL.

Currai is a post-call analytics destination. Do not configure it as a live relay for tool calls, transfers, assistant requests, or call-control events.

Step 4: Complete a real test call

After publishing the assistant, make and complete one real call.

Use a request that produces a clear user and assistant exchange, such as:

I need to change the delivery address for an order I placed today.

Wait for the assistant to respond, then end the call normally.

A webhook test in Vapi may confirm that authentication works, but it does not prove that Currai can reconstruct a complete conversation. A finished call is required for end-to-end verification.

Step 5: Verify the connection in Currai

Return to:

Currai → Settings → Connectors

The Vapi connector should move from Waiting for first call to Receiving data.

Review the conversation

Open Analyse → User Stories and confirm that the caller and assistant messages appear as a readable conversation.

Inspect call events

Open Observe → Events and look for the call-boundary event and normalized conversation turns created from the final Vapi report.

Review detected intents

Open Analyse → Intents to see recurring user requests, goals, and problems found across conversations.

Review violations

Open Analyse → Violations to find assistant responses that conflict with the behavioral, product, or compliance rules you define.

What Currai imports

For each completed Vapi call, Currai creates:

One deterministic session for the complete call
One call-boundary event
One normalized event for each conversation turn
Product and quality signals generated from the conversation

If the final user message does not receive an assistant response, Currai can mark that turn as unsuccessful.

Vapi may retry a webhook when delivery times out or fails. Currai uses stable identities so repeated delivery does not create duplicate sessions or consume duplicate event credits.

If Vapi already sends webhooks to your server

Vapi applies Server URL priority rules and sends an event to the applicable destination.

If your assistant, phone number, or account already sends reports to your backend, replacing that URL could interrupt the existing integration.

In that situation:

Keep your existing Vapi Server URL.
Receive the end-of-call-report in your application.
Forward the unchanged report to the Currai webhook endpoint.
Add the Currai Bearer authorization to the forwarded request.

This allows your current workflow to continue while Currai receives completed calls for analysis.

Local development

Vapi cannot send webhooks directly to localhost.

To test a locally running application, expose it through a secure HTTPS tunnel such as ngrok and use the complete development endpoint.

For production, use the generated endpoint under:

https://www.currai.app

Temporary tunnel URLs should not be used for production delivery.

Troubleshooting

401 Unauthorized

Confirm that the assistant uses the credential containing the current Currai secret.

If you rotated the secret, the old value stopped working immediately. Replace it in the Vapi credential.

The connector remains on “Waiting for first call”

Check that:

The assistant was published after configuration
The call was completed normally
end-of-call-report is enabled
The correct Server URL is attached
The Currai authorization credential is selected

Currai receives the call but shows no conversation

Confirm that the final Vapi report includes structured messages in artifact.messages. A plain transcript may not contain enough information to reliably reconstruct the user and assistant turns.

Delivery error

Open the connector card in Currai and inspect the displayed error.

An event_limit_exceeded error means the workspace has no remaining event allowance for additional calls.

Usage and privacy

Each imported call uses:

One event for the call boundary
One event for every normalized conversation turn

Connector traffic follows the workspace's normal event allowance.

Currai stores the connector secret as a hash rather than readable text. The connector is also designed to avoid retaining raw caller phone numbers and recording URLs.

Your team should still ensure that voice data collection follows your consent, privacy, and retention policies.

What to do after connecting Vapi

Once calls are arriving, use Currai to:

Discover why customers contact your voice assistant
Identify recurring user frustration
Detect prohibited or unsupported assistant claims
Monitor unsuccessful and abandoned turns
Create alerts for important failures
Compare quality across assistants
Find evidence for prompt, tool, and workflow improvements

Instead of manually listening to calls or reviewing random transcripts, your team receives organized product and quality signals backed by real conversation evidence.

Start monitoring your Vapi assistant

The Currai Vapi Connector helps you analyze completed voice conversations without installing another SDK or changing the live call experience.

Vapi continues to operate the assistant. Currai helps you understand how that assistant performs with real users.

Read the Vapi Connector documentation →

Start using Currai →

AI Observability for Lovable Apps: Monitor, Test, and Improve Prompts with Currai

Yasser — Thu, 18 Jun 2026 21:15:05 +0000

AI Observability for Lovable Apps: Monitor Prompts, Traces, and Evaluations with Currai

Building AI applications has never been easier.

Tools like Lovable allow developers and founders to create AI-powered products in minutes. Whether you're building a chatbot, AI assistant, recommendation engine, AI agent, or prediction app, generating the application is often the easy part.

The real challenge starts after launch.

How do you know what prompts are being sent to the model?
How do you debug unexpected AI responses?
How do you compare prompt variations and determine which performs better?
How do you evaluate output quality over time?
How do you track token usage and costs?

This is exactly why we built Currai.

What is Currai?

Currai is an AI observability platform that helps teams understand, test, and improve AI applications in production.

It provides:

Prompt tracing
AI request monitoring
Session tracking
Prompt versioning
A/B testing
LLM evaluations
Cost and token analytics
OpenTelemetry support

Instead of guessing why your AI application produced a particular response, Currai lets you inspect the entire execution flow.

The Problem With AI Applications

Traditional monitoring tools were built for APIs, databases, and backend services.

AI applications introduce a completely different set of challenges:

Prompt changes can significantly impact output quality
Model updates can affect behavior
Hallucinations are difficult to track
User conversations are hard to debug
Prompt experiments are often unmanaged
Quality evaluation is usually manual

When something goes wrong, application logs alone don't provide enough visibility.

You need observability designed specifically for AI systems.

Trace Every AI Request

Currai captures every prompt, model response, latency metric, token usage, and cost.

You can inspect:

System prompts
User prompts
Model outputs
Execution traces
Tool calls
Metadata

This makes debugging AI applications dramatically easier.

Run Prompt A/B Tests

Prompt engineering remains one of the most effective ways to improve AI quality.

With Currai, you can compare multiple prompt variants and determine which performs best.

Instead of relying on intuition, you can make decisions using real data.

Whether you're testing:

Different system prompts
Different model providers
Different retrieval strategies
Different output formats

Currai helps you measure the impact.

Evaluate Prompt Quality

Currai includes evaluation workflows that help measure output quality automatically.

You can define evaluation criteria and continuously monitor performance as prompts evolve.

This is especially useful when shipping AI features to production and ensuring quality remains consistent over time.

Understand Usage and Costs

AI costs can grow quickly.

Currai helps you monitor:

Token consumption
Request volume
Latency
Errors
Cost trends

Everything is tied back to the actual traces that generated those metrics.

Example: Building a World Cup 2026 Prediction App with Lovable

To demonstrate how Currai works, I built a FIFA World Cup 2026 prediction application using Lovable.

The app allows users to select two national teams and generate an AI-powered match prediction.

While the application is running, Currai captures:

Every LLM request
Prompt inputs
Model responses
Prompt experiments
Evaluation results
Trace metadata

This makes it easy to understand how the AI behaves and improve prediction quality over time.

Why AI Observability Matters

As AI applications become production systems, observability becomes a necessity rather than a luxury.

Without visibility, you're effectively debugging blind.

Whether you're building:

AI Agents
Chatbots
Copilots
RAG Applications
Customer Support Assistants
Internal AI Tools

Understanding how your AI behaves is critical.

Currai was built to provide that visibility.

Getting Started with Currai

Getting started takes only a few minutes.

Create an account at https://www.currai.app
Generate your API keys
Install the Currai SDK
Instrument your AI application
Start viewing traces, experiments, and evaluations

You can begin monitoring your AI workflows immediately.

Demo Video

In the video below, I show how to build a World Cup 2026 prediction app with Lovable and use Currai to:

Trace every AI request
Compare prompt variations with A/B testing
Evaluate response quality
Debug model outputs
Monitor costs and performance

Learn More

Website: https://www.currai.app
Documentation: https://www.currai.app/docs

If you're building AI products and want better visibility into prompts, traces, evaluations, and experiments, give Currai a try.