Audit the AI agents you already run

Evidova reads the conversations your agents already have, on the platform they already run on, and scores every one. Nothing about your stack changes.

Each platform below has a page of its own, setting out what we read from it and what we leave alone. A platform with no page yet still works the same way: connect it in the app and the conversations arrive.

Search platforms

All 12chat 8voice 4

BolnaWebhookEvidova reads the call Bolna reports when it is over, and scores the conversation inside it.CrispPolledEvidova asks Crisp what changed, reads the conversations that did, and scores them.ElevenLabsWebhookEvidova reads the post-call transcript ElevenLabs already sends, and scores the call it describes.GupshupWebhookEvidova reads the messages Gupshup already posts to your webhook, and scores the conversation they belong to.IntercomPolledEvidova asks Intercom what changed, reads each conversation in full, and scores it.MessengerWebhookEvidova reads the same webhooks Meta already sends your page, and scores the conversations in them.RetellWebhookEvidova reads the call Retell has already finished, and scores what the agent actually said.ZendeskWebhookEvidova reads the messages Zendesk sends as they land, and scores the conversation they belong to.TwilioWebhookEvidova reads the messages Twilio already posts to your webhook, and scores the conversation they belong to.VapiWebhookEvidova sits outside Vapi and scores the calls your agent already made.WATIWebhookEvidova reads the messages WATI posts as they happen, and scores the conversation they belong to.WhatsAppWebhookEvidova reads the same webhooks Meta already sends your agent, and scores the conversations in them.

What differs between them

Every platform here ends up scored by the same rules. What changes is how much each one hands us to score, and that is set by the platform rather than by us.

Audio, on 8 of 12

Where the platform hands over the recording, a score can rest on how something was said and not only on what the transcript holds. Where it does not, the transcript is all there is, and checks about tone or interruption cannot run.

Millisecond timestamps, on 6 of 12

Seconds are enough to measure a reply time and too coarse to order two turns that landed in the same second. On a second-precision platform, turn order is reconstructed from the transcript rather than read off the clock.

Pushed to us, on 10 of 12

A webhook platform posts each conversation as it finishes, so a score follows within seconds. The rest are read on a schedule, so their conversations arrive minutes later. Neither changes the score, only when it lands.

Signed deliveries, on 6 of 12

Where a platform signs what it sends, we verify it and refuse anything that fails. Where it publishes no signing scheme, the token in your webhook URL is the only thing protecting it, so it should be treated as a password and rotated if it leaks.

All twelve are set out side by side, with the exact behaviour of each, on the platform matrix.

Platform matrix

What stays the same

The rules you approve are the rules every conversation is held to, whichever platform it arrived on. A rule written for your phone line applies to the same question asked over WhatsApp, so one agent handled well on one channel and badly on another shows up as a difference in the score rather than a difference in the ruler.

English, Hindi, Tamil, Telugu and Hinglish are scored by the same rules across voice and chat. Nothing about your agent changes on any of these platforms: we read what it has already produced, and we never sit between it and your customer.