Kore.ai Integration

​

Overview

​

XO Platform

​

Prerequisites

• A Kore.ai Contact Center AI app on the XO Platform, with the flow you want to test.
• A phone number added to that app and attached to the flow.
• An API scope mapping for the app (created below) that lets Cekura read sessions, chat history, and debug logs.

​

Set up Kore.ai

​

Connect Kore.ai in Cekura

1

Select Kore.ai

In Cekura, go to Agents, create an agent, and select Kore.ai as the provider. If you don’t see it, click More options to expand the full provider list.

2

Enter the Kore.ai credentials

Fill in the fields under Integration Settings:

Field	Value
Platform	Leave on XO.
Kore.ai Client Secret	The Client Secret from the API scope mapping. Stored encrypted and never shown again.
Kore.ai Client ID	The Client ID from the API scope mapping (cs-...).
Kore.ai Bot ID	The App ID of your Contact Center AI app (st-...).
Kore.ai Host	Optional. Leave blank for the standard Kore.ai cloud. Only set it if your account runs on a dedicated or regional Kore.ai host, and enter the domain only (no https://).

3

Add the phone number

Under Connections, turn on Telephony, then enter the Kore.ai phone number attached to your flow under Telephony Settings. Leave Inbound on so Cekura dials your number.

4

Save the agent

Save the agent. The secret field shows Client Secret Configured once it is stored; use Update to rotate it later.

​

Run a test

1. Looks up the app’s sessions that started within a short window around the time Cekura placed the call.
2. Fetches the chat history of each candidate and confirms the match by comparing timing and conversation content, so parallel test calls to the same number are each paired with the right Kore.ai session.
3. Fetches the debug logs for the matched session.
4. Replaces its own transcript with the Kore.ai one and attaches the Kore.ai session ID to the result.

​

Troubleshooting

• Authentication fails: check that the API scope mapping uses HS256, and that the Client ID and Client Secret come from the same mapping. The Bot ID must be the App ID (st-...) of the app that owns the mapping.
• No Kore.ai transcript on the result: make sure the mapping has the App Sessions and Chat History scopes, and that the number under Telephony Settings is the one attached to the flow you tested.
• Transcript present but no Function Call entries: add the Debug Logs scope to the mapping. Without it Cekura still imports the conversation but cannot see the flow’s service-node calls.
• Host errors: if you set Kore.ai Host, enter only the domain, for example platform.kore.ai.

​

Configure through the API

{
  "assistant_provider": "koreai",
  "transcript_provider": "koreai",
  "koreai_client_secret": "<client-secret>",
  "koreai_data": {
    "client_id": "cs-...",
    "bot_id": "st-..."
  },
  "contact_number": "+15551234567"
}

​

Agent Platform (Artemis)

​

Prerequisites

• A Kore.ai Agent Platform project containing the voice agent you want to test, reachable on a phone number.
• The host your project runs on, for example ind-agents.kore.ai. Enter the domain only, without https://.
• A project-bound API key for that project with the session:read and analytics:read permissions. Ask your Kore.ai administrator to issue one; it starts with abl_.
• If Cekura dials from its own numbers, make sure the Kore.ai project accepts calls from them. Kore.ai only creates a session once the call reaches the agent, so an IVR or allow-list in front of it must let the Cekura number through.

​

Connect Kore.ai Agent Platform in Cekura

​

Run a test

1. Lists the project’s voice sessions that started between 30 seconds before Cekura placed the call and 60 seconds after it ended. Kore.ai indexes a finished session a few seconds after the call, so Cekura retries for up to two minutes if nothing is there yet.
2. Fetches each candidate’s conversation history and confirms the match by comparing timing and conversation content, so parallel test calls to the same number are each paired with the right session.
3. Handles transferred calls. If the agent handed the caller to a human queue, Kore.ai’s session covers only the middle of the call: it starts after the IVR and ends at the transfer. Cekura places that slice on its own timeline and matches it against the same stretch of its transcript, so the session is still paired even though the two transcripts differ in length.
4. Fetches the matched session’s tool calls, failed LLM and tool calls, and per-call LLM usage, and attaches the session ID to the result.

​

Troubleshooting

• Authentication fails: the API key is invalid, expired, or missing the session:read or analytics:read permission. Kore.ai returns the same error for a key bound to a different project, so also check Kore.ai Project ID.
• No Kore.ai transcript on the result: confirm the call reached the agent. Kore.ai does not create a session for a call that ended in the IVR or was rejected by an allow-list, and the result then keeps Cekura’s own transcript.
• Transcript stops early: the agent transferred the call. Kore.ai logs nothing after a transfer; see the note above.
• Chat runs are refused: Agent Platform agents can only be tested over voice. Switch Platform to XO if you need Kore.ai chat testing.
• Host errors: enter only the domain, for example ind-agents.kore.ai.

​

Configure through the API

{
  "assistant_provider": "koreai",
  "transcript_provider": "koreai",
  "koreai_data": {
    "platform": "artemis",
    "artemis_api_key": "abl_...",
    "artemis_host": "ind-agents.kore.ai",
    "artemis_project_id": "<project-id>"
  },
  "contact_number": "+914068511282"
}

{
  "provider": {
    "type": "koreai",
    "credentials": {
      "config": {
        "platform": "artemis",
        "artemis_api_key": "abl_...",
        "artemis_host": "ind-agents.kore.ai",
        "artemis_project_id": "<project-id>"
      }
    }
  }
}