Kilevel
Dashboard

Language / Sprache

KILEVEL © 2026

CRM and systems

Voice agents that know the customer and update the record

Connect an agent to your CRM so it can look up the caller during the conversation and send a clean result back when it ends, without anyone reading a transcript.

This page inEnglishDeutschTürkçe

The problem

An agent that knows nothing about the caller has to ask for everything again, and a call that ends as a transcript leaves someone copying details into the CRM by hand.

Both problems sit at the seam between the conversation and your systems, and every company's systems look different.

Who it is for

  • Support teams: Recognising the customer and their open case.

  • Sales teams: Keeping lead records up to date after every call.

  • Operations: Creating tickets, orders or appointments from a conversation.

  • Developers: Connecting a system of your own through a documented interface.

What it handles

  • Look-up before hello

    The caller's number is available from the first turn, so the agent can find the record before it speaks.

  • Actions during the call

    Check stock, read an order, create a ticket: the agent calls your endpoint and uses the response in its next reply.

  • Results after the call

    Structured outputs fill in the fields you define, and a webhook or automation delivers them.

  • Starting from your CRM

    A hosted webhook lets your CRM or form tool start an automation.

  • Matching records

    The identifiers you pass in stay on the record untouched, so matching a result needs no look-up.

How it works

  1. Choose the connection

    An MCP server if your CRM offers one, otherwise a custom HTTP tool against its API.

  2. Describe the tools

    A name, a description the model reads to decide when to use it, and the parameters it fills in.

  3. Limit where they run

    A tool can be limited to certain states of a flow, so a refund tool appears only once the caller has been verified.

  4. Define the outcome

    Set out the fields you want, as a structured output.

  5. Send it back

    Use the OnAIAnalysisReady webhook or an automation that posts the result to your CRM.

What you need

  • A CRM with an HTTP API or an MCP server
  • Credentials with only the access the agent needs
  • The look-ups and actions the agent should perform
  • The fields you want back after a call
  • Someone who can configure an API or a webhook

What it connects to

  • Custom HTTP tools: Any endpoint you can describe, with method, headers and body.

  • MCP servers: A server's tools become available to the flow like any other tool. One-click connections exist for several known services, and any other MCP server can be connected by its address.

  • Webhooks: OnSessionBegin, OnSessionEnd and OnAIAnalysisReady.

  • Automations and API: Multi-step follow-ups on a visual canvas, and a REST API.

What to expect

  • Callers are not asked for what you already know about them.
  • Records are updated with the same fields after every call.
  • Less copying from transcripts into other systems.

Limits

  • Connections are general-purpose: there is no dedicated Salesforce or HubSpot connector, so your CRM needs an HTTP API or an MCP server, and someone has to set it up.
  • Your CRM's response time becomes part of the conversation. For slow actions, use asynchronous tools or let the agent acknowledge the wait.
  • Voice transcripts can be wrong. Have tools that write data require a confirmed transcript.
  • Give the agent the narrowest access that works. Scoping connections to one application helps here.

How to evaluate it

  • Start read-only: Connect one look-up and listen to how the agent uses it.

  • Use test endpoints: Try tools against the echo endpoints before pointing them at a real system.

  • Add one write: Send a single field back and check it in the CRM.

  • Test the failures: See what the agent says when your system is slow or returns nothing.

Questions

Does Kilevel have a Salesforce or HubSpot connector?

Not a dedicated one today. You connect through their APIs with custom HTTP tools, or through an MCP server where one is available to you. The result can be just as useful, but it takes some setup.

Can the agent read customer data during a call?

Yes. A tool calls your endpoint during the conversation, and the response comes back to the model for its next reply.

How do results get back into the CRM?

Structured outputs produce the fields, and a webhook or an automation delivers them. Most integrations only need the OnAIAnalysisReady webhook.

Do we need developers?

Someone has to be comfortable with your CRM's API or MCP server. After that, the tools are configured in the dashboard.

Does Kilevel have its own CRM?

Kilevel keeps contacts and conversation records and has a built-in calendar with public booking pages. If you already run a CRM, connect it and keep it as the system of record.

See it on your own calls

Start with a test number and a small set of real questions. Nothing has to change on your main line until you decide it should.