Install Xebit VoiceKit
Requirements, install steps, the admin console, API keys, updates and air gapped notes.
Showing instructions for Xebit VoiceCloud.Showing instructions for Xebit VoiceKit, on your own infrastructure. Switch with the selector at the top. Updated 22 October 2026.
Xebit VoiceKit is the complete Xebit platform, installed and run on your own infrastructure: your cloud account, your data centre, or a fully air gapped network. It deploys on Docker or Kubernetes, uses your own models and your own carrier, and is the same artefact Xebit runs in Xebit VoiceCloud. Xebit does the first install with your team; this page explains what that involves so you can prepare.
This page is for Xebit VoiceKit. You have Xebit VoiceCloud selected: see Get started on Xebit VoiceCloud instead.
Requirements
| You provide | Standard on prem | Air gapped |
|---|---|---|
| A container platform | Docker, or a Kubernetes cluster | Docker, or a Kubernetes cluster |
| Compute and storage | Yes, sized for your concurrency, plus storage for recordings and transcripts | Yes, as for on prem |
| AI models | Your own model licences; inference goes to that provider | Local models running on your hardware |
| GPUs | Not required | Yes, for local inference |
| Telephony | Your carrier or SBC, over standard SIP or WebSocket | Your carrier or SBC, reachable inside the network |
| An owner | An engineer who will run it day to day | An engineer who will run it day to day |
Xebit never provides telephony, numbers or messaging. Your agents call and answer over your own SIP trunks.
Air gapped deployments need GPUs, because the models run locally. Standard on prem does not, because inference goes to the model provider you hold the licence with.
Before you install: set the pass mark
Agree the numbers the deployment must meet before anything is installed: the concurrency you need, your cost ceiling, the response time you accept and the languages you will run. Xebit scopes the deployment and your licence band against them. Your concurrency and quotas are then set by your own deployment and licence band, not by Xebit.
Install steps
- Prepare the environment. Provision the Docker hosts or Kubernetes cluster, persistent storage for recordings, transcripts and logs, a host name for the platform (for example
xebit.your-company.internal) and a TLS certificate for it. - Open the network paths. Allow SIP signalling and media between your carrier or SBC and the Xebit SIP gateway, browser access to the admin console for your team, and WebRTC for anyone who will join calls from a browser. On standard on prem, also allow outbound traffic to your model provider.
- Load the release. Xebit provides the release artefacts and your licence. Load them into your registry or onto your hosts. For air gapped, bring them in through your usual transfer process and mirror them into an internal registry.
- Configure the platform. Set the licence, the host name, storage, and the models: provider credentials for on prem, or the endpoints of your local models for air gapped.
- Deploy. Deploy with your existing tooling, on Docker or Kubernetes. It drops into an existing deployment pipeline like any other service.
- Sign in to the admin console and confirm every component reports healthy.
- Connect your carrier. Point your SIP trunk or SBC at the SIP gateway and route numbers on your trunk to agents. See Carrier and SIP Trunk Issues if calls do not arrive.
- Make a test call from an agent to your own phone, and check the recording, transcript and summary are stored.
Where the admin console and API keys live
- The admin console is served from your own host, for example
https://xebit.your-company.internal. Everything this guide describes for the dashboard happens there. - API keys are created in the admin console. Keys from one Xebit VoiceKit deployment do not work on another, and Xebit VoiceCloud keys do not work on Xebit VoiceKit. See API Keys.
- The API base URL is
https://<your-xebit-voicekit-host>/v1. Every API example in these docs works once you swap the base URL. See Auth Setup.
How to update
Support and all updates are included in the licence. There is no separate maintenance contract.
- Read the release notes for the new version.
- Load the new release into your registry or hosts (for air gapped, through your transfer process).
- Deploy it to a staging environment first if you have one, and run a test call.
- Roll it out to production with your usual pipeline. Plan the change for a quiet calling window, and keep the previous release to hand so you can redeploy it if needed.
- Watch Observability after the rollout for regressions in response time or failed calls.
Air gapped notes
- Models run locally on GPUs you provide, and nothing leaves the network.
- Releases come in through your own transfer process, not over the internet.
- Webhooks, CRM updates and workflow actions can only reach systems inside the network. A CRM hosted on the public internet is not reachable.
- Observability reports to the admin console, and to Slack, Teams or email only if those are reachable inside the network.
- Xebit cannot see into the deployment. When you need help, share the logs and call IDs your security process allows.
Next steps
- Create an Agent
- Set up incident reporting
- Connect your systems
- More on the offering: Xebit VoiceKit