On-premises deployment

Use cases

When it makes sense

• Regulators or your security team require recordings and personal data to stay inside the company – typical for banks, healthcare, and government
• You need to connect to internal systems without going over the internet
• You have your own IT staff and a server room or data center
• AI agents and the language model must run inside your network, with no data sent outside
• You already have a PBX and phone lines you want to keep

Learn more

What you need

• Servers or virtual machines – how many depends on the number of concurrent agents
• A separate disk or network storage for call recordings
• A GPU or a Mac mini – if all AI must run in-house
• Network links to your phone carriers and to agent workstations
• Backups, monitoring, and updates – handled by your IT team or by the vendor under contract

Example

Sizing example

Say you have 20 agents per shift, 2,000 hours of calls a month, and you keep recordings for a year.

You'll need two servers with 4 cores and 16 GB of RAM each – one for telephony, one for everything else. Recordings take 2,000 × 12 × 90 ÷ 1,000 = 2,160 GB, so you need a 4 TB disk. If you need fault tolerance, each server gets a standby, bringing the total to 4 servers.

Run your own numbers in the server calculator.

Comparison

On-premises vs cloud

In CONTACT.WORK

How it works in CONTACT.WORK

On-site installation takes 2-7 days, depending on whether it's a single server or a cluster. Nothing is left out: everything available in the cloud – from telephony to AI agents, analytics, and reports – works on your servers too. Internal systems connect through the API and webhooks, and an existing Asterisk connects through the API connector.

The server calculator shows how many servers you need for your number of agents. Above 100 agents, the telephony server needs 32 GB of RAM; above 200 agents, we size the configuration for your project. Plan for about 90 GB per thousand hours of recordings.

There's no charge for the platform itself: you pay for the support plan and the installation work, and the servers and storage are yours. AI models can run on your own hardware; if they run with us as a service, they're billed separately.