Key takeaways

  1. Self-hosted means the app, database and AI run on infrastructure you control.
  2. The usual reason is residency, isolation or a network that cannot talk to the public internet.
  3. Most “on-prem CRMs” still send AI prompts to a cloud model. Ask that question separately.
  4. Qivity runs on Docker with PostgreSQL and Redis. Lilly is an optional add-on that then runs on your infrastructure, with egress optional.

A self-hosted CRM is software you run on infrastructure you control: your VMs, your Kubernetes cluster, your private cloud. The application, the database and — if the product is honest — the AI assistant all live there.

It is not the same as “we have a data centre region in Mumbai.” Residency of a SaaS database is useful. It is not self-hosting.

When self-hosted CRM is the right call

  • A regulator or a customer contract names the processing location more tightly than a vendor region can satisfy.
  • The network is air-gapped, or egress is an incident.
  • You already operate PostgreSQL and would rather extend that than open a new SaaS attack surface.
  • Procurement will not accept a foundation-model subprocessor, and the CRM’s assistant has to keep working anyway.

If none of those are true, a well-run hosted deployment is simpler. Self-hosting is an operations choice, not a badge.

The AI catch

Plenty of on-premise CRMs still call out to a hosted model. You have then paid for a private database and a public prompt. A private AI CRM keeps inference inside the same boundary. Qivity’s self-hosted option includes Lilly on local models. The architecture comparison is the short version.

What you actually operate

Qivity’s footprint is deliberately boring: Docker, PostgreSQL, Redis. Backups are database backups. Upgrades are image pulls. There is no proprietary appliance and no requirement for a partner to “stand up the org.”

You still need someone who can run containers and restore a database. If that person does not exist, do not self-host as a political gesture. Use hosted, keep the AI local to the deployment, and revisit when the team is ready.

Questions for the vendor

  1. Does the AI run in the same install, with egress off?
  2. Which database do we own, and can we dump it?
  3. What is the upgrade path without a professional-services week?
  4. SSO, audit export, backup retention — in the product or as a custom build?

Talk to us if self-hosting is a requirement rather than a preference. Bring the constraint (air-gap, region, DPA) rather than a generic “we want on-prem.” The constraint decides the install.