Kav
עברית Try the assistant
Menu
Trust & safety

Where is Kav hosted — in the cloud or on servers in Israel?

Kav is cloud-neutral: it runs as containers with no dependency on one cloud vendor. Here is how it's deployed and what that means for where your data lives.

By The Kav team Published 2 min read

Kav isn't tied to any one cloud. It runs as a set of containers with no proprietary cloud service in its critical path, so it can be deployed on different infrastructure without being rewritten. This page doesn't name a single hosting location, because where Kav runs for your organization is something we set out with you directly — ask us, and we'll give you the specific answer for your deployment.

How Kav is deployed

  • Containers throughout. The database (Postgres), message broker (RabbitMQ), cache (Redis) and S3-compatible file storage all run as containers, deployed today with Docker Swarm. The planned path from there is Kubernetes, then AWS.
  • No managed cloud service in the critical path. This is a build rule, not an aspiration: nothing is adopted unless it has both a container form and an equivalent AWS managed service (for example RDS, Amazon MQ, ElastiCache and S3). Moving clouds is an infrastructure decision, not a rewrite.
  • Monitoring is self-hosted too. Logs, traces and metrics go to Grafana, Tempo, Loki and Prometheus running alongside the platform — not to an outside monitoring service.

Your own instance

Organizations normally share a platform, kept apart row by row by database-level security policies. An organization that needs its own instance gets a separately deployed stack — not a separate database bolted onto the shared one.

Where your data actually lives

  • Your systems stay the source of truth. Kav keeps no copy of your databases: a question about personal data is a live call to your own system.
  • The AI model is configured per organization. Supported providers include OpenAI, Azure OpenAI, Anthropic, Bedrock, Vertex, or an OpenAI-compatible endpoint you run yourself.
  • Not every field may reach a model. Each field's sensitivity class decides whether it may be sent to a language model at all.