> For the complete documentation index, see [llms.txt](https://k-ai.gitbook.io/knowledge-ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://k-ai.gitbook.io/knowledge-ai/the-k-ai-platform/deployment.md).

# Deployment models

K-AI supports three deployment modes; the choice is driven by customer context (residency, ops responsibility, regulatory constraint). All three run the same platform code — only the substrate differs.

## SaaS (multi-tenant)

Hosted on managed Kubernetes in France. Per-customer Kubernetes pods inside a shared cluster, with full pod-level isolation. Standard ingress with automated TLS termination. Customer URLs are exposed under `.kai-studio.ai` (per-subdomain routing for the management and API surfaces).

The SaaS mode is operated by K-AI. No customer infrastructure is required. Onboarding a new customer creates a new K-AI Instance — a new pod, a new Postgres schema, a new vector index, a new storage bucket — all isolated from every other customer (see [Security & isolation](/knowledge-ai/the-k-ai-platform/security.md)). Time-to-first-document is typically same-day once the connector configuration is agreed.

Data residency: France. EU residency available on request.

## On-premise

Identical Kubernetes architecture deployed in the customer's own cluster. Templates, procedures, hardware requirements, and installation guides ship with the platform — see [Operate — On-premise installation](/knowledge-ai/operate/on-premise.md) for the build-and-deploy walkthrough.

The on-premise mode supports **air-gapped environments**: bundled deployment scripts produce and consume offline Docker image archives, so a deployment in a network without internet egress is a supported path rather than an exception. Customers control their own update cadence and can pin specific platform versions.

Operations responsibility is on the customer (or the customer's chosen integration partner). K-AI provides reference templates, runbooks, and remote support.

## Snowflake Native App (SPCS)

Deployed as Snowflake Container Services inside the customer's own Snowflake account. Documents live in Snowflake Stages, embeddings in `VECTOR` columns, management metadata in dedicated schemas. **Data never leaves Snowflake** — this is the architectural property that drives the choice for customers with Snowflake-resident document policies.

The Snowflake Native App is a regular Snowflake Marketplace listing in the customer's region of choice; deployment is initiated from the customer's Snowflake account. A Tier-1 logistics group runs K-AI on Snowflake Native App today.

The Snowflake mode is the most constrained of the three: it inherits Snowflake's resource model (warehouses, compute pools, account-level governance) and its operational SLA. In return it inherits Snowflake's compliance footprint at the substrate level.

## Comparison

| Mode                     | Residency                                                               | Ops responsibility                            | Time-to-deploy                                | Integration constraints                                                                                 | Compliance fit                                                                                                                                     |
| ------------------------ | ----------------------------------------------------------------------- | --------------------------------------------- | --------------------------------------------- | ------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- |
| **SaaS**                 | France (EU residency on request)                                        | K-AI                                          | Same day for first instance                   | Outbound HTTPS to connector sources                                                                     | GDPR-aligned; default for EU customers without specific residency mandates                                                                         |
| **On-premise**           | Customer-controlled (any region, any data center, air-gapped supported) | Customer (or integration partner)             | Days to weeks (depends on cluster readiness)  | Customer cluster must meet sizing requirements; air-gapped deployments use offline image archives       | Required for highly classified data (e.g. *secret-défense* in the French defence sector), classified environments, and most public-sector mandates |
| **Snowflake Native App** | Customer's Snowflake region                                             | Customer (within Snowflake's operating model) | Hours once the Marketplace listing is enabled | Customer must have an active Snowflake account; document estate must be (or be made) Snowflake-resident | Aligned with customers whose data policy mandates Snowflake residency (BFSI, large enterprises with Snowflake-first data strategies)               |

## Next steps

For ops-grade depth — sizing tables, monitoring stacks, backup and restore procedures, upgrade paths — see [Operate — Deployment models](/knowledge-ai/operate/deployment-models.md). For the on-premise installation walkthrough specifically, see [Operate — On-premise installation](/knowledge-ai/operate/on-premise.md).