Sensitive workloads on your infrastructure, routine workloads on our cloud — with one unified UX and a single permission model across both. For organizations that don't fit neatly into "all cloud" or "all on-prem".

When hybrid makes sense

• Mixed data classifications — some data is regulated (patient records, classified documents), some is not (marketing, public content)
• Progressive migration — you’re moving from managed to on-prem but need to run both during transition
• Cost optimization — high-volume routine workloads on your owned GPUs, occasional workloads on pay-per-token cloud
• Regional split — EU cloud for HQ, on-prem in a specific country where sovereignty is mandatory

How it works

Two (or more) StellarBase deployments run in parallel — one on our managed cloud, one in your data centre. A federation layer ties them together:

• Unified identity — single sign-on works across both deployments
• Unified UI — users see one StellarBase, not two. Navigation seamlessly crosses deployments based on what the user has access to.
• Cross-deployment search — a query in one side returns results from both (subject to permissions)
• Data placement control — each Base is explicitly assigned to a deployment. Data doesn’t move between deployments unless you move it.

What lives where

Typical hybrid split:

On our EU cloud	On your infrastructure
Public marketing content	Privileged legal matters
Sales research	Patient records
Training materials	Classified project documents
Vendor documentation	Trade-secret engineering data
HR onboarding content	M&A deal rooms

The split is your decision — we enforce what you configure.

Identity & permissions

A single identity provider (your SSO) handles authentication for both deployments. Your existing group structure maps to roles in each deployment. A user’s experience:

1. Log in once via SSO
2. See a unified list of Bases — visually identical, regardless of which deployment backs them
3. Switch between Bases seamlessly
4. Cross-Base search returns hits from both deployments (filtered by permissions)

Data flow boundaries

By default, no data crosses the deployment boundary. A document in your on-prem deployment stays there. A search that spans both returns metadata and preview snippets only — the full document is rendered locally.

You can explicitly configure cross-deployment flows where needed:

• An agent in the cloud deployment can query the on-prem knowledge base (retrieval only, no full-document export)
• A workflow can move specific data between deployments with explicit approval
• Shared dictionaries and rule sets can be synchronized

LLM routing

Agents can be configured to use different LLMs in different contexts:

• On-prem Base → local LLMs only (no internet call)
• Cloud Base → StellarCloud models (our EU infrastructure)
• Either → external commercial LLM via StellarGate (with PII anonymized)

Routing is per-agent, per-Base. A single workflow that touches both deployments can use a local LLM for the on-prem steps and a cloud LLM for the public-data steps.

Networking

The federation layer requires controlled connectivity between the two deployments — typically a VPN tunnel or private peering. Traffic between deployments is:

• mTLS-authenticated
• Metadata-only by default (no bulk data movement)
• Rate-limited per your policy
• Fully audit-logged

For air-gapped segments that have no connectivity at all, use two separate deployments without federation — users simply log in to each separately. The platform supports this — it’s just not “hybrid” in the unified sense.

Audit & compliance

Each deployment maintains its own audit log. The federation layer logs cross-deployment events (cross-deployment search, cross-deployment workflow triggers) on both sides. For compliance, you get a unified view by exporting and merging — or by streaming both into your SIEM.

Billing

Two components:

• Managed cloud: per-token / per-unit as usual
• On-prem: annual licence sized to the deployment

Unified invoicing available — one invoice, itemized by deployment.

Migration & lifecycle

Hybrid is often a stepping stone:

• Pilot → Hybrid → Full on-prem for organizations planning a full self-host migration
• On-prem → Hybrid → Managed for organizations moving workloads to cloud
• Hybrid indefinitely for organizations whose workloads naturally split

Related

• Managed EU Cloud
• On-Premise
• Auth & RBAC — cross-deployment identity