Deployment
Deployment Overview
Our cloud, your cluster, or somewhere in between. Same platform, same features, same UX across every deployment shape. Pick based on where your data needs to live.
The options
| Mode | Best for | Docs |
|---|---|---|
| Managed EU cloud | Pilots, teams without ops capacity, any data that can live in EU cloud | Managed EU Cloud |
| On-Premise | Regulated industries, IP-sensitive workloads, data that can’t leave your DC | On-Premise |
| Air-gapped | Classified workloads, defence, intelligence, zero-egress requirements | Air-gapped |
| Hybrid | Mixed classifications — sensitive on-prem, routine on our cloud | Hybrid |
| Docker & Kubernetes | Technical deep-dive for self-hosted deployments | Docker & Kubernetes |
What stays the same
Across every deployment:
- The same features — knowledge base, agents, workflows, chat, search, collaboration
- The same UI (admin + user-facing)
- The same APIs
- The same connectors (subject to network reachability)
- The same security model — RBAC, zero-trust, audit log
What changes
| Dimension | Managed | On-prem | Air-gapped |
|---|---|---|---|
| Time to first production | Hours | Days to weeks | Weeks |
| You operate… | Nothing | Everything | Everything |
| Internet egress required | Yes (our cloud) | Optional | None |
| Billing model | Per-token + monthly | Annual licence | Annual licence |
| Updates | Continuous | On your schedule | Via signed data packs |
| Auto-scaling | Automatic | You configure | You configure |
| Support response | 24/7 | 24/7 + on-site if needed | On-site coordination |
Choosing
Start on managed cloud if…
- You’re piloting and need to show value this week
- Your data is not regulation-sensitive
- You don’t have a dedicated infrastructure team
- You want to pay for what you use, not commit to a licence
Go on-premise if…
- Your regulator requires data to stay in your infrastructure
- Your volumes make per-token pricing uneconomical
- You already run Kubernetes for other workloads
- You need to tune performance (GPU choice, batch sizes) yourself
Go air-gapped if…
- Your workload is classified
- You cannot have any outbound internet connectivity
- Regulatory bodies (NBU, DoD, BSI) require it
- You’re dealing with national-security-adjacent data
Go hybrid if…
- You have clearly separable data classifications
- You want agility on less-sensitive workflows without compromising on regulated ones
- You’re migrating from managed to on-prem and need a bridge
Infrastructure requirements (self-hosted)
| Scale | Nodes | GPU | Storage |
|---|---|---|---|
| Pilot (< 50 users) | 3x medium | 1x L4 / L40S | 500 GB SSD |
| Department (< 500 users) | 6x large | 2x L40S / H100 | 5 TB SSD |
| Enterprise (< 5,000 users) | 12+ large | 4+ H100 | 20 TB SSD |
Exact sizing depends on your corpus size, query rates, and chosen LLMs. Our SRE team does capacity planning with your infrastructure team during onboarding.
Deployment lifecycle
- Planning — capacity sizing, network topology, security review
- Environment setup — Kubernetes cluster, Postgres, object storage, GPUs
- Install — Helm chart, configuration, first login
- Identity integration — SSO, SCIM, group mapping
- Connector setup — data sources wired up
- Validation — pilot workloads, security review, audit log verified
- Go-live — rollout to users
- Ongoing — updates, scaling, new workloads