On-premises
Installed in your own datacentre
Installed in your own datacentre or server room. Standard for pharma sites and BFSI institutions with existing infrastructure.
Deployment & Sizing
What It Actually Takes to Run Private AI in Your Datacentre
Most private-AI conversations stall on one question: what hardware do we need? Here is the honest answer, with no hedging. Most first deployments run on a single GPU server. You do not need a cluster.
01
Three ways to install, all inside your perimeter.
Private cloud
Your own tenancy
Your own tenancy in AWS, Azure or an India-region provider. Suits organisations without on-site server capacity but with data-residency requirements.
Air-gapped
No network route to the internet
No network route to the internet at any point. Updates applied via signed offline packages. Standard for defence-adjacent, IP-critical and highest-sensitivity deployments.
02
Start on one GPU. Scale only when concurrency demands it.
| Tier | Users | Documents | Indicative hardware |
|---|---|---|---|
| Pilot | 5–20 | Up to ~50k pages | Multi-core CPU, 64 GB RAM, NVMe SSD, single 24–48 GB GPU |
| Department | 20–100 | ~500k pages | 128 GB RAM, NVMe array, single 48 GB GPU (L40S/A6000 class) |
| Enterprise | 100+ | Multi-million pages | Multi-GPU node or small cluster, HA configuration |
Final sizing is confirmed in the Readiness Assessment against your actual document mix — OCR-heavy scanned archives change the arithmetic significantly.
03
Open-weight models, permissively licensed, your counsel's choice.
FortiqAI is model-agnostic. We deploy open-weight models that run entirely on your hardware, and we default to permissively licensed families (Apache 2.0 and similar) because they are the lowest-friction option for an enterprise legal review. Where a restrictive community licence is acceptable to your counsel, we can deploy those too.
Models can be swapped as better options release, without re-architecting your deployment or re-ingesting your documents. You are not locked to one vendor's model roadmap.
04
Where your documents already live.
SharePoint
Network file shares
S3-compatible object storage
PostgreSQL
Internal DMS platforms
REST and GraphQL APIs
PDF / DOCX / PPTX / XLSX / CSV
OCR for scanned archives
05
Patched on your schedule, not ours.
Signed update packages with a documented application process. Three supported modes: manual offline patching for air-gapped sites, scheduled secure update windows, or one-time supervised connections under your control. No background sync. No automatic updates.
06
Discovery to go-live.
01
Discovery
Understand your documents, workflows, and environment.
02
Preparation
Environment prepared to spec.
03
Installation
System installed in your infrastructure.
04
Ingestion
Documents ingested and indexed.
05
Go-Live
Access configured, validated, trained, and handed over.
Sizing
Get a sizing recommendation.
Tell us about your document volume and we'll tell you what hardware you actually need — no oversizing.