DEPLOYMENT
Three deployment models. Same AI, same product. What changes is where your deal data, data rooms and credit files live and where the inference runs.
THE 3 MODES
Amaril is the same application across every configuration. What changes is where the app runs, where the AI inference runs, and who owns the hardware. The page does not ask you to choose: a sales call helps you map the mode to your regulatory perimeter, from MiFID II to GDPR.
01 · ON-PREM
A machine dedicated to you, hosted by us in EU data centers. No shared cloud: your deal team works on a dedicated machine, fully managed by us, with custom domain and brand. Nothing to operate, no hardware to buy, time-to-value measured in hours.
For teams that want to start now without managing infrastructure.
02 · DEDICATED ON-PREM
Amaril installed inside the customer's private cloud · AWS, Azure or GCP, EU region. The app runs in your perimeter, under your contract with the hyperscaler. AI inference local on your cloud GPUs, or via API · your call. We handle deployment and runbooks; you hold the keys and deal data never leaves your account.
For financial institutions that already run a private cloud and want the deployment inside their own perimeter.
03 · SOVEREIGN
Everything on the customer's physical hardware, inside their network. AI inference local on on-site GPUs (NVIDIA DGX small/medium or HP DL380 + 2× RTX 6000 Ada). Air-gappable: no outbound calls, no egress. The strictest mode, for data rooms, MNPI and ultra-sensitive credit files.
For firms that cannot have a single bit of a deal or credit file leave the rack.
Not sure which mode fits? You are not alone · even bank and fund CISOs mix up Dedicated On-Prem and Sovereign. A 30-minute scoping call clears it up.
HARDWARE
For Sovereign, where the institution wants AI on its own rack behind the firewall, we rely on datacenter-grade enterprise hardware. Two reference architectures, picked by workload and budget.
SMALL · MEDIUM SOVEREIGN
NVIDIA integrated system for sovereign AI inference. We use it in Sovereign when the workload is small-to-medium. Desktop or rack form factor depending on size.
ENTERPRISE SOVEREIGN
HP DL380 enterprise server with two NVIDIA RTX 6000 Ada Generation cards. Our reference for enterprise Sovereign: 96 GB combined VRAM, sufficient for production models on deal teams and desks up to hundreds of concurrent analysts.
The HP DL380 + RTX 6000 Ada option is available with HP 5-year financing at preferential rates: it turns a significant capex into a predictable monthly fee, aligned with the IT budget cycle.
THREE COMPONENTS · ONE TECHNOLOGY
The three Amaril components share the same agentic engine and the same document corpus of financials, filings and data rooms. No incompatibility across deployment modes: when you move from On-Prem to Sovereign, the products stay the same, the sources stay connected and the due diligence workflows keep running.
01 · CORE
The flagship product: agentic search across large document corpora, analysis of financials and filings, answers always cited to the source with no hallucinations. The backbone the other two plug into.
02 · INTELLIGENCE
Mail intelligence for deal flow and investor relations: classifies, prioritizes and summarizes threads with source citation, also in Dedicated On-Prem and Sovereign modes.
One brand, three components, three deploys. No incompatibility between tiers.
FAQ
The questions our sales team gets in the first 48 hours of a conversation with banks and funds.
NVIDIA TECHNOLOGY PARTNER
Reference architectures and AI deployment expertise on NVIDIA infrastructure across all three modes: from On-Prem (DGX managed by us) to Dedicated On-Prem (customer cloud) to Sovereign (DGX or HP DL380 + RTX 6000 Ada on the customer rack).
