Security Suite Production sizing
Overview
This sizing covers Prompt Guardian, Shadow AI, MCP Gateway, AI Gateway, and the Admin Center.
Most customers use their own LLM platform or AWS Bedrock. No local GPU is required. A standard deployment uses:
One Windows server for the Admin Centre, APIs, Prompt Guardian, and reporting.
One Linux Docker host for the AI Gateway and MCP Gateway.
Initial Sizing
Profile | Typical users | Windows server | Linux host | Shadow AI proxy (Windows Server) | Storage | GPU (Optional for local classification) |
|---|---|---|---|---|---|---|
Small | Up to 500 | 4 vCPU, 8 GB RAM | 4 vCPU, 8 GB RAM | May share the Windows server | 100 GB SSD per server | 1× NVIDIA L4 24 GB; T4/A2/A10 24 GB acceptable |
Standard | 500-2,500 | 8 vCPU, 16 GB RAM | 8 vCPU, 16 GB RAM | 1 dedicated node: 4 vCPU, 8 GB RAM | 200 GB SSD per server | 1× NVIDIA L4 24 GB; A10/A10G 24 GB acceptable |
Large | 2,500-10,000 | 16 vCPU, 32 GB RAM | 16 vCPU, 32 GB RAM | 2 Servers: 8 vCPU, 16 GB RAM each | 300 GB SSD per server | NVIDIA L40S 48 GB or 2× L4 24 GB |
X-Large | 50k - 70k | 32 vCPU, 64 GB RAM | 32 vCPU, 64 GB RAM | 4 Servers: 32 vCPU, 64 GB RAM | 1 TB SSD per server | 2–4× L40S 48 GB or H100 80 GB |
Important Notes
User ranges are planning guides. Peak requests, proxy traffic, policy complexity, and audit retention may require more resources.
The browser extension does not require a dedicated server.
Use dedicated proxy nodes for medium and large deployments.
For high availability, use two Windows servers, two Linux hosts, two proxy nodes, and a highly available database.
For more than 10,000 users or high proxy traffic, contact AGAT for sizing.
If local models are required, see Hardware Sizing - For Private AI.