​

Overview

We understand that enterprise customers pursuing AI adoption have three key requirements:

1. Data localization
2. Compute localization
3. Reduced network surface area for ingress (private networking)

Inferable is designed to meet all these requirements.

​

On-Premise Execution and Data Localization

Inferable offers robust on-premise execution, giving you full control over your compute and data:

• Local Runtime Environment: Inferable SDK sets up a dedicated runtime environment within your own infrastructure.
• Workflow Execution: All function calls are executed entirely within your local environment.
• Secret Management: Sensitive information like API keys and passwords remain within your infrastructure.

This approach addresses concerns about data sovereignty and compliance with regional data protection regulations.

​

Private Networking

Inferable’s architecture allows your compute to run on private networks without requiring any incoming connections:

• Outbound Connections Only: The Inferable SDK initiates all connections from within your infrastructure to the Inferable control plane.
• Long-Polling Mechanism: Your services periodically check for new tasks, eliminating the need for open inbound ports.

Benefits include:

• Reduced attack surface
• Mitigation of Man-in-the-Middle (MITM) attacks
• Simplified firewall rules
• Ability to deploy in private subnets

​

Zero-Retention of Private Data with Sentinel

For enterprise customers requiring an additional layer of data protection, our Sentinel feature offers:

• Complete data localization when interacting with external servers
• Masking of all data sent to Inferable control-plane
• Guaranteed zero-retention of private data
• Ability to audit all traffic to and from your services and the Inferable API

Sentinel acts as a secure intermediary, tokenizing sensitive data before it leaves your infrastructure and detokenizing it upon return.

​

Bring Your Own Model (BYOM)

With Sentinel, enterprise customers can bring their own models. Inferable is model-agnostic, but requires a MoM (mixture of models) that are available commercially via AWS Bedrock. (Support for other providers is in the works.)

​

SOC2 Certification

We are currently undergoing SOC2 Type I and Type II certification processes.