Key Facts
- Who: Asset intelligence platform Axonius and Amazon Web Services (AWS).
- What: Implementation of secure, multi-tenant AI agents using Amazon Bedrock AgentCore.
- Architecture: Axonius utilized a "silo" deployment model to ensure strict tenant isolation.
- Impact: The initiative aims to reduce manual security, audit, and compliance burdens by up to 50%.
- Technical Detail: Bedrock AgentCore uses dedicated microVMs for each session to maintain structural separation.
Axonius, a leading asset intelligence platform, has successfully integrated secure multi-tenant AI agents into its infrastructure using Amazon Bedrock AgentCore. The implementation allows Axonius to provide agentic AI capabilities to its customers while maintaining the strict data isolation required for sensitive security and IT environments. By leveraging AWS’s managed service, Axonius seeks to help security teams prioritize risks and coordinate fixes across massive, fragmented datasets.
Scaling AI for the Enterprise SaaS Model
Independent Software Vendors (ISVs) like Axonius face unique challenges when introducing AI agents, specifically regarding how to manage workloads across multiple customers, or "tenants." According to an AWS Machine Learning blog post, organizations adding agentic workloads must balance security, scalability, time to market, and cost tracking. For SaaS providers, these challenges must be managed at the individual tenant level to ensure that data from one customer never interacts with another.
Axonius operates by reconciling data from more than 1,400 systems into a single authoritative source of truth. The company currently manages hundreds of isolated customer environments on AWS. Its first AI agent is designed to interpret the state of large enterprise environments, identifying gaps and risks by analyzing millions of data points from dozens of integration sources. This tool is intended to allow junior analysts to perform complex analyses that previously required hours of manual labor from senior staff.
Navigating Multi-Tenancy Patterns
When architecting AI agents for SaaS, AWS identifies three common patterns: silo, pool, and bridge. In a silo model, tenants are provided with dedicated resources, meaning each tenant has a dedicated agent within the Bedrock AgentCore runtime. The pool model involves tenants sharing resources, where one agent serves multiple users, isolated only by unique session IDs. The bridge model is a hybrid approach, using some shared components, such as a shared Amazon Bedrock Knowledge Base, while keeping other elements, like the agent runtime, in silo mode.
Axonius opted to maintain its existing "silo" deployment methodology. In this architecture, each customer workload resides within a dedicated Amazon Virtual Private Cloud (VPC). Each VPC contains its own Application Load Balancer (ALB), Network Load Balancer (NLB), databases, and compute infrastructure. According to the AWS report, Axonius chose this path to ensure that its AI agents remained scoped exclusively to a single customer’s data, meeting the high security standards required for sensitive asset intelligence.
Technical Requirements and Integration
To successfully deploy AI agents within this siloed framework, Axonius addressed several critical technical requirements:
- Tenant Isolation: Agents must be strictly limited to the data of the specific customer they serve. Amazon Bedrock AgentCore supports this by allocating a dedicated microVM for each session, providing structural separation at the runtime level.
- Identity and Authentication: Axonius needed to integrate the agent’s identity flow with its existing authentication and authorization modules, which reside on Amazon Elastic Compute Cloud (EC2) instances allocated to each tenant.
- Cost Tracking: Because model invocation is the primary driver of agentic costs, Axonius prioritized the ability to track these expenses per tenant. This granular tracking is essential for determining pricing for AI-enhanced service offerings.
- Lifecycle and Observability: The company integrated agentic workloads into its existing Continuous Delivery (CD) workflows. To manage a large fleet of agents across hundreds of silos, Axonius required observability tools capable of issuing alarms and providing tracing for debugging.
Industry and User Impact
The integration of AI agents into the Axonius platform is expected to have a significant impact on the productivity of Security and IT operations. By automating the analysis of millions of data points, Axonius claims the system can reduce the manual burden of security, audit, and compliance by as much as 50%.
For developers and architects in the SaaS space, the Axonius use case provides a blueprint for deploying agentic AI in highly regulated or security-conscious industries. It demonstrates how AWS Bedrock AgentCore can be used to scale AI features without abandoning established multi-tenant isolation strategies like the silo model. The use of microVMs within the AgentCore runtime provides a layer of security that addresses the "noisy neighbor" and data leakage concerns often associated with shared AI infrastructure.
What's Next
While Axonius has detailed its architectural choice and the integration of these agents into its SaaS platform, no specific timeline for the rollout of additional AI agents or features was announced. The company continues to use AWS infrastructure to manage its isolated customer environments. Organizations looking to implement similar architectures can reference the Bedrock AgentCore documentation for guidance on siloed and pooled deployment strategies.