What's New

Amazon Bedrock AgentCore Runtime now supports a bring-your-own file system (BYOFS) capability, allowing developers to attach Amazon S3 Files file systems and Amazon EFS access points directly to their agent runtimes. The runtime automatically mounts the specified file system into every agent session at a developer-defined path, enabling standard file I/O operations without any custom mount code, privileged containers, or pre-session download orchestration. This feature complements the existing managed session storage (currently in public preview) by addressing the distinct use case of sharing persistent, pre-existing data across sessions, microVM lifecycles, and multiple agents.

How It Works

• When configuring an AgentCore Runtime, developers supply an access point ARN — either an Amazon S3 Files access point or an Amazon EFS access point — along with a target mount path.
• The runtime must be deployed within a VPC to enable the necessary network connectivity for NFS-based mounting.
• At session initialization, AgentCore Runtime automatically mounts the file system at the specified path inside the agent's microVM, making the data immediately accessible via standard POSIX file operations.
• For S3 Files, changes are bidirectionally synchronized between the mounted file system and the underlying S3 bucket, and the data can also be accessed concurrently via S3 APIs.
• For EFS, the access point provides a shared NFS file system with purpose-built network file storage semantics.
• Both options deliver sub-millisecond latency for active data and enforce NFS close-to-open consistency, ensuring that data written and closed by one client is visible to subsequent opens by other clients.

Why It's Important

• This capability fundamentally changes how AI agents interact with persistent and shared data, eliminating a significant operational burden that previously required custom infrastructure work.
• Agents can now load shared skills, prompt templates, tool libraries, and curated reference datasets at session start without redundant re-downloads on every new session initialization, reducing both latency and egress costs.
• Long-running agentic workflows gain the ability to persist intermediate results and resume across session boundaries or microVM restarts, which is critical for complex, multi-step tasks.
• Perhaps most significantly, it enables genuine multi-agent collaboration patterns where one agent produces outputs that another agent consumes as inputs — all operating on the same shared file system in near real time.

How It's Different

• Prior to this feature, developers who needed persistent or shared file data in AgentCore Runtime had two suboptimal paths: either rely on the managed session storage (still in public preview), which is automatically provisioned but scoped to individual sessions and not designed for cross-session or cross-agent sharing, or implement custom download and orchestration logic before each session could begin working with the data.
• The BYOFS feature is explicitly designed for data the developer already owns and wants to share broadly — across sessions, microVM lifecycles, and multiple distinct agents — whereas managed session storage is optimized for ephemeral, per-session scratch space.
• Additionally, the dual support for both S3 Files (with its S3 API compatibility and automatic sync) and EFS (with its purpose-built NFS semantics) gives developers flexibility to choose the storage backend that best matches their existing data architecture and access patterns.

When to Prefer It

• Choose bring-your-own file system when you have pre-existing datasets, tool libraries, knowledge bases, or prompt templates stored in S3 or EFS that need to be shared consistently across multiple agent sessions or multiple distinct agents without duplication or re-download overhead.
• It is the right choice for long-running workflows that must checkpoint intermediate state and resume in a future session, or for pipeline architectures where agents act as producers and consumers of shared data.
• Prefer S3 Files when your data already lives in S3 and you want the flexibility of both file-system access and S3 API access with automatic synchronization.
• Prefer EFS when you need a high-performance, shared NFS file system with strong consistency guarantees and your workload involves frequent concurrent reads and writes from multiple agents simultaneously.
• Continue using managed session storage for purely ephemeral, per-session scratch data that does not need to persist or be shared beyond a single session's lifecycle.

Availability

• The bring-your-own file system feature is generally available (no preview qualifier is mentioned in the announcement, distinguishing it from the managed session storage feature which remains in public preview).
• It is supported across all 15 AWS Regions where Amazon Bedrock AgentCore Runtime is available; the full region list is published on the AWS Supported Regions page for AgentCore Runtime.
• A key configuration prerequisite is that the AgentCore Runtime must be deployed within a VPC, as VPC connectivity is required to mount NFS-based file systems.
• Developers get started by providing an access point ARN (either an S3 Files access point or an EFS access point) during runtime configuration.