Preview availability and limitations

Amazon Bedrock Managed Agents, powered by OpenAI is available in preview through regional bedrock-mantle endpoints in US East (N. Virginia), US West (Oregon), and US East (Ohio).

Supported workflows in this guide

• Create, retrieve, list, and delete sessions.

• Submit text messages and cancel the current turn.

• Read durable items and stream progress events.

• Execute commands on self-hosted compute or AgentCore Runtime.

• Discover filesystem-based skills.

• Use environment-based STDIO MCP servers.

Preview boundaries

Area	Preview behavior
Endpoint	Use bedrock-mantle. The BMA preview does not use the bedrock-runtime endpoint.
Models	Use BMA-supported OpenAI models available in the target Region and account.
Cross-Region inference profiles	Not supported by this BMA preview workflow.
Input	The documented session input surface is text.
Subagents	Not supported.
Programmatic tool calling / code mode	Not included in the supported preview workflow.
Dedicated turn-list and turn-retrieval APIs	Not available in the deployed preview used by these examples. Correlate turn IDs through events and items.
Long-term memory	No built-in long-term memory integration. Session conversation and workspace files have separate lifecycles.
Service-managed session-data KMS key	The preview does not expose a customer-managed key setting.
Console	These procedures use APIs and supplied deployment examples.

Acceptance of a configuration field does not establish that its feature is supported for a preview workload. Use the documented surface and test compatibility before relying on additional provider APIs.

Example configuration limits

These values describe the example and its API schema. They are not a general statement of account-level service quotas.

Setting	Value
Item-list page size	1–100; default 20
Capability-directory entries	At most 32 unique paths
Example result polling interval	2 seconds, configurable with BMA_POLL_INTERVAL
Example result polling timeout	300 seconds, configurable with BMA_POLL_TIMEOUT
AgentCore example idle timeout	28,800 seconds
AgentCore example maximum compute lifetime	28,800 seconds

Account request limits, model capacity, token limits, and underlying resource quotas also apply. A larger polling timeout does not increase a service quota or extend an execution environment's maximum lifetime.

Compatibility

Use a compatible exec-server version and the request shapes documented for BMA. OpenAI-hosted Agents APIs and BMA do not necessarily expose the same features or change at the same time.

The public preview retains the bedrock-mantle:CreateAgentSession style of IAM action names shown in security. When upgrading examples or client libraries, review API and permission changes together.