Amazon Bedrock AgentCore
App : A product you use directly
AWS managed runtime to put an agent into production: per-session isolated execution, memory, tool gateway, identity, browser, code sandbox, and traces.
Accepts any framework (Strands, LangGraph, CrewAI, ADK...) and any model. Check the region: Frankfurt offers all services, Paris only a subset.
Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.
What it does, concretely
Putting an agent into production requires far more than writing it: you need to manage session isolation, permissions on every tool call, memory across conversations, decision traceability, and scaling. Amazon Bedrock AgentCore handles this infrastructure so teams can focus on the agent's business logic.
The platform brings together modular services that can be used together or independently: a serverless runtime with per-microVM isolation, short- and long-term memory shareable across agents, a gateway that converts APIs or Lambda functions into MCP-compatible tools, a cloud browser, a code execution sandbox, an identity service compatible with existing providers (Cognito, Okta, Azure Entra ID), and an observability layer in OpenTelemetry format. Newer services also cover automated evaluation, A/B test optimization, access policy management, and an agent registry.
Amazon Bedrock AgentCore is aimed at developers and engineering teams deploying agents in production on AWS, regardless of the framework used (LangGraph, CrewAI, Strands, LlamaIndex, OpenAI Agents SDK, Google ADK) and regardless of the model (Claude, Gemini, GPT, Nova, Llama, Mistral).
This service is not suited to those looking for a no-code solution or a consumer-facing interface: it is developer infrastructure, accessible via CLI, API, and SDK, which requires familiarity with AWS and agent development.
What it offers
- ·Serverless runtime: each agent session runs in an isolated microVM with fast cold starts, asynchronous workload support, and multi-agent capabilities. (paid)
- ·Harness: a managed agent loop that orchestrates model calls, tool execution, memory, and response generation through a single API call. (paid)
- ·Memory: short-term storage for multi-turn conversations and long-term storage persisting across sessions, shareable across multiple agents. (paid)
- ·Gateway: converts APIs, Lambda functions, or existing MCP servers into agent-accessible tools, with built-in authentication and access control management. (paid)
- ·Cloud browser: a managed browsing environment that lets agents interact with websites, fill forms, and extract information. (paid)
- ·Code Interpreter sandbox: isolated execution of Python, JavaScript, and TypeScript code for analysis or generation tasks. (paid)
- ·Observability: detailed per-step agent traces in OpenTelemetry format, integrated with Amazon CloudWatch for debugging and production monitoring. (paid)
- ·Evaluation and optimization: automated agent quality assessment and A/B testing service for system prompts and tool descriptions. (paid)
In practice
- Price
- Paid
- Commercial use
- Commercial use: yes
- Available on
- cloud
- FR interface
- No
Security and data
- Rating
- Security: one caveat, see the card
- Your data
- Data with the vendor
- Maintenance
- Active
Links and repository checked on 18 October 2026 · added on 27 September 2026
How to install Amazon Bedrock AgentCore
pip install bedrock-agentcore-starter-toolkit puis agentcore configure et agentcore launch (compte AWS requis)