Amazon Bedrock Agents
AWS's fully managed platform for building autonomous AI agents inside Amazon Bedrock — with RAG via Knowledge Bases, tool calling via action groups (Lambda/OpenAPI/MCP), multi-agent collaboration, long-term memory, and sandboxed code interpretation.
Regional availability
- us-east-1
- us-west-2
- us-east-2
- eu-central-1
- eu-west-1
- eu-west-3
- eu-north-1
- ap-northeast-1
- ap-northeast-2
- ap-southeast-1
- ap-southeast-2
- ap-south-1
- ca-central-1
- sa-east-1
- AWS GovCloud (US)
Description
History and evolution
AWS announced Agents for Amazon Bedrock as a preview in July 2023, and the service reached general availability in November 2023 (re:Invent) alongside Knowledge Bases. In 2024, AWS added short-term session memory, and in December 2024 — multi-agent collaboration (a supervisor agent coordinating specialized sub-agents). In 2025, AWS extended the offering with Amazon Bedrock AgentCore — a complementary runtime for deploying agents built in any open-source framework (LangGraph, CrewAI, Strands Agents) with isolated execution sandboxes. Bedrock Agents (managed builder) and AgentCore (deploy/operate) are positioned as two layers of the same AWS agent offering.
Agent architecture
An agent is defined by a foundation model (Claude, Llama, Mistral, Nova, DeepSeek available in Bedrock), instruction prompts, and action groups — sets of APIs/functions delivered as Lambda + OpenAPI schema or return-of-control. RAG is handled by Bedrock Knowledge Bases — managed RAG with Titan/Cohere embeddings and support for vector stores: OpenSearch Serverless, Aurora PostgreSQL (pgvector), Pinecone, Redis Enterprise, MongoDB Atlas. A multi-agent supervisor orchestrates sub-agents via LLM-based reasoning or deterministic routing rules.
Security and enterprise
Bedrock Agents runs inside the customer's VPC with VPC Endpoints (AWS PrivateLink), supports KMS encryption (CMK), IAM authentication, Guardrails (content filtering, PII redaction, prompt injection protection), and Bedrock Model Invocation Logs to CloudWatch / S3. Code Interpretation runs in an isolated Firecracker sandbox. Available in AWS regions including us-east-1, us-west-2, eu-central-1, ap-northeast-1, and AWS GovCloud.
MLOps LifecycleMLOps LifecycleFull model lifecycle: registry, feature store, prompt management, monitoring and human-in-the-loop.
Prompt Management
Monitoring
Human-in-the-Loop
Data & KnowledgeData & Knowledge ManagementData connectors, vector database integration, native vector search and data management (PII, provenance, synthetic data).
ApplicationsAI ApplicationsDomains and use cases this platform is best suited for — from RAG and fine-tuning to scientific research.
SecurityEnterprise SecurityCertifications, access controls and data-protection features essential for corporate deployments and cloud privacy compliance.
Developer EcosystemDeveloper EcosystemDeveloper resources: available SDKs, supported programming languages, and infrastructure features and model-deployment methods.
Pricing & Business ModelPricing & Business ModelBilling models (usage-based, provisioned throughput), resource limits and SLA parameters (uptime, support tiers).
Pricing models
Resource quotas
SLA & Support