One AI Gateway for Your Whole Estate, on Any Cloud or None
WSO2 fronts Bedrock, OpenAI, Anthropic, Gemini, and Azure OpenAI behind one proxy, and governs human-facing APIs from the same control plane. Deployed wherever your data has to live.
What "Amazon's AI Gateway" actually is
AWS does not market a single product called "Amazon AI Gateway." The capabilities of an AI gateway are split across Amazon Bedrock AgentCore Gateway (a managed MCP server endpoint with semantic tool discovery, GA October 2025) and Amazon Bedrock (the model platform that supplies LLM proxying, Guardrails, Knowledge Bases, and Intelligent Prompt Routing for Bedrock-hosted models). Here, we treat both together as Amazon's AI gateway surface.
| Pillar |
|
Amazon Bedrock AgentCore + Bedrock |
|---|---|---|
| Head-to-head snapshot, by pillar | ||
| Agentic ready | AgentCore Gateway is itself an MCP server endpoint and converts OpenAPI, Smithy, and Lambda functions into MCP tools in a few lines of code; AgentCore Runtime already supports A2A. Neither vendor has a developer-facing discovery hub, but WSO2's MCP Hub adds a catalog, subscriptions, and credential issuance on top of its own MCP proxy. | AgentCore ships MCP-native and A2A-capable today, ahead of WSO2's Q3 2026 A2A roadmap, but with no consumption-side catalog. |
| Universal governance | One control plane governs API, AI, and MCP traffic, including WSO2, AWS, Azure, Kong, and Envoy gateways. | AgentCore is an AI-focused stack; non-AI API governance lives in AWS API Gateway, a separate product with its own model and no shared control plane. |
| Freedom and flexibility | Deploys on any cloud, on-prem, hybrid, or air-gapped, from a single Apache 2.0 codebase. | Bedrock and AgentCore are regional AWS managed services with no self-hosted or on-premises option, even as AWS expands GovCloud and sovereign-cloud coverage. |
| Monetization | Native, Moesif-powered analytics, FinOps, and monetization, battle-tested at Siemens, You.com, and Oxford University Press. | No native capability to resell or monetize AI APIs to your own customers. AWS billing tracks your AWS spend across separately metered components, not a product for billing downstream consumers. |
| Capability | Benefit |
|
Amazon Bedrock AgentCore + Bedrock |
|---|---|---|---|
| Agentic ready | |||
| MCP-native gateway | Turn APIs and functions into MCP tools automatically | Auto-generates MCP servers from existing API specs in a few clicks, with a productized MCP Hub for discovery, subscriptions, and credential issuance. | AgentCore Gateway is itself an MCP server endpoint, converting OpenAPI, Smithy, and Lambda functions into MCP tools in a few lines of code, with built-in semantic search for tool discovery and an open-source MCP server on GitHub. There's no developer-facing catalog or subscription flow. |
| A2A (agent-to-agent) | Govern agent-to-agent traffic | On the AI Gateway roadmap for Q3 2026. | AgentCore Runtime already supports A2A, with a documented protocol contract and A2A server deployment. |
| Design-time AI readiness | Catch APIs that agents cannot use before they ship | AI readiness scoring in the VS Code API Designer, Spectral rulesets, and AI compliance checks against plain-language rules. | No comparable design-time tooling. AgentCore's tooling (starter toolkit, Cedar policy language) is runtime-focused. |
| Universal governance | |||
| Unified API + AI governance | One place to govern human-facing APIs and AI traffic | One control plane governs API, AI, and MCP traffic, including WSO2, AWS, Azure, Kong, and Envoy gateways. | AgentCore is an AI-focused stack; non-AI API governance lives in AWS API Gateway, a separate product with its own model. |
| Guardrails depth | Protect data and prove compliance | 40+ open-source policies in Policy Hub: semantic prompt guard, PII masking, NeMo Guard, Granite Guardian, Bedrock, and Azure Content Safety integrations. | Bedrock Guardrails covers content filters, denied topics, PII filtering, word filters, contextual grounding, and Automated Reasoning checks (GA August 2025). The ApplyGuardrail API extends coverage to non-Bedrock and self-hosted models, and AgentCore Policy adds a Cedar-based deterministic authorization layer (GA March 2026). |
| Identity for agents and users | Know who, and what agent, is calling | WSO2 Identity Server natively secures human APIs and AI traffic with OIDC, SAML, CIBA, SCIM, and CIAM. | AgentCore Policy provides Cedar-based deterministic authorization for agent actions; broader human-facing identity and CIAM live outside AgentCore, in separate AWS identity services. |
| Freedom and flexibility | |||
| Deployment and sovereignty | Run wherever the regulator requires, including air-gapped | Self-hosted, hybrid, and SaaS deployment models, including air-gapped, from a single Apache 2.0 codebase. | Regional AWS managed service with no self-hosted or on-premises option. AgentCore reached AWS GovCloud (US-West) on 5 May 2026, and Bedrock is named among the initial AI services for the AWS European Sovereign Cloud, but there's still no air-gapped path. |
| 100% open source foundation | Meets open-source procurement and sovereignty mandates | Apache 2.0 across the platform with full function, no open core. Monetization is the one SaaS-only exception. | Bedrock and AgentCore are proprietary AWS services. AWS open sources the AgentCore MCP server, starter toolkit, and Cedar policy language, but there's no self-hostable, Bedrock-compatible gateway. |
| Provider and routing choice | Route to any model, on any infrastructure | Single proxy fronts OpenAI, Anthropic, Gemini, Bedrock, Azure OpenAI, and Mistral with shared guardrails, unified token tracking, and one credential vault, on any infrastructure. | Bedrock hosts a broad and growing provider roster (Anthropic, Meta, Mistral, OpenAI, xAI, Cohere, DeepSeek, Writer, Stability AI, Amazon's own Nova and Titan). AgentCore Gateway inference targets front third-party providers directly with pattern-based routing, but the abstraction is AWS-hosted and AWS-operated. Bedrock's own Intelligent Prompt Routing is narrower: it routes within a single model family and needs exactly two models from that family. |
| RAG infrastructure | Retrieve context without locking your data pipeline to one cloud | Native RAG ingestion pipeline with vendor-neutral vector store integrations (Pinecone, Milvus, Weaviate, or an in-memory store). | Bedrock Knowledge Bases is a genuinely broad managed RAG offering, spanning OpenSearch Serverless, OpenSearch Managed Cluster, Amazon S3 Vectors, Aurora PostgreSQL, Neptune Analytics, Pinecone, Redis Enterprise Cloud, and MongoDB Atlas, but it retrieves only for models reachable through Bedrock. |
| Monetization | |||
| Business-level AI analytics | See which customers adopt, which stick, and which will pay more | Moesif-powered adoption funnels, retention, behavioral cohorts, AI Explain, and automated customer guidance. | No product-analytics layer. Observability runs through OpenTelemetry (ADOT SDK), CloudWatch GenAI observability, and CloudTrail, which is operational telemetry, not customer adoption analytics. |
| AI FinOps and chargeback | Understand and chargeback AI cost automatically | Token and USD budgets per user, team, and app; cost dashboards with fully automated chargeback and export to external systems. | No unified budget or chargeback product. Cost tracking means reconciling separate line items: Gateway invocations at $0.005 per 1,000, semantic-search queries at $0.025 per 1,000, tool indexing at $0.02 per 100 per month, Runtime compute, and independent Bedrock token costs. |
| Enterprise-grade AI monetization | Turn AI APIs into revenue centers | Battle-tested with Siemens, You.com, and Oxford University Press. Stripe, Zuora, Chargebee, and AWS Marketplace connectors; prepaid credits and business-level quotas. | No native capability to resell or monetize AI APIs to your own customers. AWS billing tracks your AWS spend, not a product for billing your downstream consumers. |
| Cross-gateway monetization | Monetize the estate you already run, not just one vendor's gateway | Moesif gateway plugins meter Kong and AWS traffic today, billed from one control plane. | Scoped to AWS-billed usage only; no capability to meter or bill traffic running on other vendors' gateways. |
Why teams choose WSO2 AI Gateway
over Amazon's AI stack
Deployment options
Resources
WSO2 AI Gateway
A multi-cloud AI gateway fronting Bedrock, OpenAI, Anthropic, Gemini, and Azure OpenAI.
Deployment Flexibility
Run AI Gateway on any cloud, on-prem, or hybrid, including air-gapped deployments.
API Portal & MCP Hub
Productized MCP Hub for AI-agent tool discovery.
Frequently asked questions
No. AWS splits AI gateway capabilities across Amazon Bedrock AgentCore Gateway (a managed MCP server endpoint, GA October 2025) and Amazon Bedrock itself (model access, Guardrails, Knowledge Bases, and Intelligent Prompt Routing). WSO2 ships one AI Gateway component covering both concerns.
No. Bedrock is a regional AWS managed service with no self-hosted or on-premises deployment option, though AWS continues to expand GovCloud and sovereign-cloud availability. WSO2 deploys identically across any cloud, on-prem, hybrid, or Kubernetes environment, including air-gapped.
No. Intelligent Prompt Routing routes within a single model family and requires exactly two models from that family. WSO2's failover and weighted routing work across any combination of providers.
No. Bedrock and AgentCore are proprietary AWS-managed services. AWS open-sources the AgentCore MCP server, CLI and Cedar policy language, but there is no self-hostable AWS-supplied Bedrock-compatible gateway. WSO2 API Platform's core runtime is fully Apache 2.0.
No. AgentCore is an AI-focused stack; non-AI API governance lives in AWS API Gateway, a separate product with its own model. WSO2 governs human-facing APIs and AI traffic from one control plane, sharing the same identity, lifecycle, and monetization framework.
Yes. The ApplyGuardrail API extends Bedrock Guardrails to self-hosted and third-party models. WSO2's guardrails are native to the gateway and run wherever you deploy it, including on-premises and air-gapped.
Amazon®, AWS®, Amazon Bedrock®, and AgentCore™ are trademarks of Amazon.com, Inc. or its affiliates. All product, company names, and marks mentioned herein are the property of their respective owners and are mentioned for identification purposes only. Last updated August 2026. * Monetization is the one SaaS-only exception to WSO2's Apache 2.0 licensing; every other core component ships fully open source.