Agent Router
Open source control plane that gives applications one OpenAI-compatible API for LLM providers and MCP tools, built on Envoy.
At a Glance
Agent Router core is open source under the Apache 2.0 license and can be self-hosted at no cost. Hosted and enterprise offerings from third parties are separate and not priced here.
Engagement
Available On
Alternatives
Listed Oct 2026
About Agent Router
Agent Router is an open-source project of the Agentic AI Foundation, formerly known as Envoy AI Gateway, built on Envoy Proxy and Envoy Gateway. It gives application teams a single OpenAI-compatible API for hosted and self-hosted models and for MCP tools, while platform teams manage providers, credentials, routing, quotas and failover centrally. The latest listed release is v1.2.0, and the code is licensed under Apache 2.0.
What It Is
Agent Router is a gateway and control plane for AI and agent traffic. Providers, credentials, tools and limits are described once in configuration, which Agent Router turns into Envoy configuration; Envoy then carries the traffic. The docs state it is not an agent framework and does not tell an agent how to reason. Its value is clearest when AI traffic crosses provider, team, security, cost or operational boundaries.
How It Works
An agent points its OpenAI-compatible client at the router, so switching models means changing the model name. The documented capabilities include:
- Routing to providers such as Anthropic, AWS Bedrock, Vertex AI, Azure OpenAI and self-hosted vLLM.
- Provider fallback, model name virtualization, and token limits per team, app or model.
- Provider credentials kept in the router rather than in agent code.
- A single MCP tool catalog drawn from many MCP servers and filtered by caller identity.
- Inference-aware routing with Kubernetes InferencePool support.
- Observability following OpenTelemetry GenAI conventions.
Setup Path
On a laptop, running aigw run with a provider API key starts a local router at localhost:1975, and MCP servers can be added with a mcp-servers config file. The same configuration can be deployed on Kubernetes with Envoy Gateway. A hosted option is offered by Tetrate, and the site lists other commercial solutions built on the project.
Current Status
The project was renamed from Envoy AI Gateway; the docs state that the CRDs, API group, aigw CLI, container images and Helm charts are unchanged, so there is nothing to migrate.
Community Discussions
Be the first to start a conversation about Agent Router
Share your experience with Agent Router, ask questions, or help others learn from your insights.
Pricing
Open Source (Self-hosted)
Agent Router core is open source under the Apache 2.0 license and can be self-hosted at no cost. Hosted and enterprise offerings from third parties are separate and not priced here.
- Apache License 2.0
- Self-hosted via the aigw CLI or on Kubernetes with Envoy Gateway
- OpenAI-compatible API routing to 17 providers
- Traffic management, provider fallback and token limits
- MCP tool routing
Capabilities
Key Features
- OpenAI-compatible API across LLM providers
- Provider fallback and failover
- Model name virtualization
- Token limits per team, app or model
- Centralized upstream credential management
- MCP gateway with a unified, identity-filtered tool catalog
- Inference-aware routing with Kubernetes InferencePool support
- OpenTelemetry GenAI observability
- Standalone CLI (aigw run) and Kubernetes deployment
- Two-tier gateway pattern for self-hosted model clusters
