Powerful traffic handling, made usable
Your agent speaks one API. You describe providers, credentials, tools, and limits once; Agent Router turns that into Envoy configuration, and Envoy handles every request with the reliability it has proven in production for a decade.
One integration point for your agent
Provider APIs, credentials, failover, tool wiring, and token limits move out of your agent and into configuration — on the traffic handling Envoy has proven in production for a decade, without requiring Envoy expertise.
One API, every provider
Route OpenAI-compatible requests to Anthropic, Bedrock, Vertex, Azure, or self-hosted vLLM. Switching models is changing the model name.
Traffic management
Provider fallback, model name virtualization, and token limits per team, app, or model. Your agent never counts a token.
Credentials stay out of your code
Your agent authenticates to the router and nothing else. Provider API keys and AWS, GCP, and Azure credentials stay with Agent Router, rotated in one place.
One router for every MCP tool
One tool catalog from many MCP servers, filtered by who is asking — sized for each agent, so context stays small and wrong-tool calls stay rare.
Inference-aware routing
Route to self-managed models with Kubernetes InferencePool support and inference-aware endpoint picking.
Observability
See what your agent actually did — which model answered, how many tokens, time to first token, where a fallback kicked in — with no instrumentation in your agent. OpenTelemetry GenAI conventions.
From laptop to production
One command on your laptop. The same configuration ships to a dedicated gateway or a Kubernetes cluster — only the packaging changes.
Read the guide# 1 — start the router locally OPENAI_API_KEY=sk-... aigw run # 2 — the only change your agent needs export OPENAI_BASE_URL=http://localhost:1975/v1 # 3 — add MCP tools (same mcpServers file as Cursor / Claude Desktop) aigw run --mcp-config mcp-servers.json
Latest from the project
Envoy AI Gateway is becoming Agent Router, an Agentic AI Foundation project
On 10 September, Envoy AI Gateway officially joins the Agentic AI Foundation as Agent Router. Same code, same maintainers, same APIs. Here is what changes, what does not, and where to find us.
Multi-User MCP in Production: Per-User Identity at the Gateway Layer
How Envoy AI Gateway's header forwarding gives MCP tool calls per-user identity without OAuth infrastructure — and when you need OAuth instead.
Announcing Envoy AI Gateway 1.0 — A Stable, Production-Ready AI Gateway
Envoy AI Gateway 1.0 is here — the first stable release of the open source AI gateway built on CNCF Envoy Gateway. A committed-stable control-plane API, 16 AI providers, an MCP gateway, multimodal support, and enterprise observability, all backed by a growing community.
Built in the open
Weekly meetings, public design proposals, and maintainers from four companies. Showing up with a production problem is a valuable contribution.