Agent Router
Agent Router is an open-source, Envoy-powered gateway that gives application teams one consistent, OpenAI-compatible API for hosted and self-hosted models and MCP tools. It lets platform teams centrally manage provider credentials, routing, quotas, failover, authorization, and usage attribution.
At a Glance
- Enterprise platform and infrastructure teams
- Organizations operating multiple AI applications or teams
- Regulated industries and enterprises needing centralized security and governance
- Teams using hosted LLM providers and self-hosted inference together
- +1 more
AI Tools by Agent Router
(1)
Agent Router
Open Source AI Gateway for LLMs
Discussions
No discussions yet
Be the first to start a discussion about Agent Router
Latest News
Agent Router 1.2 released with hardened multi-user MCP security, credential isolation, safer sessions, and production fixes.
Envoy AI Gateway became Agent Router and joined the Agentic AI Foundation; the code, maintainers, APIs, CLI, and license remained the same.
Multi-User MCP in Production: per-user identity at the gateway layer through explicit header forwarding and OAuth guidance.
Envoy AI Gateway 1.0 announced as a stable, production-ready open-source AI gateway with 16 providers and a full MCP gateway.
Products & Services
Apache 2.0 open-source AI and agent traffic gateway built on Envoy Proxy and Envoy Gateway. It provides one OpenAI-compatible interface across model providers, self-hosted inference, and MCP servers.
Standalone CLI that runs the router locally on Linux or macOS with one command and uses the same configuration model that can be deployed to Kubernetes or a gateway service.
Gateway for Model Context Protocol servers with multiplexing, tool routing and filtering, streamable HTTP, OAuth/JWT authentication, CEL authorization, header forwarding, and session-aware handling.
Hosted service built on Agent Router that supplies a managed endpoint for routing to models and MCP tools without customer-managed infrastructure.
Market Position
Agent Router positions itself as a neutral, open-source, Envoy-based traffic and governance layer rather than an agent framework: Agent Router configures providers, credentials, routing, tools, and limits while Envoy carries traffic. Its competitive set includes LiteLLM, Portkey, Kong AI Gateway, Bifrost, and managed AI gateways such as Vercel AI Gateway. Its differentiators are Envoy's production traffic foundation, Kubernetes-native control-plane APIs, multi-provider routing, enterprise policy controls, and first-class MCP gateway capabilities.
Founding Story
The project originated in summer 2024 when Bloomberg engineers identified the need for an internal AI platform built on open-source Envoy and Kubernetes. Bloomberg brought the enterprise traffic-management problem to the Envoy community, and Tetrate joined to turn the vision into an open, multi-company project rather than a feature locked behind a single vendor's enterprise license. Public development began in October 2024 as Envoy AI Gateway; it later became Agent Router and joined the Agentic AI Foundation in September 2026.
Business Model
Revenue Model
The core Agent Router project is free and open source under Apache 2.0. Commercial monetization is provided by ecosystem offerings built on it, including hosted Agent Router services and self-managed enterprise distributions from partners such as Tetrate, Nutanix, and Paper Compute Co.
Target Markets
- Enterprise platform and infrastructure teams
- Organizations operating multiple AI applications or teams
- Regulated industries and enterprises needing centralized security and governance
- Teams using hosted LLM providers and self-hosted inference together
- Developers and operators deploying MCP-based agents
- Centralized enterprise AI platform gateways for multiple application teams
- Multi-provider applications that need switching, A/B testing, or failover without application rewrites
- Governed access to LLMs with quotas, usage attribution, cost control, and provider credentials managed centrally
- Secure MCP-based agent systems requiring per-user authorization, tool filtering, and observability
- Self-hosted model serving clusters using Kubernetes, KServe, vLLM, or inference pools
- Local development and prototyping with an OpenAI-compatible endpoint
- Bloomberg
- Tencent Cloud
- Tetrate
- Nutanix