Higress
Higress is a CNCF Sandbox open-source, AI-native API gateway built on Istio and Envoy. It provides a unified, production-grade entry point and governance layer for LLM APIs, AI Agents, MCP Servers, Kubernetes ingress, microservices, and inference workloads, with both a free self-hosted edition and Alibaba Cloud commercial offerings.
At a Glance
- Enterprises deploying AI applications and Agents
- AI model providers and API platforms
- SaaS companies and developer-platform teams
- Kubernetes and cloud-native platform teams
- +3 more
AI Tools by Higress
(1)
Higress
AI Native API Gateway for LLMs
Discussions
No discussions yet
Be the first to start a discussion about Higress
Latest News
Higress released v2.2.5 with 133 updates, including new features, fixes, optimizations, and MCP protocol compatibility improvements.
Higress published v2.2.4-related updates for the revised MCP protocol, Gateway API 1.6, and Gateway API Inference Extension.
Higress MCP Server updated to support MCP 2026-07-28 while maintaining compatibility with existing protocols.
CNCF announced that Higress had passed the TOC vote and joined CNCF as a Sandbox project, highlighting enterprise AI gateway and Nginx Ingress migration capabilities.
Products & Services
Free Apache-2.0-licensed, self-hosted gateway based on Istio and Envoy. It supports Docker and Kubernetes deployment, an out-of-the-box console, Wasm extensions, LLM API governance, MCP proxying/hosting, ingress, microservice routing, security, and observability.
Managed commercial service built from Higress capabilities, providing Serverless Standard, enterprise, and dedicated/premium options for LLM APIs, Agents, and MCP Servers.
AI-focused gateway layer for unified access to 100+ mainstream models, protocol conversion, multi-model routing and fallback, token governance, semantic caching, content safety, authentication, and observability.
Gateway and hosting capabilities for native MCP services and HTTP-to-MCP conversion, with unified authentication, rate limiting, audit logs, observability, protocol translation, and tool routing.
Market Position
Higress positions itself as an Envoy/Istio-based, production-oriented AI-native gateway rather than only an LLM proxy. Its differentiation is the combination of Kubernetes ingress, microservice gateway, LLM governance, MCP hosting, Wasm extensibility, Gateway API/Inference Extension support, and an operational console. Official comparison materials identify OneAPI and LiteLLM as AI-gateway alternatives; broader gateway competition includes APISIX, Kong, NGINX Ingress, Envoy Gateway, Spring Cloud Gateway, and OpenRouter. Higress emphasizes high-throughput, long-connection-safe configuration updates, enterprise security, and unified model/tool governance.
Leadership
Executive Team
Yiquan Dong
Project Maintainer
Project-wide governance, technical direction, releases, and community stewardship; affiliated with Trip.com.
Yuanxiao Zhao
Project Maintainer
Project-wide governance, technical direction, releases, and community stewardship; affiliated with Alibaba Cloud.
Founding Story
Higress originated inside Alibaba to solve production problems caused by Tengine reloads disrupting long-lived connections and by insufficient gRPC/Dubbo load balancing. The project was subsequently open-sourced and is now governed as a vendor-neutral CNCF project; CNCF records the first commit as October 27, 2022.
Business Model
Revenue Model
The community edition is free and open source. Commercial revenue comes from Alibaba Cloud AI Gateway managed services, including usage-based Serverless Standard, production enterprise service, and dedicated/premium deployments with managed operations, dedicated endpoints, custom domains, encryption, and SLA/support.
Pricing Tiers
Self-hosted open-source edition with community support and documentation.
Fully managed, request-based automatic scaling; intended for trials and PoCs.
Production tier with 99.95% SLA, dedicated endpoints, custom domains, and transport encryption.
Dedicated/premium commercial support, elastic resources, and negotiable SLA for larger workloads.
Target Markets
- Enterprises deploying AI applications and Agents
- AI model providers and API platforms
- SaaS companies and developer-platform teams
- Kubernetes and cloud-native platform teams
- Microservice and API-platform operators
- Financial services, retail, education, healthcare, travel, manufacturing, and human-resources organizations
- Enterprise AI gateway for LLM, Agent, and MCP traffic
- Kubernetes ingress controller and migration from ingress-nginx
- Microservice gateway for REST, gRPC, Dubbo, and service-registry-based routing
- MCP Server hosting, API-to-MCP conversion, and private MCP marketplaces
- Inference-aware routing for GPU/model workloads
- Security, authentication, WAF, content safety, and API governance
- Alibaba / Alibaba Cloud: Qwen applications, Bailian model APIs, PAI, and internal HSF-to-MCP deployment
- Ctrip: AI gateway for multi-model and MCP access, authentication, traffic control, and model mapping
- Junrun HR: MCP-enabled digital employees, with more than 1,000 digital employees reported in the case study
- SOFA Community: SOFA AI Gateway built on the open-source Higress kernel