Portus-Gateway
Portus is an open-source Kubernetes gateway written in Rust that carries HTTP, LLM and MCP traffic on one data plane. It implements the Kubernetes Gateway API and adds AI-provider routing, token/call budgets, API keys, MCP federation and OAuth-aware tool routing.
At a Glance
- Kubernetes platform and infrastructure teams
- Organizations operating APIs and microservices
- Teams deploying internal or customer-facing LLM applications
- AI-agent and MCP platform operators
- +2 more
AI Tools by Portus-Gateway
(1)
Portus
Kubernetes AI and MCP Gateway
Discussions
No discussions yet
Be the first to start a discussion about Portus-Gateway
Latest News
Portus published its portus-gateway.dev landing page and updated conformance reporting with the release tag and public contact.
Portus 0.2.12 added standalone mode with ACME certificates, DNS backends, weighted backend traffic and hot-reload fixes.
Portus 0.2.11 added per-unit key budgets, budget/refusal webhooks, trusted keys, key labels, export cursors and fallback-model handling.
Portus 0.2.10 introduced MCP federation with namespaced tool names and multi-server session routing.
Products & Services
Rust implementation of Gateway API v1.6.2 experimental-channel resources, including HTTPRoute, GRPCRoute, TLSRoute, TCPRoute, UDPRoute, ListenerSet and BackendTLSPolicy. The controller compiles CRDs directly into protobuf configuration and streams it to per-Gateway data planes over mTLS gRPC.
Gateway for Anthropic Messages and OpenAI chat-compatible providers. It routes using request-body model/stream fields, swaps client keys for provider credentials, meters JSON and streaming tokens, issues Portus API keys and enforces token/call budgets through the optional ledger.
Streamable HTTP gateway for Model Context Protocol servers, with JSON-RPC method/tool routing, per-key tool allow lists, call budgets, session pinning and OAuth/JWT verification. Federation combines multiple MCP servers behind one endpoint using namespaced tools.
Runs the data plane from a YAML file without Kubernetes, retaining routing, policies and TLS features; supports DNS backends, ACME http-01/tls-alpn-01 certificates and hot reload without restart or dropped connections.
Market Position
Portus positions itself as a high-performance Rust alternative to conventional Kubernetes/API gateways, combining Gateway API routing, AI-provider governance and MCP governance in one data plane. Its published benchmark compares it directly with agentgateway and reports higher throughput in the tested configuration; the architecture also differentiates it from multi-component gateways by keeping key, budget and token checks off the request path and by supporting Rama or Pingora.
Leadership
Founders
Tom Meaney (GitHub: Nodstuff)
Backend Go Engineer; GitHub profile lists NGINX and Cork, Ireland. He is the Portus-Gateway organization’s public contact and the author of the repository’s Portus commits and release/conformance work.
Founding Story
Portus appears to have started as a personal open-source project to provide a high-performance, fully conformant Kubernetes Gateway API implementation in Rust, then expanded into a unified gateway for Kubernetes routes, LLM providers and MCP tool servers. The project emphasizes a single data plane, local request-path decisions and a swappable network stack.
Business Model
Revenue Model
Open-source, self-hosted software distributed under Apache License 2.0. The project publishes Helm/OCI installation artifacts and container images; no hosted service, subscription, paid tier or commercial pricing is described in the official materials.
Target Markets
- Kubernetes platform and infrastructure teams
- Organizations operating APIs and microservices
- Teams deploying internal or customer-facing LLM applications
- AI-agent and MCP platform operators
- Security and platform teams needing local authentication, authorization, rate limits and usage budgets
- Self-hosters running gateways on VMs or Docker/Compose
- Kubernetes ingress and service-to-service HTTP, gRPC, TLS, TCP and UDP routing
- Putting LLM provider access behind organization-owned API keys, model routing and token budgets
- Centralizing and governing MCP tool servers for AI agents
- Federating several MCP servers behind one endpoint
- OAuth-protected agent tools with tenant, group and scope-based authorization
- High-throughput, low-memory self-hosted gateway deployments