An AI gateway for agents, models, and MCP servers
Tetrate Agent Router is the commercial edition of Agent Router, the open source AI gateway built on Envoy, now a project of the Agentic AI Foundation. Tetrate led the work to bring Envoy to AI builders and remains a core maintainer.
Applications point at one endpoint that speaks both the OpenAI and Anthropic APIs, and agents at one Model Context Protocol (MCP) endpoint. Route to any model, fail over automatically, enforce rate limits and guardrails, track spend against budgets, and attribute spend by team and agent. Control which MCP tools each agent can reach, and manage policies centrally across a fleet of AI gateways. Provider-neutral, from a quick trial through enterprise rollout.
Get started by product
Already running open source Agent Router? Compare features →Agent Router Service
Multi-tenant and fully hosted by Tetrate. Nothing to install.
Best for trying routing quickly, with no infrastructure.
Agent Router Enterprise,
Fully Managed
Team-scale governance on a dedicated instance, hosted end to end by Tetrate.
Best for getting admin controls without running infrastructure.
Agent Router Enterprise,
Self-Hosted Data Plane
Same controls and dedicated management plane, but with data path on self-managed infra.
Best for teams that must keep data on premises.
Choosing the right Agent Router product
Within Agent Router Enterprise, Fully Managed and Self-Hosted Data Plane share the same features and differ only in where the data plane runs: Fully Managed is hosted end to end by Tetrate, and Self-Hosted Data Plane keeps AI traffic on customer-managed infrastructure while Tetrate still hosts the management plane.
Agent Router (open source) | Tetrate Agent Router Service | Tetrate Agent Router Enterprise | |
|---|---|---|---|
| Who runs the gateway | You, in your own Kubernetes cluster | Tetrate, multi-tenant | Tetrate (Fully Managed) or you (Self-Hosted Data Plane) |
| Configured with | Kubernetes CRDs and YAML | Developer console and API | Admin console, API, and CLI |
| AI gateway with multi-model routing and auto-failover | ✓ | ✓ | ✓ |
| OpenAI-compatible and Anthropic APIs: works with existing code | ✓ | ✓ | ✓ |
| Connect to 200+ models across provider families | ✓ | ✓ | ✓ |
| MCP gateway: connect agents to tools securely | ✓ | ✓ | ✓ |
| Token-based rate limits | ✓ | ✓ | ✓ |
| Per-key token usage and cost logs | ✓ | ✓ | |
| Usage dashboards and request logs in a UI | ✓ | ✓ | |
| Hosted management plane, nothing to install | ✓ | ✓ | |
| Cross-team cost attribution, showback, and chargeback | ✓ | ||
| Budgets and spend ceilings by team | ✓ | ||
| Admin controls: model and MCP access profiles by team | ✓ | ||
| Runtime AI guardrails: PII redaction and policy enforcement | ✓ | ||
| Enterprise SSO: every request carries authenticated identity | ✓ | ||
| Central policy across a fleet of gateways | ✓ | ||
| Audit trail and SIEM export | ✓ | ||
| Upgrades, support, and SLA | Community | Slack support | Enterprise with 1 hour SLA |
Connect to any provider and tool
One endpoint fronts every model provider and slots into the SDKs, agents, and observability stack already in use.
Supported providers
The 200+ models and provider families the gateway routes to, how each provider is authenticated, and where the live model catalog is published.
Browse providers →Integrations
Everything that plugs into the gateway: SDKs and frameworks, coding agents, guardrail vendors, observability backends, and MCP clients.
Explore integrations →