Tetrate Agent Router Enterprise
This integration is community-maintained. Reference: ory/integrates/agent-identity/tetrate.
Tetrate Agent Router Enterprise is an enterprise AI gateway built on Envoy AI Gateway for routing and controlling AI-agent traffic. Pair it with Ory for dynamic, granular control over enterprise AI agents: Ory OAuth2 & OpenID Connect (Hydra) issues OAuth2/OIDC tokens for agents and Ory Permissions (Keto) holds the fine-grained authorization policies, while Tetrate enforces those policies on live traffic, down to parameter-level control over Model Context Protocol (MCP) tool calls.
This entry reflects an announced partnership. See Tetrate and Ory bring dynamic, granular control to enterprise AI agents.
How it works
Two layers, each owning a distinct concern:
- Authorization / policy layer (Ory) — decides what an agent or user may access at the resource level. AI agents are
first-class identities authenticated through Ory. Ory OAuth2 & OpenID Connect issues OAuth2/OIDC tokens (
client_credentials), and Ory Permissions holds the fine-grained policies. - Network / gateway layer (Tetrate) — enforces those policies on live traffic when agents call models, tools, and enterprise services. When an agent invokes an MCP tool, Tetrate Agent Router Enterprise evaluates the request against the Ory-defined policies.
Because enforcement is parameter-level, Tetrate can gate not only which tools an agent may use but also which request parameters are allowed, based on policies defined in Ory Permissions.
Step-up authorization
If a request exceeds a risk threshold, the system can:
- Pause the request,
- Trigger an authentication and approval flow through Ory, and
- Issue short-lived elevated access.
For example: an agent processes routine refunds automatically but requires approval for larger amounts, or handles standard records access but requires additional authorization for sensitive data.
Notable
- Tetrate Agent Router Enterprise is built on Envoy AI Gateway — see the Envoy integration for the underlying pattern of validating Ory tokens at the edge.
- Closely related to the Skyfire integration: both treat AI agents as first-class Ory identities, differing in where verification and enforcement sit.
- Agent-identity is an emerging space — this integration reflects an announced partnership and is a reference architecture rather than a turnkey product.
