Portkey launches Agent Gateway for production AI governance
Portkey has launched Agent Gateway in beta, a production infrastructure layer that gives enterprises centralized governance, observability, and access control over AI agents and their MCP server calls.
Score breakdown
Teams shipping autonomous agents can replace ad-hoc, hand-rolled governance patches with a single production gateway that enforces access control, budget limits, and security guardrails — including full MCP call tracing — without touching existing agent or client code.
- 01Portkey's existing AI Gateway processes over 1 trillion tokens per day across thousands of teams.
- 02Agent Gateway is now available in beta as a production layer for autonomous AI agents.
- 03Agents are registered to receive a governed Portkey endpoint; no changes to agent or client code are required.
Portkey, whose AI Gateway already handles over 1 trillion tokens per day for thousands of teams, has released Agent Gateway in beta — a purpose-built production layer for autonomous AI agents. According to author Siddharth Sambharia, the motivation stems from a shift in enterprise AI: teams are no longer shipping simple LLM-backed apps but are instead deploying agents that chain tool calls, consume MCP servers, and operate autonomously over longer timescales with real-world consequences. The recurring problem every enterprise hits is the same: no unified answer to who can call an agent, what it costs, what it did, and who is responsible when something goes wrong.
Agent Gateway addresses this by letting teams register any agent and receive a governed Portkey endpoint.
Agent Gateway addresses this by letting teams register any agent and receive a governed Portkey endpoint. All traffic flows through a single production stack from that point, with no changes required to existing agent or client code. The feature set includes scoped governance and access control per agent, team, and user; budget and usage limits enforced org-wide with instant policy updates (no redeployment needed); complete traces across every agent run including every MCP call; 40+ metrics out of the box; automatic fallbacks and load balancing to absorb provider outages; and 50+ security guardrails covering PII detection, PHI filtering, and content moderation. An Agent Registry provides each agent with a permanent governed endpoint, its own access controls, and a Skills and Capabilities definition — functioning as a service catalog with auth and RBAC native to every entry.
Agent Gateway is available in beta today, with a technical walkthrough webinar scheduled for Friday, April 24, covering architecture, access control, and policy enforcement.
Key facts
- 01Portkey's existing AI Gateway processes over 1 trillion tokens per day across thousands of teams.
- 02Agent Gateway is now available in beta as a production layer for autonomous AI agents.
- 03Agents are registered to receive a governed Portkey endpoint; no changes to agent or client code are required.
- 04Governance features include scoped access control per agent, team, and user, with instant org-wide policy enforcement and no redeployment needed.
- 05Full observability includes complete traces across every agent run — including every MCP call — plus 40+ out-of-the-box metrics.
- 06Security guardrails cover PII detection, PHI filtering, and content moderation, with 50+ guardrails available without custom setup.