The Agentic AI Foundation (AAIF) today revealed that an open source stateful artificial intelligence (AI) gateway has been added to the list of projects the consortium now governs.
While AI gateways play a critical role in providing access to backend services, the core technology itself doesn’t provide any differentiated value, says Mazin Gilbert, executive director of the AAIF. Instead of requiring IT vendors to build the same gateway functionality multiple times over, the time has come for multiple vendors to standardize on an open source agentgateway project that provides a unified data plane to centralize the management of HTTP and gRPC traffic, he adds.
At the moment, there are simply too many fragmented approaches to enabling interoperability across the AI stack, notes Gilbert. In contrast, the agentgateway project has already drawn support from Microsoft, Cisco, Apple, Amazon Web Services (AWS) and Solo.io. The core goal of the platform-agnostic project is to support routing, federation, and traffic management across Model Context Protocol (MCP) and Agent-2-Agent (A2A) protocols in a way that makes it simpler to switch between large language models (LLMs).
Additionally, the project provides access to a set of security controls that address JWT authentication, API key auth, role-based access controls, external authorization, mTLS encryption, cross-origin resource sharing (CORS) and protections against malicious tool behavior along with access to metrics, tracing, and logs that can be used to observe AI application and AI agent workflows.
Finally, the AI gateway will include support for providing access to multiple federated MCP servers, dynamic configuration and updates using the Common Expression Language (CEL) and governance tools to enable rate limiting, content-based routing, prompt guards, budget controls, and model aliasing.
Built in the Rust programming language to improve security, the expectation is there will be multiple distributions of the AI gateway that individual vendors will support in much the same way there are multiple distributions of Linux and Kubernetes, says Gilbert. “An AI gateway is a key part of the AI infrastructure stack,” he says. “But at the end of the day it’s just a gateway.”
As an arm of the Linux Foundation, the AAIF emerged late last year to centralize governance across multiple open source AI projects. There are now more than 200 members of the AAIF, which makes it the fastest growing open source consortium to have ever been launched, notes Gilbert. “Our goal is to harmonize the industry around a set of open standards,” he adds.
For example, the AAIF, in addition to standardizing AI gateways and overseeing the advancement of MCP and A2A, is also working on defining a set of standards that will enable AI agents to conduct transactions, notes Gilbert.
Despite widespread adoption of AI tools and platforms, it’s still early days. When it comes to AI infrastructure there are few to none de facto standards. That means inevitably there are a lot of redundant initiatives that have been launched by organizations that are, in the final analysis, trying to address the same set of core fundamental requirements. It remains to be seen how many of those efforts will be subsumed into a larger open source project, but the one thing that is clear is that pressure to reduce the total cost of AI will lead to an inevitable rationalization of many of those initiatives.
