The Agentic AI Foundation has added agentgateway as a hosted project under the Linux Foundation.
agentgateway is an open-source gateway for traffic linked to AI and software systems, including Model Context Protocol, agent-to-agent communication, large language model inference, and HTTP and gRPC services. It is released under the Apache 2.0 licence.
The move makes agentgateway the fourth hosted project in the Agentic AI Foundation, which positions itself as a neutral home for standards, protocols and software related to so-called agentic AI systems. These systems are increasingly built as networks of tools, models, databases and services that interact across multiple environments.
As companies test and deploy these systems in production, they face infrastructure challenges that differ from conventional web traffic management. These include routing requests between models and tools, monitoring activity across workflows, and applying security and governance rules when agents interact with external services.
agentgateway is intended to provide a single layer for managing that traffic. According to the project description, it allows organisations to run AI-related traffic and traditional application traffic through one operational setup instead of maintaining separate infrastructure stacks.
This allows platform teams to apply existing controls, such as authentication, observability, routing rules and governance policies, across both conventional applications and AI workloads. The project supports JWT authentication, API key authentication, role-based access control, external authorisation, mutual TLS and CORS, along with logging, tracing and metrics.
It also supports MCP and A2A traffic management, model switching across providers, policy controls using Common Expression Language, and a feature described as MCP virtualisation, which combines multiple MCP tool servers into a single access point. The software can run on bare metal, virtual machines, containers and Kubernetes environments, and supports dynamic updates without downtime through xDS.
The foundation said the project strengthens the broader ecosystem by adding an infrastructure layer for AI traffic management, security and interoperability. It joins other foundation work focused on standards and tooling for agent-based systems.
Community backing is already substantial. The project says it has more than 300 active contributors from more than 60 organisations, including CoreWeave, Red Hat, Solo.io, Adobe, Salesforce, Amdocs and Microsoft.
One notable aspect is the project's focus on model independence. As businesses use different model providers or combine proprietary and open-weight models, infrastructure that can route traffic across them without locking teams into a single supplier has become more important.
Another is governance at the tool and model level. The software includes rate limiting, budget controls, prompt guards, content-based routing and model aliasing, reflecting broader business concerns about controlling costs and reducing risk as AI agents gain greater autonomy in connecting to other systems.
Industry backdrop
The addition of projects such as agentgateway also highlights how the AI software market is moving beyond model development into operational tooling. While much of the earlier industry focus centred on model performance, companies are now paying more attention to the layers needed to deploy and manage AI systems in production.
That has created demand for software that sits between models, applications and tools, much as earlier generations of infrastructure software emerged to manage APIs, cloud workloads and microservices. In this case, the challenge is broader because AI systems can trigger actions, call external tools and coordinate with other agents rather than simply respond to user requests.
David Soria Parra commented on what the addition means for the foundation's project base. "The addition of agentgateway marks a major milestone for the AAIF ecosystem. As organizations deploy increasingly sophisticated AI systems, they need infrastructure that can provide visibility, governance, and operational control across agent workflows and tool interactions. agentgateway helps fill that gap with an open, high-performance platform designed specifically for these emerging workloads," said David Soria Parra, Chair, AAIF Technical Committee.
Lin Sun said the decision reflected the project's original direction. "Donating agentgateway to AAIF is the natural next step for a project built around open connectivity and interoperability. Our goal from day one has been to help organizations manage the operational realities of AI systems - security, governance, observability, and reliability - through open infrastructure. Bringing the project into AAIF ensures it can evolve through open collaboration and neutral governance alongside the broader ecosystem," said Lin Sun.