Agentgateway wants to be the traffic cop for AI’s increasingly busy roads

Solo.io has introduced agentgateway, an open-source project that brings AI traffic and ordinary API traffic under one gateway. It supports protocols including HTTP, gRPC, MCP and A2A, giving teams a shared control point for the services and agents now queuing up to talk to one another.

Watch Desk analysis

What happened

Solo.io says agentgateway is built in Rust with Tokio and Hyper, technologies chosen for high-performance network services. The project is designed to handle large-language-model workloads alongside conventional application traffic, rather than forcing organisations to maintain separate gateways for each new AI-shaped protocol.

The company has donated agentgateway to the Agentic AI Infrastructure Foundation. Solo.io says the move is intended to provide vendor-neutral governance and a common home for work on enterprise AI infrastructure, security and observability.

Why it matters

AI systems are acquiring more ways to communicate, including model-context protocol connections and agent-to-agent exchanges. That is useful, but it also creates a thicket of identity, routing, policy and monitoring problems. A gateway that can see these different kinds of traffic could give operators one place to apply controls and inspect what is happening.

The practical promise is less glamorous than a new model release, which is precisely why it matters. When an agent calls a tool, hands work to another agent or reaches a conventional API, somebody still has to decide whether that request is allowed, where it goes and how its performance is measured. Infrastructure tends to become important just after everyone discovers they need it.

Our read

Agentgateway is a credible answer to a real plumbing problem: AI traffic is becoming too varied to manage as a collection of one-off integrations. Donating the project to a foundation is also a sensible attempt to avoid making a piece of shared infrastructure look like a single vendor’s toll booth.

The harder test comes next. “Open source” and “vendor-neutral” describe the starting arrangement, not proof that the project will attract broad adoption or make production systems safer. Teams should look for independent contributors, clear policy controls, useful observability and evidence that the gateway can handle agent traffic without becoming another operational bottleneck.

What to watch

• Foundation governance:
  whether contributors and adopters extend beyond Solo.io.
• Protocol coverage:
  how reliably agentgateway handles MCP, A2A and established API traffic together.
• Security controls:
  whether identity, authorisation and traffic inspection are practical in production.
• Operational evidence:
  whether users publish results on latency, scale and failure handling.

Discussion spark: Should AI infrastructure be governed through neutral open-source foundations by default, or is a strong commercial steward more likely to make gateways secure and usable?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.