Camel 4.23 gives AI agents Kamelet-aware catalog and validation tools

Apache Camel 4.23 will extend its agent-facing catalog and validation tools to understand Kamelets, addressing a weakness exposed when the project asked a local model to build integrations from Camel’s reusable route templates.

In the project’s third benchmark round, Camel maintainer Claus Ibsen reports that the original test model passed only one of 15 Kamelet tasks. After the project made Kamelets discoverable through the Camel catalog tools and taught the validator to check them, the same model passed 47 of 50 benchmark prompts.

Discovery was the missing layer

Kamelets package common integration actions such as connecting to services or transforming data. The benchmark found that having those components in the runtime was not enough: an agent also needed a structured way to find the right template and verify the route it produced.

The Camel post says the updated catalog tooling now exposes Kamelet metadata alongside Camel components. Validation was expanded so an agent can catch invalid Kamelet usage before attempting to run a generated integration.

That change shifted the result beyond simple syntax completion. Ibsen reports that the model authored a Kamelet itself in 19 of 20 relevant attempts, while a staged benchmark that introduced the required capabilities step by step reached a 92% success rate.

What changes in 4.23

The Kamelet-aware discovery and validation changes are scheduled for Apache Camel 4.23. For teams building coding agents or natural-language integration tools around Camel, the practical change is that Kamelets become part of the same machine-readable feedback loop already used for ordinary Camel components.

The benchmark remains a project-authored test of one local-model workflow, not a general measure of agent reliability. Its useful result is narrower: better domain tools can materially improve an agent’s ability to select and validate integration primitives without changing the underlying model.