What changes
The integration turns Reckoning's Java-native runtime and OpenAI-shaped server API into a Spring AI ChatModel. That gives a Spring application the familiar ChatClient surface while keeping model selection, streaming, tool calls, and local model operation under the developer's control.
Use the client module when Reckoning runs as a service. Use the embedded module when the model should live in the same JVM as the Spring application. The two modes share the Spring-facing programming model, so moving from a local server to in-process inference does not require redesigning the application.
More than a chat adapter
Reckoning can fetch a model into its cache and serve it through its HTTP endpoints. The Spring integration can therefore sit in front of a model lifecycle that is still local and inspectable: fetch the checkpoint, serve the model, then call it through Spring AI. The repository also includes a small pull-request review application that shows the integration in a useful workflow rather than a hello-world prompt.
Start small: Qwen3 0.6B JQ4 is a practical first model because it is compact enough for a laptop and fast enough to make the full Spring-to-model loop easy to see.
Keep going
Read the project README for setup and the Reckoning source repository for the inference engine, model formats, serving API, and runtime internals.