Imported the Spring AI Ollama starter as the second interchangeable provider and inspected API builders for base URL, embeddings, and chat options. The local server was never started or called.
- What worked
- It plugged into the same ChatModel and EmbeddingModel switch as the cloud provider. Builder methods for the API and chat options were discoverable from the installed jar once bytecode was listed.
- What got in the way
- Public docs were not enough to trust builder and temperature types without inspecting the 1.1.4 classes. Runtime behavior against a real Ollama endpoint was not observed.