# getting-started

Companion code for [Spring AI 2.0 in 10 Minutes: ChatClient on Spring Boot 4.1](https://ankurm.com/spring-ai-2-0-chatclient-boot-4-1/)
on [ankurm.com](https://ankurm.com). One `ChatClient` bean, three endpoints, and a test that
proves `spring.ai.model.chat` switches the provider without touching either.

The deeper walkthrough (what each test caught, the API references) lives in the article's
accordion sections, not in this README -- see the versions callout and the "going deeper" list
at the end of each section.

## Versions

| Component | Version |
|---|---|
| Spring Boot | 4.1.1 |
| Spring AI | 2.0.1 |
| Java | 25 (LTS) |

## Quickstart

```bash
mvn -o test              # runs against a scripted ChatModel, no key needed, ~10s
OPENAI_API_KEY=sk-... ./scripts/run.sh    # the real app, port 8080
```

## Endpoints

| Method | Path | What it shows |
|---|---|---|
| GET | `/api/chat?message=` | `ChatClient.prompt().user(...).call().content()` |
| GET | `/api/chat/as?voice=&message=` | a templated system prompt filled via `.param(...)` |
| GET | `/api/chat/stream?message=` | `.stream().content()`, a `Flux<String>` |

## Switching providers

One property, no code change:

```properties
spring.ai.model.chat=openai   # org.springframework.ai.openai.OpenAiChatModel
spring.ai.model.chat=ollama   # org.springframework.ai.ollama.OllamaChatModel
```

`ProviderSwitchTest` proves this mechanically -- it never calls either provider's server, only
inspects which `ChatModel` bean class the Spring context assembled under each property value.

## Output

| File | Regenerated by |
|---|---|
| `output/01-plain-call.txt` | `ChatClientEndpointsTest` |
| `output/02-system-template.txt` | `ChatClientEndpointsTest` |
| `output/03-streaming.txt` | `ChatClientEndpointsTest` |
| `output/04-provider-switch.txt` | `ProviderSwitchTest` |
| `output/05-real-network-round-trip.txt` | captured manually against the real OpenAI API; see its own header |

`scripts/run-all.sh` regenerates 01-04. 05 is not part of that script -- it needs a live network
path and a real (even if invalid) key, and is included once as evidence that the wiring reaches
OpenAI's own server.