Add docker-model-runner module: Spring AI OpenAI starter against Docker Model Runner and Ollama with a benchmark runner (time to first token, decode tokens/second, container memory, tool-call success); real docker/model-runner v1.2.9 CPU image and Ollama 0.40.13 on a shared 2-vCPU sandbox, no GPU

Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017ic4b4VQVEJWetTT1dYzyJ

README.md

@@ -32,5 +32,6 @@ Runnable companion code for the Spring AI articles on [ankurm.com](https://ankur
 | [`graphrag-neo4j/`](graphrag-neo4j) | Entities and relations extracted with `ChatClient.entity(...)` into a typed Neo4j graph with a vector index on entity names (Testcontainers, `neo4j:2026.10.1-community`; also run on 5.26), then multi-hop retrieval by span linking, variable-length Cypher paths and path ranking, against plain vector RAG over the same 35 chunks and 9 questions: complete evidence 7/9 for the graph, 5/9 for vector k=8, 2/9 for k=4, with the traps reproduced (duplicate entity under an alias, an embedding merge that joins the wrong incidents, a reply with a sentence before the JSON fence, a hostile relation type, hop explosion). **The extraction step is a rule-based stand-in, not a language model**; Neo4j, embeddings and Cypher are real. Spring Boot 4.1.2, Spring AI 2.0.2, Java 25. | [GraphRAG in Java with Neo4j and Spring AI](https://ankurm.com/graphrag-java-neo4j-spring-ai-knowledge-graph/) |
 | [`image-gen/`](image-gen) | `ImageModel` and `ImagePrompt` with `OpenAiImageOptions` (what the real `OpenAiImageModel` puts on the wire, a per-call options object that silently resets the model, a `List<ImageMessage>` prompt that throws), a prompt gate with `ModerationModel` that fails closed, and safe storage in MinIO: sniffed content type, byte and pixel caps, server-chosen keys, a fetcher for provider URLs, presigned reads. A per-image price table from the vendors' pages dated 2026-10-28. No image provider called: the `ImageModel` and `ModerationModel` are hand-written fakes and a mock server; the MinIO tests use a real server through Testcontainers. Spring Boot 4.1.2, Spring AI 2.0.2, Java 25. | [Image Generation and Vision in Spring AI: ImageModel, Prompt Options and Safe Storage](https://ankurm.com/spring-ai-image-generation-imagemodel-s3-minio-safe-storage/) |
 | [`batch-llm/`](batch-llm) | 100,000 synthetic support tickets classified by a `ChatClient` inside a Spring Batch 6 chunk step: a Resilience4j rate limiter and retry around the model call (and where Spring Batch's own retry and skip belong instead), replies that are not valid JSON skipped and recorded, a restart after `kill -9` in a second JVM that ends with a byte-identical result, and a cost ledger row committed with every chunk. **The model is a deterministic simulated `ChatModel`, not a language model; token counts are estimates and prices are configurable placeholders.** Spring Boot 4.1.2, Spring Batch 6.0.6, Spring AI 2.0.2, Resilience4j 2.4.0, Java 25. | [Spring AI + Spring Batch: Classifying 100,000 Support Tickets with an LLM](https://ankurm.com/spring-ai-spring-batch-llm-ticket-classification/) |
+| [`docker-model-runner/`](docker-model-runner) | The Spring AI OpenAI starter pointed at Docker Model Runner and at Ollama by changing only `base-url` and the model name (no `/v1` is appended: the path is used exactly as given), plus a benchmark runner: time to first token, decode tokens/second, container memory from `docker stats` and tool-call success over 10 runs. Measured with the real `docker/model-runner` v1.2.9 CPU image and Ollama 0.40.13 on one shared 2-vCPU sandbox, Qwen2.5-0.5B Q4_K_M; no GPU, and Docker Desktop's `docker model` path was not used. Spring Boot 4.1.2, Spring AI 2.0.2, Java 25. | [Spring AI with Docker Model Runner vs Ollama: Local LLMs Benchmarked on the Same Laptop](https://ankurm.com/spring-ai-docker-model-runner-vs-ollama/) |
 
 Upgrading from Spring AI 1.x: [migration guide](https://ankurm.com/spring-ai-1-to-2-migration-guide/).