June 2026 delivered a focused enhancement to the Ollama embedding pathway in spring-ai, adding configurable embedding options to improve control, parity, and performance. The change introduces new fields to OllamaEmbeddingOptions (numCtx, f16KV, logitsAll), enabling fine-grained tuning of embedding requests and aligning the embedding API with the chat API and llama.cpp parameters.

May 2026 monthly summary for mudler/LocalAI focused on stability and performance of the React UI behind reverse proxies. Delivered a comprehensive fix for prefix recovery, asset URL rewriting, safety validation, and test coverage to ensure reliable loading under proxy configurations. This work reduces deployment risk and support overhead for customers running LocalAI behind reverse proxies. Key business value includes improved user experience when proxied paths are stripped, and smoother deployments across environments. Technical accomplishments include implementing a BasePathPrefix-aware flow that honors X-Forwarded-Prefix, rewriting index.html asset URLs to route through the proxy, validating prefixes to prevent unsafe URLs, and expanding test coverage (unit tests for BaseURL with pre-stripped paths and an end-to-end test for proxy-stripped scenarios).

April 2026: mudler/LocalAI delivered a REST-compliance enhancement for the Backend Monitor endpoint. The /backend/monitor GET now accepts ?model=<name> as a query parameter instead of reading from the body, aligning with REST and enabling Swagger/OpenAPI tooling. Commit 607efe5a4c4967b7b3e7024b00e26197576f48dc updated the handler and preserved backward compatibility by falling back to body binding when the query param is absent. Swagger annotations, OpenAPI specs, and docs were regenerated to reflect the change. Impact: easier client integration, improved codegen reliability, and better monitoring tooling support.