LightRAG: inherit GraphRAG retrieval config via lightRAG()

[ronaldmannak]

Follow-up to #2. Makes a LightRAGEngine created via GraphRAG.lightRAG() inherit the owning instance's retrieval Config, so the LightRAG path stays consistent with GraphRAG.ask/search on the same configured graph. (In #2, LightRAG deliberately ignored these settings; this threads them through as the more intuitive default for a factory method on a configured instance.)

What's inherited

Config setting	Effect on LightRAG (via lightRAG())
topKResults	Becomes the engine's defaultTopK, used when retrieve/ask are called without an explicit topK
similarityThreshold	Applied to semantic (cosine) hits in both stores
approach == "semantic"	Drops BM25 (includeKeyword: false) so results are semantic-only
approach == "keyword"	Still rejected — LightRAG's dual-level design requires embeddings

How

• New LightRAGSearchOptions (semanticThreshold, includeKeyword) is threaded into InMemorySemanticSearcher / LightRAG.chunkSearcher / communitySearcher and applied in the HybridRetriever.search call.
• LightRAGEngine gains searchOptions and defaultTopK; retrieve/ask take topK: Int? = nil and resolve to defaultTopK.
• GraphRAG.lightRAG() populates both from self.config.

Constructing LightRAGEngine directly still uses permissive defaults (full hybrid, no threshold, topK 10), so the subsystem remains usable standalone — the inheritance only applies when derived from a configured GraphRAG.

Tests

• lightRAGInheritsConfiguredTopK — withTopK(1) → engine.defaultTopK == 1 and capped results.
• lightRAGSemanticApproachRuns — withApproach("semantic") → includeKeyword == false, engine builds and answers without throwing.

README updated to document the inheritance. Offline-only, builds under Swift 6 strict concurrency.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VUzx3BzstYuB68txcHvAUh

---

Generated by Claude Code

[gemini-code-assist]

Code Review

This pull request updates LightRAGEngine to inherit retrieval settings (such as topKResults, similarityThreshold, and whether to include keywords based on the approach) from the parent GraphRAG instance's configuration, ensuring consistency across retrieval paths. Feedback suggests using Task.detached when building the searchers to offload heavy initialization work to the global concurrent pool, preventing the LightRAGStoreCache actor from being blocked.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.