lazy-graph-rag · service and data flow

Hybrid seed retrieval plus an LLM-free cached concept graph built lazily from RagBase chunks

lazy-graph-rag service and data flowThe query path combines hybrid seed search with a deterministic concept graph cached by corpus fingerprint on an Atlas volume. ATLAS COMPOSE STACK · RAG PLUGIN + WEAVIATE + NAMED CACHE VOLUMEQUERY PATH · RETRIEVAL AND CACHE CHECK RUN CONCURRENTLY LAZY GRAPH PIPELINE · NO LIGHTRAG · NO NEO4J · NO EXTRACTION LLM 1request2dispatch3embed 4A4B67 5A · read all RagBase chunks5B · cache hit5C · cache miss → build + write8 · answer + sources + cache/index/traversal metadata CallerOpen WebUI / API LiteLLMmodel: lazy-graph-rag Lazy graph pluginorchestrate · assemble · metrics Query embeddingnomic-embed-text Hybrid seed searchseed_k candidates Weaviate · RagBasehybrid seed chunks Budgeted expansionrelevance budget · context cap light_genexpanded evidence → answer Full chunk scanread corpus state RagBase chunkscontent fingerprint input Graph cache volumefingerprint → cached index Deterministic builderconcept + co-occurrence graph MESSAGE CONTRACT, CACHE LIFECYCLE, AND TUNING 1–4 · SEED RETRIEVALquestion → embedding → hybrid seed candidatesseed_k controls graph entry points.Uses the shared RagBase corpus. 5 · CACHE OR BUILDall chunks → stable content fingerprintCache hit reuses the deterministic index.Cache miss builds once and writes the volume. 6 · BUDGETED EXPANSIONseed nodes → relevance-budgeted neighborsmax_context_chunks bounds evidence size.Graph is machine-oriented and query-supporting. 7–8 · GENERATE + RETURNlight_gen receives expanded source chunks.Response exposes cache and traversal metadata.No extraction-model call is required.