n8n-adaptive-rag · service and data flow

Workflow classifier routes each request to an existing vanilla or agentic endpoint and preserves its evidence

n8n-adaptive-rag service and data flowA plugin adapter invokes an n8n webhook that classifies a query, routes it to vanilla or agentic RAG, and shapes the downstream response. N8N WORKFLOW · SEEDED AND RUN INSIDE THE ATLAS COMPOSE PROJECTCLASSIFY → ROUTE → INVOKE → SHAPE 6A · simple6B · complex8 · normalized answer + downstream sources/trace + adaptive route metadata CallerOpen WebUI / API LiteLLMmodel: n8n-adaptive-rag Adaptive adaptertimeout · normalize · metrics Webhookinput: {query} ClassifyLiteLLM → simple | complex Routeselect target alias vanilla-ragdense retrieval endpoint agentic-ragtool-selection endpoint Shapepreserve answersources · traceroute metadata 1request2dispatch3POST {query} 4classify5simple | complex MESSAGE CONTRACT, ROUTING, AND STATE 1–3 · ENTER WORKFLOWOpenAI chat request → adapter → webhook {query}n8n is an orchestrator, not a retriever.Workflow is seeded before readiness. 4–5 · CLASSIFY + ROUTEclassifier model returns simple or complexCode node maps the label to a model alias.Selection happens independently per request. 6 · EXECUTE DOWNSTREAMsimple → vanilla-ragcomplex → agentic-ragTheir own retrieval and model calls remain intact. 7–8 · SHAPE + RETURNWorkflow preserves downstream evidence fields.Adapter adds routing and timing metadata.No knowledge is persisted by n8n.