import type { ParsedWorkflow, ParsedNode } from './types.js';
/**
 * Find the sub-node connected INTO `agentName` via a reverse `ai_*` port. In n8n
 * an AI sub-node (model / memory / output parser) is the connection *source* and
 * the agent is the *target*, so we scan every source's named port for one whose
 * slots target the agent.
 */
export declare function findAgentSubNode(parsed: ParsedWorkflow, agentName: string, port: string): ParsedNode | undefined;
/**
 * Follow a chain of reverse `ai_*` ports hop by hop, returning the terminal
 * sub-node (or `undefined` if any hop is missing). A RAG retrieval chain nests
 * its sub-nodes — `chainRetrievalQa —ai_retriever→ retriever —ai_vectorStore→
 * store —ai_embedding→ embeddings` — so resolving the store namespace from the
 * chain node means walking `['ai_retriever', 'ai_vectorStore']`.
 */
export declare function walkAiChain(parsed: ParsedWorkflow, fromName: string, ports: string[]): ParsedNode | undefined;