Build an OpenAI RAG system with document upload, semantic search and caching
Build a sophisticated RAG system by uploading documents, performing semantic searches, and caching responses for efficiency. This workflow connects a Webhook for document upload and queries, n8n's AI nodes for text extraction, splitting, embedding with OpenAI, and storing in PGVector, and Postgres for caching.
Workflow JSON
{"meta":{"instanceId":"48aac30adfc5487a33ef16e0e958096f27cef40df3ed0febcbe1ca199e789786"},"nodes":[{"id":"772b7ee1-fa79-424e-91ee-b042a7c098c7","name":"Webhook Trigger","type":"n8n-nodes-base.webhook","position":[-1392,128],"webhookId":"3e81bb1b-3b9b-4cf1-ba3a-d0c4b93543e3","parameters":{"path":"rag-system","options":{},"httpMethod":"POST","responseMode":"lastNode"},"typeVersion":2.1},{"id":"f254f90f-fc8b-4e69-b680-2faa6ea14a2b","name":"Workflow Configuration","type":"n8n-nodes-base.set","position":[-1120,128],"parameters":{"options":{},"assignments":{"assignments":[{"id":"id-1","name":"chunkSize","type":"number","value":1000},{"id":"id-2","name":"chunkOverlap","type":"number","value":200},{"id":"id-3","name":"topK","type":"number","value":5},{"id":"id-4","name":"tableName","type":"string","value":"documents"},{"id":"id-5","name":"cacheTableName","type":"string","value":"query_cache"}]},"includeOtherFields":true},"typeVersion":3.4},{"id":"de7dfd1a-89f5-4579-8723-a6025a788f51","name":"Route by Action","type":"n8n-nodes-base.switch","position":[-736,128],"parameters":{"rules":{"values":[{"outputKey":"Upload","conditions":{"options":{"leftValue":"","caseSensitive":true,"typeValidation":"strict"},"combinator":"and","conditions":[{"operator":{"type":"string","operation":"equals"},"leftValue":"={{ $json.body.action }}","rightValue":"upload"}]},"renameOutput":true},{"outputKey":"Query","conditions":{"options":{"leftValue":"","caseSensitive":true,"typeValidation":"strict"},"combinator":"and","conditions":[{"operator":{"type":"string","operation":"equals"},"leftValue":"={{ $json.body.action }}","rightValue":"query"}]},"renameOutput":true}]},"options":{"fallbackOutput":"extra"}},"typeVersion":3.4},{"id":"04c7dba5-11fe-4320-a72c-0dde1ec79bbd","name":"Extract Text from Document","type":"n8n-nodes-base.extractFromFile","position":[-448,752],"parameters":{"options":{},"operation":"pdf"},"typeVersion":1.1},{"id":"3cc951b4-1db9-44f5-bc0e-94dba28aae25","name":"Text Splitter","type":"@n8n/n8n-nodes-langchain.textSplitterRecursiveCharacterTextSplitter","position":[-96,1184],"parameters":{"options":{},"chunkSize":"={{ $('Workflow Configuration').first().json.chunkSize }}","chunkOverlap":"={{ $('Workflow Configuration').first().json.chunkOverlap }}"},"typeVersion":1},{"id":"8b938e6a-5a85-4271-aa1f-260802704c93","name":"Document Loader","type":"@n8n/n8n-nodes-langchain.documentDefaultDataLoader","position":[-160,1024],"parameters":{"options":{},"jsonData":"={{ $json.text }}","jsonMode":"expressionData","textSplittingMode":"custom"},"typeVersion":1.1},{"id":"2c678aa9-751c-47fd-a5f5-5c85e40a5b9e","name":"OpenAI Embeddings","type":"@n8n/n8n-nodes-langchain.embeddingsOpenAi","position":[192,1056],"parameters":{"options":{}},"typeVersion":1.2},{"id":"c0432b95-ccf1-4cdd-995e-7cb3af38437d","name":"Store Embeddings in PGVector","type":"@n8n/n8n-nodes-langchain.vectorStorePGVector","position":[-112,752],"parameters":{"mode":"insert","options":{},"tableName":"={{ $('Workflow Configuration').first().json.tableName }}"},"typeVersion":1.3},{"id":"92c79b58-e8c9-4496-be91-fe7d0983c16c","name":"Log Upload to Cache","type":"n8n-nodes-base.postgres","position":[528,752],"parameters":{"query":"INSERT INTO upload_log (user_id, document_name, uploaded_at) VALUES ($1, $2, $3)","options":{"queryReplacement":"={{ $('Webhook Trigger').first().json.body.user_id }},={{ $('Webhook Trigger').first().json.body.document_name }},={{ $now.toISO() }}"},"operation":"executeQuery"},"typeVersion":2.6},{"id":"c5d4dcaa-0f64-44ee-93b9-6d6994a62397","name":"Respond Upload Success","type":"n8n-nodes-base.respondToWebhook","position":[848,752],"parameters":{"options":{},"respondWith":"json","responseBody":"={\n \"status\": \"success\",\n \"message\": \"Document uploaded and processed\",\n \"user_id\": \"={{ $('Webhook Trigger').first().json.body.user_id }}\",\n \"document_name\": \"={{ $('Webhook Trigger').first().json.body.document_name }}\"\n}"},"typeVersion":1.5},{"id":"d7e575a7-e112-4848-8521-c24b4adc50a6","name":"Check Query Cache","type":"n8n-nodes-base.postgres","position":[-384,-64],"parameters":{"query":"SELECT answer, created_at FROM query_cache WHERE user_id = $1 AND query_hash = MD5($2) AND created_at > NOW() - INTERVAL '1 hour'","options":{"queryReplacement":"={{ $('Webhook Trigger').first().json.body.user_id }},={{ $('Webhook Trigger').first().json.body.query }}"},"operation":"executeQuery"},"typeVersion":2.6},{"id":"a6cd4561-8ba5-43a6-a679-d6259f64078c","name":"Cache Hit or Miss","type":"n8n-nodes-base.switch","position":[-112,-64],"parameters":{"rules":{"values":[{"outputKey":"Cache Hit","conditions":{"options":{"leftValue":"","caseSensitive":true,"typeValidation":"strict"},"combinator":"and","conditions":[{"operator":{"type":"number","operation":"gt"},"leftValue":"={{ $json.length }}","rightValue":0}]},"renameOutput":true},{"outputKey":"Cache Miss","conditions":{"options":{"leftValue":"","caseSensitive":true,"typeValidation":"strict"},"combinator":"and","conditions":[{"operator":{"type":"number","operation":"equals"},"leftValue":"={{ $json.length }}","rightValue":0}]},"renameOutput":true}]},"options":{}},"typeVersion":3.4},{"id":"1576da8b-432c-42c9-a0e4-487523bf314b","name":"Retrieve Relevant Chunks","type":"@n8n/n8n-nodes-langchain.vectorStorePGVector","position":[432,1088],"parameters":{"options":{"metadata":{"metadataValues":[{"name":"user_id","value":"={{ $('Webhook Trigger').first().json.body.user_id }}"}]}},"tableName":"={{ $('Workflow Configuration').first().json.tableName }}"},"typeVersion":1.3},{"id":"559a79ea-ce0a-4fa1-96e6-f9439799a48a","name":"OpenAI Chat Model","type":"@n8n/n8n-nodes-langchain.lmChatOpenAi","position":[224,240],"parameters":{"model":{"__rl":true,"mode":"list","value":"gpt-4.1-mini"},"options":{},"builtInTools":{}},"typeVersion":1.3},{"id":"4952c071-c20b-4e4e-b753-e0fe93413d26","name":"Answer Query with Context","type":"@n8n/n8n-nodes-langchain.agent","position":[288,16],"parameters":{"text":"={{ $('Webhook Trigger').first().json.body.query }}","options":{"systemMessage":"You are a helpful AI assistant that answers questions based on the provided document context.\n\nYour task is to:\n1. Analyze the user's question carefully\n2. Review the relevant document chunks provided from the vector database\n3. Provide a clear, accurate answer based ONLY on the information in the documents\n4. If the answer is not in the provided context, say \"I don't have enough information to answer that question based on the available documents.\"\n5. Always cite which document or section your answer comes from when possible\n\nBe concise but thorough. Maintain a professional and helpful tone."},"promptType":"define"},"typeVersion":3},{"id":"a7bea884-a935-4de3-8770-a52806d1abd4","name":"Save to Query Cache","type":"n8n-nodes-base.postgres","position":[768,16],"parameters":{"query":"INSERT INTO query_cache (user_id, query_hash, query, answer, created_at) VALUES ($1, MD5($2), $3, $4, NOW()) ON CONFLICT (user_id, query_hash) DO UPDATE SET answer = EXCLUDED.answer, created_at = NOW()","options":{"queryReplacement":"={{ $('Webhook Trigger').first().json.body.user_id }},={{ $('Webhook Trigger').first().json.body.query }},={{ $('Webhook Trigger').first().json.body.query }},={{ $json.output }}"},"operation":"executeQuery"},"typeVersion":2.6},{"id":"c380d8d7-769c-4bfe-994e-d54dffcc4c7a","name":"Respond with Answer","type":"n8n-nodes-base.respondToWebhook","position":[1168,16],"parameters":{"options":{},"respondWith":"json","responseBody":"={\n \"answer\": \"={{ $('Answer Query with Context').first().json.output }}\",\n \"cached\": false,\n \"user_id\": \"={{ $('Webhook Trigger').first().json.body.user_id }}\"\n}"},"typeVersion":1.5},{"id":"efcc43e4-7a1d-4174-85dd-f73c55960ba6","name":"Format Cached Response","type":"n8n-nodes-base.set","position":[416,-576],"parameters":{"options":{},"assignments":{"assignments":[{"id":"id-1","name":"answer","type":"string","value":"={{ $json.answer }}"},{"id":"id-2","name":"cached","type":"boolean","value":true},{"id":"id-3","name":"user_id","type":"string","value":"={{ $('Webhook Trigger').first().json.body.user_id }}"}]}},"typeVersion":3.4},{"id":"767873ee-0740-4460-8d1a-79a9b37e1d8b","name":"Respond with Cached Answer","type":"n8n-nodes-base.respondToWebhook","position":[880,-576],"parameters":{"options":{},"respondWith":"json","responseBody":"={\n \"answer\": \"={{ $json.answer }}\",\n \"cached\": true,\n \"user_id\": \"={{ $json.user_id }}\"\n}"},"typeVersion":1.5},{"id":"4f8320ed-a474-4f49-b6d6-baf644f3c937","name":"Sticky Note1","type":"n8n-nodes-base.stickyNote","position":[-1472,48],"parameters":{"color":7,"width":592,"height":240,"content":"## Input Layer\nReceive upload or query via webhook and Set chunking, topK, and database tables"},"typeVersion":1},{"id":"0aafeb7b-06f6-45c9-bfe8-322e95550240","name":"Sticky Note","type":"n8n-nodes-base.stickyNote","position":[-800,16],"parameters":{"color":7,"width":304,"height":288,"content":"## Action Routing\nRoute request to upload or query flow"},"typeVersion":1},{"id":"34da2fb8-619f-4e21-b2d1-4b7e0192f2ae","name":"Answer questions with a vector store","type":"@n8n/n8n-nodes-langchain.toolVectorStore","position":[368,240],"parameters":{},"typeVersion":1.1},{"id":"479de263-883a-4197-9ae2-09ed26b05b5c","name":"Sticky Note2","type":"n8n-nodes-base.stickyNote","position":[-448,-192],"parameters":{"color":7,"width":470,"height":288,"content":"## Cache Check\nCheck if query result exists in cache"},"typeVersion":1},{"id":"afface0a-2435-4831-aee3-0c5683218a53","name":"Sticky Note4","type":"n8n-nodes-base.stickyNote","position":[336,-704],"parameters":{"color":7,"width":272,"height":320,"content":"## Cache Response Formatting\nFormat cached query results into a consistent response structure."},"typeVersion":1},{"id":"2fe09f60-1e2b-42fd-9115-34af03fcfee3","name":"Sticky Note5","type":"n8n-nodes-base.stickyNote","position":[768,-736],"parameters":{"color":7,"width":304,"height":320,"content":"## Response Layer\nReturn answer or cached result via webhook"},"typeVersion":1},{"id":"0adeaf21-cd46-4e59-96a7-2285c91d79a9","name":"Sticky Note6","type":"n8n-nodes-base.stickyNote","position":[1056,-64],"parameters":{"color":7,"width":304,"height":240,"content":"## Response Layer\nReturn answer or cached result via webhook"},"typeVersion":1},{"id":"9efa23a4-2f27-4b7b-b729-c9a0c02194ce","name":"Sticky Note7","type":"n8n-nodes-base.stickyNote","position":[656,-64],"parameters":{"color":7,"width":304,"height":272,"content":"## Cache Storage\nStore query and response for reuse"},"typeVersion":1},{"id":"75a4c942-f12b-42ae-96a1-6a36542acf80","name":"Sticky Note8","type":"n8n-nodes-base.stickyNote","position":[176,-80],"parameters":{"color":7,"width":464,"height":448,"content":"## Answer Generation\nGenerate answer using context + AI model"},"typeVersion":1},{"id":"02e378e4-dd93-4800-ae87-1d14490b2727","name":"Sticky Note9","type":"n8n-nodes-base.stickyNote","position":[-512,672],"parameters":{"color":7,"width":304,"height":240,"content":"## Document Processing\nExtract text from uploaded documents"},"typeVersion":1},{"id":"f21786fe-b403-4aa0-b484-caa8bcd67e64","name":"Sticky Note10","type":"n8n-nodes-base.stickyNote","position":[-160,640],"parameters":{"color":7,"width":304,"height":272,"content":"## Chunking Engine\nSplit text into overlapping chunks"},"typeVersion":1},{"id":"6a5abc30-754c-4e9b-8213-f79186b80a35","name":"Sticky Note11","type":"n8n-nodes-base.stickyNote","position":[400,976],"parameters":{"color":7,"width":304,"height":240,"content":"## Embeddings Storage\nGenerate embeddings and store in PGVector"},"typeVersion":1},{"id":"7a1009b0-7395-4041-bde5-06d382318b30","name":"Sticky Note12","type":"n8n-nodes-base.stickyNote","position":[368,624],"parameters":{"color":7,"width":672,"height":272,"content":"## Upload Logging\nTrack document uploads in database"},"typeVersion":1},{"id":"32a6aac8-009a-4cac-829b-ca853c9f2f38","name":"Sticky Note13","type":"n8n-nodes-base.stickyNote","position":[-2224,-256],"parameters":{"width":448,"height":576,"content":"## How it works\nThis workflow implements a complete Retrieval-Augmented Generation (RAG) system for document ingestion and intelligent querying.\n\nUsers can upload documents or send queries via webhook. Uploaded files are processed by extracting text, splitting it into chunks, generating embeddings, and storing them in a vector database (PGVector).\n\nWhen a query is received, the workflow first checks a cache for recent answers. If no cache is found, it retrieves relevant document chunks using vector search and generates a contextual answer using an AI model.\n\nResponses are cached for faster future queries and returned to the user via webhook.\n\n## Setup steps\n- Configure webhook endpoint for upload and query actions\n- Add OpenAI API credentials for embeddings and chat\n- Set up Postgres with PGVector extension\n- Create tables for documents and query cache\n- Adjust chunk size, overlap, and topK settings"},"typeVersion":1}],"pinData":{},"connections":{"Text Splitter":{"ai_textSplitter":[[{"node":"Document Loader","type":"ai_textSplitter","index":0}]]},"Document Loader":{"ai_document":[[{"node":"Store Embeddings in PGVector","type":"ai_document","index":0}]]},"Route by Action":{"main":[[{"node":"Extract Text from Document","type":"main","index":0}],[{"node":"Check Query Cache","type":"main","index":0}]]},"Webhook Trigger":{"main":[[{"node":"Workflow Configuration","type":"main","index":0}]]},"Cache Hit or Miss":{"main":[[{"node":"Format Cached Response","type":"main","index":0}],[{"node":"Answer Query with Context","type":"main","index":0}]]},"Check Query Cache":{"main":[[{"node":"Cache Hit or Miss","type":"main","index":0}]]},"OpenAI Chat Model":{"ai_languageModel":[[{"node":"Answer Query with Context","type":"ai_languageModel","index":0}]]},"OpenAI Embeddings":{"ai_embedding":[[{"node":"Store Embeddings in PGVector","type":"ai_embedding","index":0},{"node":"Retrieve Relevant Chunks","type":"ai_embedding","index":0}]]},"Log Upload to Cache":{"main":[[{"node":"Respond Upload Success","type":"main","index":0}]]},"Save to Query Cache":{"main":[[{"node":"Respond with Answer","type":"main","index":0}]]},"Format Cached Response":{"main":[[{"node":"Respond with Cached Answer","type":"main","index":0}]]},"Workflow Configuration":{"main":[[{"node":"Route by Action","type":"main","index":0}]]},"Retrieve Relevant Chunks":{"ai_vectorStore":[[{"node":"Answer questions with a vector store","type":"ai_vectorStore","index":0}]]},"Answer Query with Context":{"main":[[{"node":"Save to Query Cache","type":"main","index":0}]]},"Extract Text from Document":{"main":[[{"node":"Store Embeddings in PGVector","type":"main","index":0}]]},"Store Embeddings in PGVector":{"main":[[{"node":"Log Upload to Cache","type":"main","index":0}]]},"Answer questions with a vector store":{"ai_tool":[[{"node":"Answer Query with Context","type":"ai_tool","index":0}]]}}}How to Import This Workflow
- 1Copy the workflow JSON above using the Copy Workflow JSON button.
- 2Open your n8n instance and go to Workflows.
- 3Click Import from JSON and paste the copied workflow.
Don't have an n8n instance? Start your free trial at n8nautomation.cloud
Related Templates
Auto-create TikTok videos with VEED.io AI avatars, ElevenLabs & GPT-4
Automate the creation and distribution of trending TikTok videos using AI avatars. This workflow connects Telegram, Perplexity, OpenAI, ElevenLabs, VEED.io, and BLOTATO to generate scripts, synthesize voice, create video, and publish across multiple social platforms. Content creators and marketers can rapidly produce engaging short-form video content without manual editing.
Summarize Google Sheets form feedback via OpenAI's GPT-4
Efficiently summarize Google Sheets form feedback using OpenAI's GPT-4. This productivity workflow connects Google Sheets to retrieve form submissions, aggregates the responses into arrays, and then leverages OpenAI's advanced GPT-4 model to generate concise summaries. The generated summaries, initially in Markdown, are then converted to HTML for easy readability before being sent directly to your inbox via Gmail. This automation is ideal for product managers, marketing teams, or customer support departments who regularly collect feedback through Google Forms and need a quick, intelligent overview of responses without manually sifting through every entry. It dramatically reduces the time and effort spent analyzing feedback, allowing for faster insights and more informed decision-making.
Use skills In n8n agent node
This template gives you a framework to use skills in any n8n agent. You can use this as a starting point and add any other tools or patterns needed for your use case. What are “skills”? Skills are a context management standard created by Anthropic for use in Claude code. Basically, instead of having a HUGE system prompt, skills split that into lots of small, structured instruction files that tell an agent how to do a specific kind of task. Instead of stuffing a massive prompt full of rules, the agent: finds the relevant skill file reads it and follows the steps inside It’s a simple pattern that makes managing system prompts for general purpose agents much more straightforward. See an example of a skills repo here. What this workflow does Responds to messages in n8n Chat (or Chat Hub) Builds an “available skills” index from one or more GitHub repos Lets the agent browse folders + fetch skill files (Markdown) as needed Uses the skill content to guide how it completes tasks How it (roughly) works A chat message comes in. The workflow lists directories in the configured skills repos (root + if present), filters out noise, and merges everything into one directory map. That directory map gets injected into the agent’s system prompt so it knows what skill files exist. When it needs instructions, it uses tools: List Files by Path Name to explore folders * Get a File From GitHub to pull the skill file as raw text (no base64) Same “skills” pattern as the CLI tools The flow is: find a skill → open it → follow the steps, the same way it works in the CLI tools, but running inside n8n, so you don’t need to download or install anything locally. How to set it up (Required) Add your GitHub credentials to each node that needs it (Required) Add your OpenRouter credentials to the chat node or replace with the provider of your choosing (Optional) Add more repos to (any skills GitHub repo works as long as your credentials have access to it, such as any public repo) (Optional) Add more tools and turn it into whatever agent you actually need