n8n-nodes-tool-vector-store-custom-prompt

A custom n8n community node that extends the Vector Store Question Answer Tool with editable prompt templates. This node allows you to customize the QA prompt used in RAG (Retrieval-Augmented Generation) workflows, providing more control over how AI agents answer questions using vector stores.

Features

• ✅ Customizable Prompt Template - Edit the QA prompt via UI (unlike the original node)
• ✅ Two Prompt Modes - Support for both "stuff" (PromptTemplate) and "chat" (ChatPromptTemplate) modes
• ✅ Strict Mode - Prevents AI from making inferences and requires explicit evidence from context
• ✅ Source Return - Optional return of document sources with metadata
• ✅ Full LangChain Integration - Uses VectorDBQAChain with StuffDocumentsChain
• ✅ AI Tool Compatible - Works seamlessly with n8n AI Agents

Installation

Install this node in your n8n instance:

npm install n8n-nodes-tool-vector-store-custom-prompt

Usage

Basic Setup

1. Add the Vector Store QA Tool (Custom Prompt) node to your workflow
2. Connect a Vector Store node to the first input
3. Connect a Language Model (Chat Model) to the second input
4. Configure the node parameters:
   • Description: Describe what data is in your vector store (e.g., "imóveis, bairros e empreendimentos")
   • Limit (Top-K): Number of documents to retrieve (default: 4)
   • Prompt Template: Your custom QA prompt (must include {context} and {question})
   • Strict Mode: Enable to prevent AI inferences (recommended)

Connecting to AI Agent

1. Add an AI Agent node to your workflow
2. Enable the tool from the Vector Store QA Tool node
3. The tool will be available with a sanitized name based on your node name

Example Prompt Template

Você é um assistente de QA (perguntas e respostas) que responde com base APENAS no contexto fornecido.

REGRAS CRÍTICAS:
1) Use SOMENTE as informações presentes no CONTEXTO.
2) Se a resposta não estiver explicitamente no CONTEXTO, responda exatamente: "Não encontrei essa informação no contexto disponível."
3) NÃO inferir bairros/regiões próximas, "na região", "perto de", "equivalentes" ou similares, a menos que o CONTEXTO cite isso explicitamente.
4) NÃO inventar códigos/IDs de imóveis e NÃO supor características que não estejam no CONTEXTO.
5) Se a pergunta exigir filtros (ex.: bairro X, 3 quartos, 2 vagas), só traga itens que tenham esses critérios explícitos no CONTEXTO.

CONTEXTO:
{context}

PERGUNTA:
{question}

RESPONDA de forma direta, curta e objetiva.

Node Parameters

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | Description | string | "imóveis, bairros e empreendimentos" | Description of the vector store content (used in tool description) | | Limit (Top-K) | number | 4 | Maximum number of documents to retrieve in similarity search | | Prompt Mode | options | "stuff" | Prompt mode: "stuff" (PromptTemplate) or "chat" (ChatPromptTemplate) | | Document Variable Name | string | "context" | Variable name in template that receives the context (must match {context} in prompt) | | Strict Mode | boolean | true | When enabled, injects reinforcements to prevent inference and require explicit evidence | | Return Sources | boolean | false | When enabled, returns a SOURCES section with document metadata | | Prompt Template | string | (see default above) | Editable prompt used to combine documents from vector store |

Differences from Original Node

| Aspect | Original toolVectorStore | This Node | |--------|---------------------------|-----------| | Prompt | Fixed LangChain default (not editable) | Editable via UI | | Customization | Limited | Full control over instructions, tone, response format | | Use Cases | Generic | Specific business needs (e.g., real estate, internal docs) |

Requirements

• n8n: Version 1.0.0 or higher
• langchain: ^0.3.0 (peer dependency)
• @langchain/core: ^0.3.0 (peer dependency)

Development

Prerequisites

• Node.js v22 or higher
• npm

Setup

# Install dependencies
npm install

# Build the project
npm run build

# Run in development mode
npm run dev

# Lint code
npm run lint

Troubleshooting

Node doesn't appear in n8n

1. Make sure you ran npm install
2. Check that the node is listed in package.json under n8n.nodes
3. Restart n8n
4. Check console for error messages

"Cannot find module 'langchain'"

This usually means langchain dependencies are not installed. Ensure you have:

• langchain ^0.3.0
• @langchain/core ^0.3.0

These should be available in your n8n environment.

Invalid prompt template

The node validates that your prompt template includes:

• {context} (or the name configured in "Document Variable Name")
• {question} (required in "stuff" mode)

Make sure both placeholders are present in your template.

License

MIT

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Support

For issues and questions:

• n8n Community Forum
• GitHub Issues

Related

• n8n Documentation
• LangChain Documentation
• n8n Community Nodes