n8n Agents

Purpose: This repository houses multiple n8n agents. Currently included:

• PDF From Drive to Pinecone Vector Store (RAG data loader and QA)
• ElevenLabs Voice AI From Database (voice responses powered by RAG)

Repository scope

• This repository will host multiple n8n agents over time (e.g., data loaders, RAG utilities, voice assistants).
• Each agent will live in its own folder with an importable .json workflow and optional diagram/docs.

Agents in this repo

• PDF From Drive to Pinecone Vector Store: loads PDFs from Google Drive, chunks text, embeds, and upserts into Pinecone; enables QA over your documents.

• ElevenLabs Voice AI From Database: exposes a webhook that answers questions using a Pinecone vector store and returns voice audio via ElevenLabs.

How to import any agent

1. In n8n: Workflows → Import from File → select the agent .json under its folder.
2. Create/assign credentials (OpenAI, Pinecone, Google Drive, ElevenLabs as applicable).
3. Update node parameters (index name/namespace, model IDs, folder IDs, etc.).
4. Execute the workflow or trigger the webhook to test.

Prerequisites

• n8n (cloud or self-hosted) >= 1.0
• Google Drive access to the target folder/files
• Pinecone account with an index created (metric: cosine or dot-product; dimension should match your chosen embedding model)

Quick Start

1. Import the workflow into n8n

   • In n8n UI: Workflows → Import from File → select PDF From Drive to Pinecone Vector Store/PDF-From-Drive-To-Pinecone.json.

2. Create/Configure credentials

   • Google Drive: OAuth2 or Service Account with read access to target files/folders.
   • Pinecone: API Key and Environment/Base URL for the target index.

3. Set required variables/nodes in the workflow

   • Google Drive search/filter: set folder ID or query so only desired PDFs are processed.
   • Text splitting/chunking: adjust chunk size and overlap for your use case.
   • Embeddings: select the embedding model used (ensure Pinecone index dimension matches).
   • Pinecone: set Index name and optional Namespace; map metadata (e.g., file name, path, page numbers).

4. Test run

   • Execute the workflow on a small set of PDFs to verify chunk counts and vector counts.
   • Confirm upserts appear in your Pinecone index dashboard.

Tips

• Dimension must match the embedding model. For example, OpenAI text-embedding-3-large is 3072; adjust your Pinecone index accordingly.
• Namespacing keeps sources isolated per project or environment.
• Start with larger chunks for semantic search; reduce size if answers miss detail.

Troubleshooting

• No files found: verify Google Drive folder permissions and the search query/filters.
• Dimension mismatch errors: recreate Pinecone index with the correct dimension for your embedding model.
• Timeouts: reduce batch size or number of parallel upserts; check n8n execution timeouts.

Repository Structure

PDF From Drive to Pinecone Vector Store/
  ├─ PDF-From-Drive-To-Pinecone.json
  └─ pinecone_workflow.png

Eleven Labs Voice AI From Pinecone Database/
  ├─ Eleven-Labs-Voice-AI-From-Database.json
  └─ voice_ai_workflow.png

License

No license specified. If you plan to share or modify, add a LICENSE file.