n8n RAG Telegram Bot: Supabase + TogetherAI + OpenRouter

Turn your Google Docs into a RAG-ready Telegram chatbot—powered by Supabase + TogetherAI + OpenRouter in n8n

This n8n RAG Telegram Bot workflow fetches content from Google Docs, converts it into vector embeddings, stores them in Supabase, and then answers chat queries using retrieval from your embedded document knowledge.

What this workflow does

• Run once to build your RAG knowledge base: the first workflow is designed to be executed only one time. It retrieves your Google Docs content, splits it into chunks, creates embeddings for each chunk, and saves everything into the Supabase embed table.
• Chunk and embed the document: using a code node, the document text is sliced into 1000-character chunks for processing. For each chunk, an httpRequest node calls the TogetherAI embedding API to generate vector embeddings.
• Start chat-based retrieval: the second workflow triggers on a user chat message (chatTrigger) and begins by sending an initial greeting.
• Create embeddings for incoming queries: it generates embeddings for the user’s message via an httpRequest call to the TogetherAI embeddings endpoint, enabling embedding-based searching against the stored document vectors.

Use cases

• Answer questions inside a Telegram chat using knowledge from a Google Docs policy, SOP, or product documentation.
• Enable SaaS support workflows where your internal docs are indexed once, then reused for fast Q&A.
• Build a lightweight RAG system for teams that want document-grounded responses without manual indexing.

Technical details

• n8n nodes: telegramTrigger (currently disabled), googleDocs, code, httpRequest, supabase, aggregate
• Document ingestion: Google Docs via Service Account authentication
• Embeddings: TogetherAI embeddings API (document chunks + user query embeddings)
• Vector storage: Supabase embed table