Build document RAG system with Kimi-K2, Gemini embeddings and Qdrant
UnknownWhat this workflow does
This AI workflow has 25 nodes and runs when a chat message arrives (plus 3 other triggers). Along the way it connects Google Gemini Chat Model, Featherless (community) and Qdrant (community). It also calls generativelanguage.googleapis.com and highwaycodeuk.co.uk over HTTP.
How it runs
Trigger: When a chat message arrives (plus 3 other triggers).
- Highway Code Expert (AI Agent): runs an AI agent, with model: Google Gemini Chat Model; 1 tool (Highway Code Manual)
- Retrieval Vectors with Gemini-Embeddings- (HTTP Request): calls generativelanguage.googleapis.com (POST)
- Download Large Document (HTTP Request): calls highwaycodeuk.co.uk
- Query Docs from Qdrant Vector Store (Qdrant): query points (search)
- Extract from File: extracts data from a file (pdf)
- Split Pages (Split Out): splits a list into separate items
- Kimi-K2 via Featherless.ai (Featherless): uses Featherless
- Get Response (Edit Fields (Set)): sets 4 fields: summary, title, chunk…
- Loop Over Items (Loop Over Items (Split in Batches)): loops over the items in batches
- Retrieval Vectors with Gemini Embeddings (HTTP Request): calls generativelanguage.googleapis.com (POST)
…plus 7 more steps. See the full node list below.
Services it uses
HTTP calls to: generativelanguage.googleapis.com, highwaycodeuk.co.uk
Before you upgrade to n8n 3.0
We can't fully verify this one: it relies on 2 community packages, and n8n 3.0 turns unverified community packages off by default (N8N_UNVERIFIED_PACKAGES_ENABLED=false).
- Check that n8n-nodes-featherless is verified or supports 3.0 (used in Kimi-K2 via Featherless.ai, Kimi-K2 via Featherless ai).
- Check that n8n-nodes-qdrant is verified or supports 3.0 (used in Add Docs To Qdrant Vector Store, Create Collection, Create Summary Index).
Nodes used (16)
- Featherless
n8n-nodes-featherless.featherless - Manual Trigger
n8n-nodes-base.manualTrigger - Extract from File
n8n-nodes-base.extractFromFile - Edit Fields (Set)
n8n-nodes-base.set - HTTP Request
n8n-nodes-base.httpRequest - Split Out
n8n-nodes-base.splitOut - Qdrant
n8n-nodes-qdrant.qdrant - Loop Over Items (Split in Batches)
n8n-nodes-base.splitInBatches - No Operation, do nothing
n8n-nodes-base.noOp - Chat Trigger
@n8n/n8n-nodes-langchain.chatTrigger - Execute Workflow Trigger
n8n-nodes-base.executeWorkflowTrigger - Google Gemini Chat Model
@n8n/n8n-nodes-langchain.lmChatGoogleGemini - Call n8n Workflow Tool
@n8n/n8n-nodes-langchain.toolWorkflow - MCP Server Trigger
@n8n/n8n-nodes-langchain.mcpTrigger - AI Agent
@n8n/n8n-nodes-langchain.agent - Execute Sub-workflow
n8n-nodes-base.executeWorkflow
From the author
Generating contextual summaries is an token-intensive approach for RAG embeddings which can quickly rack up costs if your inference provider charges by token usage. Featherless.ai is an inference provider with a different pricing model - they charge a flat…