Automation library · Money admin

Build a tax code assistant with Qdrant and Mistral

Started by someone opening a chat with it, and works across Qdrant. It puts OpenAI behind a retrieval step, so answers are drawn from documents you supply rather than from the model alone. 31 steps, 1 branch, 2 loops or waits: large, and worth reading before you trust it with anything live.

LargeCalls a model: OpenAIRetrieval

Download free
31 stepsLarge: read it before trusting it with anything live
1Branch: places the flow splits
2Loops or waits
1Service it connects to

The flow

Starts it 3Model 10Logic 12Service 6

The steps in the order they run, from what starts it.

  1. 1When clicking ‘Test workflow’Manual triggerStarts it
  2. 2Embeddings Mistral CloudEmbeddings Mistral CloudModel
  3. 3Recursive Character Text SplitterText Splitter Recursive Character Text SplitterModel
  4. 4Execute Workflow TriggerExecute Workflow triggerStarts it
  5. 5Window Buffer MemoryMemory Buffer WindowModel
  6. 6Window Buffer Memory1Memory Buffer WindowModel
  7. 7OpenAI Chat ModelLm Chat Open AiModel
  8. 8Ask ToolTool WorkflowModel
  9. 9Search ToolTool WorkflowModel
  10. 10Get Tax Code Zip FileHttp RequestService
  11. 11Qdrant Vector StoreVector Store QdrantModel
  12. 12Default Data LoaderDocument Default Data LoaderModel
  13. 13SwitchSwitchLogic
  14. 14AI AgentAgentModel
  15. 15When chat message receivedChat triggerStarts it
  16. 16Extract Zip FilesCompressionService
  17. 171secWaitLogic
  18. 18Get Mistral EmbeddingsHttp RequestService
  19. 19Use Qdrant Scroll APIHttp RequestService
  20. 20Files as ItemsSplit OutLogic
  21. 21For Each Section...Split In BatchesLogic
  22. 22Use Qdrant Search API1Http RequestService
  23. 23Get Search ResponseSetLogic
  24. 24Extract PDF ContentsExtract From FileService
  25. 25Content Chunking @ 50k CharsSetLogic
  26. 26Get Ask ResponseSetLogic
  27. 27Extract From ChapterSetLogic
  28. 28Split Out ChunksSplit OutLogic
  29. 29Map To SectionsSetLogic
  30. 30Sections To ListSplit OutLogic
  31. 31Only Valid SectionsFilterLogic

How it's built

Starts on
Someone opening a chat with it
Steps
31, with 1 branch, 2 loops or waits
Size
Large: read it before trusting it with anything live
Connects to
Qdrant
Model
OpenAI
Retrieval
Answers are drawn from documents you supply, not from the model alone

Questions

What does it need to run?

An n8n account, plus your own Qdrant account, and a key for OpenAI. You connect your own accounts after importing it; none are included.

What starts it?

It starts on someone opening a chat with it.

Does it use AI?

Yes. It calls OpenAI. Its answers draw on documents you give it, not on the model alone. Model calls are billed by the provider, not by n8n.

How long does it take to set up?

Take your time. At 31 steps it is large: read it before trusting it with anything live.

How do I import it?

Copy the workflow, open a new workflow in n8n and paste. The steps appear on the canvas, ready for your own accounts.

Before you run it

Nothing here has been tested by us. It is somebody's working automation, read and described from its own steps. Treat the connections it wants as a list of accounts it will touch, and read the 31 steps before you point it at anything that matters.