OpenAI file search answers with cited file names inline

OpenAI file search answers with cited file names inline

An n8n workflow that answers a chat question from the files in your own OpenAI vector store, then writes the name of each
file it used into the answer. The answer comes from OpenAI's Responses API and its file search tool.

What is inside

• workflow.json: the workflow as n8n exports it, eight nodes: four that do the work and four sticky notes. It holds no
  credential.
• README.md: these notes.

What we changed from the original copy

• The original ran on OpenAI's Assistants API. It called an assistant, read that assistant's thread with one request and
  looked up each cited file's name with another. OpenAI's deprecations page, read on 12 November 2026, says the Assistants
  API shut down on 26 August 2026 and names the Responses API and the Conversations API as its replacement. Those nodes
  cannot run now.
• Mindket's copy asks the Responses API directly. Chat Trigger feeds Ask OpenAI with File Search, an HTTP Request node
  that posts to https://api.openai.com/v1/responses with the file search tool and your vector store. Insert File Names
  then writes the names into the answer.
• The two lookups are gone, with the three Split Out nodes, the Set node and the Aggregate node that served them. OpenAI's
  file search guide shows each file citation carrying the file's name, so no second request is
  needed.
• The assistant's ID is gone. The vector store's ID is <YOUR_VECTOR_STORE_ID> in the request body.
• The original assistant took its model from OpenAI's dashboard, which the file does not show. The request now names
  gpt-5.6-terra, which OpenAI's model page lists as supporting file search in the Responses API.
• The old Code node swapped odd marker text in the answer for the file name. A Responses API citation says where in the text
  it belongs, so the new Code node inserts the name at that spot instead.
• Window Buffer Memory is gone, because it fed the assistant node. Each question is answered on its own.
• The notes are rewritten as short notes about each step.
• Nothing else.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• An OpenAI vector store that holds your files, and its ID. OpenAI's file search guide explains how to make one.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in Ask OpenAI with File Search.
3. In the JSON body of that node, replace <YOUR_VECTOR_STORE_ID> with your vector store's ID. Remove the angle brackets too.
4. Open the chat from Chat Trigger and ask a question that your files can answer.

How a run goes

1. Chat Trigger takes the question as chatInput.
2. Ask OpenAI with File Search posts the question with the file_search tool and your vector store. OpenAI searches the
   files and writes the answer.
3. Insert File Names reads the answer's text and its file citations. At each citation's position it adds the file's name
   in Markdown italics, and two files at one spot are joined by a comma.
4. Optional Markdown to HTML is switched off. Turn it on to change the Markdown into HTML.

Change the look

• The model: model in the JSON body of Ask OpenAI with File Search.
• How many passages OpenAI may use: add max_num_results to the tool in the body. OpenAI's guide lists it.
• The look of the note: the text in Insert File Names.
• HTML instead of Markdown: switch on Optional Markdown to HTML.

Good to know

• Each question starts fresh, because the request carries no earlier messages.
• OpenAI's API reference, read on 12 November 2026, defines a citation's index as the position in the output text where the
  citation belongs. The Code node trusts it.
• The Code node expects a message in the reply. An error body from OpenAI has none, so the run stops there.
• The question and the matching passages of your files go to OpenAI.
• The JSON holds no credential.