n8n Reranker Cohere nodeConfigure Reranker Cohere in n8n.
A vector store gives back the chunks that look close to the question, not always the ones that answer it. The n8n Reranker Cohere node reorders that list by relevance before it reaches the model. Two parameters, one port, and a Cohere API key to create.
Verified Trustpilot reviews · AI, automation & growth agency
What does the n8n Reranker Cohere node change in a RAG workflow?
The Reranker Cohere node reorders the chunks a vector store returns, from most to least relevant to the query. It is a sub-node: it has no main input, it never runs on its own, and it does nothing until you attach it to a node that exposes a reranker port. A vector store, or vector database, holds mathematical representations of your documents so an AI can search them.
Vector search scores chunks on similarity. Similarity and usefulness are not the same thing, and that gap shows up in the answers: a paragraph that repeats the words of the question ranks above the paragraph that contains the actual figure. Reranking is the pass that fixes the order before generation.
A support bot that stops quoting the wrong page. The vector store returns candidate chunks, Cohere scores them against the question, and Top N cuts the list to what the model reads. Fewer chunks in the prompt, less noise around the one that matters, and the answer stops drifting onto a neighbouring product page.
A multilingual knowledge base queried in one language and written in another. Set Model to rerank-multilingual-v3.0 and the scoring no longer collapses when the question is asked in Spanish against English documentation. This is the case where the model choice is the whole point of the node.
A retrieval step feeding an AI Agent. The agent picks its own tools, so it will call retrieval more than once in a conversation. Reranking on every call keeps each round of context short, which matters because the model provider bills every call it receives.
Know what it will not do. It does not retrieve anything by itself, it does not rewrite the chunks, and it does not change how the documents were embedded: a bad chunking strategy stays a bad chunking strategy after reranking. It also adds a Cohere call on top of your existing model calls, from OpenAI, Anthropic or Google Gemini, and each provider bills its own API under its own terms. If your vector store already returns three chunks and all three are used, there is nothing to reorder.
The node has one version, and older workflows may still show an earlier one with fewer options. Teams that want the whole retrieval chain walked through on their own data usually go through the n8n training.
What do you need to authenticate it?
- 01
Open a Cohere account with the right role
Create a Cohere account first. Roles matter here: the Trial API needs User or Owner permissions, the Production API needs Owner permissions. If someone else owns the Cohere workspace and you only hold a User seat, you can build and test, but the switch to the Production API has to come from an Owner.
- 02
Generate the API key
The key lives in the API Keys section of the Cohere dashboard. Copy it there, it is the single value the credential asks for. A credential in n8n is a stored secret: you paste the key once and workflows reference the credential, never the raw key, so nothing sensitive ends up in a node parameter or in an exported workflow file.
- 03
Create the credential in n8n and select it
In n8n, go to the Credentials menu and create a Cohere credential with the API key. Then open the Reranker Cohere node and pick that credential in its selector. It is created once and reused across every workflow on the instance, on n8n Cloud and on a self-hosted setup alike.
Where does the node plug in?
n8n AI workflows run on a root node and sub-nodes: the root node receives the workflow items on its main input, and sub-nodes attach to its ports, one type per port. Reranker Cohere is a sub-node with a single output port and no main input.
Output (what it plugs into)
- Reranker
ai_reranker
01Reranker
This is the port the node connects to on a root node, and the only connection it has. Drag from the reranker port of the node doing the retrieval, and the reordering step sits inside that retrieval instead of being a separate branch of the workflow.
Key parameters
- Required: the node never executes alone, so an unattached Reranker Cohere sits idle in the canvas and no run will ever reach it.
Which parameters does the node expose?
The Reranker Cohere node has 2 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.
Model
modelNameWhat you see in n8n
Notes & use cases
Picks the Cohere model that scores the retrieved documents against the query. This is the parameter you set first, because the language coverage of your documents decides it.
Key parameters
- Model: the model used to rerank the documents, chosen from a dropdown.
- rerank-multilingual-v3.0: one of the three values in the dropdown, next to
rerank-v3.5andrerank-english-v3.0.
rerank-multilingual-v3.0, while an English-only product knowledge base has no reason to leave the English model.Top N
topNWhat you see in n8n
Notes & use cases
Caps how many documents come out of the reranking step. Everything the vector store returned is scored, then the tail of the ranking is dropped and only the top slice continues to the model.
Key parameters
- Top N: the maximum number of documents returned after reranking, given as a number.
Need help automating Reranker Cohere with n8n?
A person reads every message.
Questions about the Reranker Cohere node
01Is the n8n Reranker Cohere node free to use?
02What do you need for the Reranker Cohere node to work?
03What are the limits of the Reranker Cohere node?
04What should you plug in first when starting out?
05n8n or Make for this kind of RAG workflow?
Get our weekly integration tips.
No spam. Unsubscribe anytime.
