n8n Embeddings Cohere nodeConfigure Embeddings Cohere in n8n.
The n8n Embeddings Cohere node turns text into vectors, the numeric form a vector store searches by meaning instead of by keyword. It is a sub-node, so it has no main input and never runs on its own. One parameter, 7 models, from 384 to 4096 dimensions.
Verified Trustpilot reviews · AI, automation & growth agency
What does the n8n Embeddings Cohere node do in a workflow?
It generates embeddings for a given text using Cohere. An embedding is a list of numbers that places a piece of text somewhere in a space of many dimensions, so two passages about the same topic land close together. The node hands that list to whatever root node it is plugged into, and that root node stores or searches the vectors. Nothing else. It has no main input, no output items of its own, and it never appears alone in a workflow.
The first scenario is indexing. A document arrives, a text splitter cuts it into chunks, and a vector store node writes each chunk to its index. That write step needs vectors, and the vectors come from this node on the embeddings port. Pick embed-english-v3.0 and every chunk becomes a 1024 dimension vector.
The second is retrieval. A question comes in, the same node embeds it, and the store compares that vector against the ones already indexed. This is why the model choice matters more than it looks: a question embedded with one model cannot be compared to chunks embedded with another. Same model on both sides, always.
The third is a question answering flow driven by an AI Agent. The agent handles the conversation and the tool calls, a chat model writes the answer, and retrieval runs underneath on vectors this node produced. Teams often mix providers here, a Cohere embedding with an OpenAI chat model, because the two jobs are separate.
When would you skip it? If your content is English only and you already embed everything elsewhere with another provider, adding a second embedding model buys you nothing and costs you a reindex. Cohere earns its place when the corpus is multilingual: the embed-multilingual-v3.0 family covers several languages in one shared space, so a French question can match an English chunk.
The limits are worth knowing before you build. This node exposes a single parameter, the model. There is no batch size, no dimension override, no truncation setting in it. Changing the model later means rebuilding the index from scratch, because vector length and geometry both change. And the Cohere side bills its own API calls under its own terms; n8n adds nothing on top. If you want a wider read on where n8n fits, the n8n review covers it, and the n8n training walks through building a retrieval flow end to end.
What do you need to authenticate?
- 01
Create a Cohere account
The node calls the Cohere API, so it needs an API key issued by Cohere. Get one from your Cohere account before you open n8n, because the node will not load the model list without a working credential attached. Nothing is installed on the n8n side: this node ships inside the n8n AI package, on n8n Cloud and on a self-hosted instance alike.
- 02
Store the key as an n8n credential
In n8n, open the Credentials menu and create a Cohere credential, then paste the key into it. A credential is a stored set of secrets that n8n keeps outside the workflow itself, which is why nobody reading an exported workflow ever sees the key. You create it once and every workflow on that instance can select it.
- 03
Attach it to the node and connect the port
Add the Embeddings Cohere node to the canvas, select the credential you just created, then drag its output to the embeddings port of a root node such as a vector store. A port is the small connector on the node body, and each port accepts one type of sub-node. Until that link exists, the node has nothing to run inside.
Where does this node plug in?
n8n splits AI work between a root node that receives the workflow items and sub-nodes that attach to its ports. This one is a sub-node, so it has no input of its own: it only offers an output that a root node consumes.
Output (what it plugs into)
- Embeddings
ai_embedding
01Embeddings
This is the single port the node exposes, and it is an output. A root node that indexes or searches vectors reads its embeddings through it.
Key parameters
- Required: a vector store node cannot write or query an index without an embeddings sub-node on this port, and the port carries vectors only, so the root node decides whether they are indexed or searched.
Which Cohere model should you pick?
The Embeddings Cohere node has one parameter. 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
Model selects which Cohere model generates the embeddings. The choice sets the vector length for everything you index, so make it before the first run rather than after.
Key parameters
- Embed-English-v3.0 (1024 Dimensions): for a corpus that is English only.
- Embed-Multilingual-v3.0 (1024 Dimensions): when questions and documents are not in the same language.
- Embed-Multilingual-Light-v3.0 (384 Dimensions): shorter vectors, lighter index, still multilingual.
- Embed-English-v2.0 (4096 Dimensions): the longest vectors on offer, kept for indexes already built on it.
Need help automating Embeddings Cohere with n8n?
A person reads every message.
Questions people ask about this node
01Is the n8n Embeddings Cohere node included for free?
02What do you need for it to work?
03What are the limits of the node?
04What do you have to connect at minimum, and where should you start?
05n8n or Make for this kind of AI flow?
Get our weekly integration tips.
No spam. Unsubscribe anytime.
