n8n Ollama nodeConfigure Ollama in n8n.
The n8n Ollama node talks to the models running on your own Ollama server. It ships 2 operations, one for text and one for images, both built on a Model selector and a shared Options collection of 25 sampling settings. Useful when the data cannot leave the building.
Verified Trustpilot reviews · AI, automation & growth agency
What does the n8n Ollama node do in a workflow?
It sends a prompt to a model hosted on an Ollama server and returns the model's answer as workflow data. The node runs like any other action node, once per incoming item, and has no sub-node ports, so nothing gets plugged underneath it. Two resources split the work: Text sends a conversation, Image takes pictures and answers questions about them.
Start with the classifier case. A support form lands in a workflow, the Text resource gets a System Message that spells out the categories, Temperature goes low so the output stops drifting, and the category comes back ready to route. Set Output Format to json and the answer parses without a cleanup step, which matters when the next node writes to Google Sheets.
Second case, document images. A scanned delivery note arrives as a binary file, which is a file attached to the item rather than text. The Image resource reads it through Input Data Field Name(s), the Text Input field carries the question, and the extracted reference goes into a reply drafted by Gmail. Hosted vision models such as OpenAI do the same job, but here the image never leaves your network.
Third case, bulk rewriting. A list of product descriptions runs through the node item by item, Max Tokens caps the length, and Seed makes a rerun produce the same text for the same prompt, so a diff between two runs means something.
When to pick something else. This node answers a prompt and stops. It holds no conversation history of its own, it calls no tools, and it makes no decisions about what to do next, so anything that needs memory or tool calls belongs to the AI Agent node instead. The node also cannot pull a model down for you: the model has to already exist on the Ollama server you point at.
Known limits. The Options collection exposes 25 settings, several of them tied to the machine rather than the prompt (Number of GPUs, Number of CPU Threads, Low VRAM Mode), which is a fair warning: a local model is only as fast as the hardware under it. Thinking only applies to models that support it, and Simplify Output is the fastest way to stop reading through a raw API envelope. If you are still choosing a platform, the n8n review covers the self-hosting side in more detail.
What do you need to connect before the node runs?
- 01
Create the Ollama credential
Open the Credentials menu in n8n and add an Ollama credential. It asks for a Base URL, defaulting to
http://localhost:11434, and an optional API Key for a proxy expecting a Bearer token. A credential is the stored connection n8n reuses instead of asking again in every workflow, so this step happens once. - 02
Attach it to the node
Drop the Ollama node on the canvas and select that credential at the top of its panel. The same credential serves both resources, so a workflow that classifies text and another that reads images share one entry and one place to fix things when the server moves.
- 03
Pick the model
Fill Model, the only required field on both operations. It is a resource locator, meaning you either choose an entry from the list n8n loads or type the identifier yourself, which is what you do when the model name comes from an earlier node as
{{ $json.model }}.
What can the Ollama node do?
The Ollama node exposes 2 operations across 2 resources. For each one: the node as you configure it in n8n, the required fields, and our field notes.
Image
1 operationAnalyze image
image.analyzeWhat you see in n8n
Notes & use cases
Feeds one or more images to a vision-capable model and returns its answer to a question you write. Good for reading what is printed on a document rather than describing a photo.
Key parameters
- Model: required, the vision model to call, picked from the list or typed in.
- Input Type:
binaryfor files carried by the item,urlfor images already online. - Input Data Field Name(s): the binary field holding the image, often
data, several names separated by commas. - URL(s): the image addresses when you chose
url, comma separated. - Text Input: the question asked about the image.
Text
1 operationMessage a model
text.messageWhat you see in n8n
Notes & use cases
Sends a conversation to the model and returns the reply. Nothing is remembered between executions, so the whole exchange you want the model to see travels in the request.
Key parameters
- Model: required, the text model that answers.
- Messages: the list of turns, each one a Content string and a Role,
userfor the question andassistantto replay an earlier answer as history. - System Message: the standing instruction that frames every turn.
- Simplify Output: on, you get the answer alone instead of the full response payload.
Need help automating Ollama with n8n?
A person reads every message.
Ollama and n8n, common questions
01Is the Ollama node included in n8n?
02What do you need for the n8n Ollama node to work?
03What are the limits of the Ollama node?
04Which operation should you start with?
05n8n or Make for running local models?
Get our weekly integration tips.
No spam. Unsubscribe anytime.
