Resources · n8n integration

n8n DeepSeek Chat Model nodeConfigure DeepSeek Chat Model in n8n.

An AI Agent in n8n does nothing until a chat model is wired to it. The n8n DeepSeek Chat Model node is that model: a sub-node, meaning a helper node that plugs into a port of a bigger node instead of running on its own. It exposes 9 parameters and connects through the ai_languageModel port.

Try the DeepSeek Chat Model node on n8nSee the 9 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n DeepSeek Chat Model node actually do?

It supplies the language model that a root node uses to generate text. In n8n's AI setup, a root node such as an AI Agent or a chain receives workflow items on its main input, and sub-nodes attach to its ports: one chat model, one memory, as many tools as you need. The DeepSeek Chat Model node fills the model slot, and nothing else. On its own, it never runs.

Think of it as the engine you drop into a chassis. Three shapes come up again and again once you start building.

The first is a conversational assistant. You wire DeepSeek Chat Model into the AI Agent node, give the agent a memory sub-node so it remembers the thread, and hand it tools so it can act. The agent decides what to do, DeepSeek writes the words. The second is structured extraction: you set Response Format to json_object, so the model returns valid JSON that the next node can read field by field, then push the result into Google Sheets or post it to Slack. The third is quiet background classification: a summarization or classifier chain calls the model once per item, with Sampling Temperature low so two identical inputs give the same answer twice.

When is another node the better call? If you already run everything on OpenAI credentials and want the same provider across a workflow, the OpenAI node covers that ground. If the model keeps returning JSON that breaks your parser, the Auto-fixing Output Parser takes a model of its own and repairs the output instead of you tightening prompts forever.

The limits are worth knowing before you build. This node has no main input and no main output: you cannot place it in the middle of a workflow and expect items to flow through it. It only attaches to a root node's model port. It carries no prompt of its own either, since the prompt belongs to the root node. And the model list is loaded live from DeepSeek using your credential, so you see the models your account can call, not a hardcoded list.

Cost sits with DeepSeek, not with n8n. The node ships with n8n, the API calls are billed by the provider under its own terms. Teams that want to compare platforms before committing usually read the n8n review first.

Connect

How do you connect DeepSeek to n8n?

  1. 01

    Open the Credentials menu

    In n8n, credentials live in their own menu, separate from any workflow. Create the DeepSeek credential there once and every workflow on that instance can reuse it. This is the same on n8n Cloud and on a self-hosted instance, so a workflow you export from one runs on the other once the credential exists on both sides.

  2. 02

    Attach it to the node

    Drop a DeepSeek Chat Model node on the canvas and pick the credential in the node panel. There is no Authentication selector to choose from on this node, so the credential is the only thing standing between you and a working model. Save the workflow before you test, otherwise n8n runs the version it last stored.

  3. 03

    Check the model list loads

    Open the Model dropdown. n8n queries DeepSeek with your credential and fills the list with the models your account can reach. If the list comes back empty, the credential is the suspect, not the node. Pick a model, then attach the node to a root node before you run anything.

Connections

Where does this node plug in?

n8n splits AI work between a root node that runs in the workflow and sub-nodes that hang off its ports. This node has no input port of its own: it exposes one output, and that output goes into a root node's model slot.

Output (what it plugs into)

  • Modelai_languageModel
01ModelThis is the only port the node uses, and it points outward: you drag from the node up into the model slot of whatever root node is doing the work.

This is the only port the node uses, and it points outward: you drag from the node up into the model slot of whatever root node is doing the work.

Key parameters

  • Required: a root node cannot generate anything without a model on this port, so the connection is not optional.
  • AI Agent: the usual destination when the model has to reason, pick a tool and answer.
  • Basic LLM Chain: the simpler route when you just want a prompt in and text out, with no tool selection.
  • Summarization Chain: takes the model to condense long documents item by item.
  • Text Classifier: uses the model to sort each incoming item into one of your categories.
Use cases
a support workflow where the same DeepSeek node feeds an agent on one branch and a classifier on another.
Parameters

Which parameters change what the model does?

The DeepSeek Chat Model node has 9 parameters. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

Model

model

What you see in n8n

Notes & use cases

Picks which DeepSeek model generates the completion. The dropdown is not a static list: n8n calls DeepSeek with your credential and shows only the models your account can use, so two people on the same instance can see different options.

Key parameters

  • Model: the model that will generate the completion, chosen from the list n8n loads from DeepSeek.
Use cases
keeping one model for a customer-facing agent and switching a second, cheaper node to another model for batch classification, without touching either root node.
02

Frequency Penalty

options.frequencyPenalty

What you see in n8n

Notes & use cases

Pushes the model away from repeating itself. Positive values penalize tokens based on how often they have already appeared in the text so far, which lowers the odds of the model repeating the same line verbatim.

Key parameters

  • Frequency Penalty: a number; the higher it goes, the less the model repeats itself.
Use cases
product descriptions generated in bulk, where every third item otherwise ends on the same closing sentence. Nudge this up before you rewrite the prompt.
03

Maximum Number of Tokens

options.maxTokens

What you see in n8n

Notes & use cases

Caps how long the answer can get. A token is a chunk of text, roughly a word fragment, and this setting fixes the completion length. Most models carry a context length of 2048 tokens, except for the newest ones, which support 32768.

Key parameters

  • Maximum Number of Tokens: a number, the ceiling on tokens generated in the completion.
Use cases
a Slack digest that has to stay short. Set the ceiling low and the model stops padding, instead of you trimming the text downstream.
04

Response Format

options.responseFormat

What you see in n8n

Notes & use cases

Decides whether the answer comes back as prose or as machine-readable data. Switch it the moment a later node has to read fields rather than display a paragraph.

Key parameters

  • Text (text): a regular text response, the default shape for anything a human reads.
  • JSON (json_object): enables JSON mode, which should guarantee the message the model generates is valid JSON.
Use cases
pulling a name, an amount and a due date out of an email body, then mapping each one with {{ $json.amount }} in the next node.
05

Presence Penalty

options.presencePenalty

What you see in n8n

Notes & use cases

Steers the model toward new ground. Positive values penalize tokens based on whether they have appeared in the text at all, which raises the likelihood of the model talking about new topics.

Key parameters

  • Presence Penalty: a number; higher values increase the chance the model moves to new topics.
Use cases
a brainstorming step that should list ten angles on a campaign rather than circle the first two. Note that it works differently from Frequency Penalty, which counts how often a token appeared instead of whether it appeared.
06

Sampling Temperature

options.temperature

What you see in n8n

Notes & use cases

Controls randomness. Lowering it results in less random completions, and as the temperature approaches zero the model becomes deterministic and repetitive. Raise it for more diverse sampling, at the cost of a higher risk of hallucinations.

Key parameters

  • Sampling Temperature: a number; near zero the answers stop varying, higher values loosen them.
Use cases
classification and data extraction want it low, so the same input gives the same label twice; ad copy variants want it higher.
07

Timeout

options.timeout

What you see in n8n

Notes & use cases

Sets the maximum amount of time a request is allowed to take, in milliseconds. Past that, the request stops instead of holding the execution open.

Key parameters

  • Timeout: a number in milliseconds, so 2000 means two seconds, not two thousand.
Use cases
a chatbot answering in a live conversation, where a slow reply is worse than a failed one. The unit trips people up more than the setting itself.
08

Max Retries

options.maxRetries

What you see in n8n

Notes & use cases

Says how many times n8n retries a request that failed before it gives up and the execution errors out.

Key parameters

  • Max Retries: a number, the maximum count of retry attempts per request.
Use cases
an overnight batch that walks through a few hundred rows. One transient failure should not kill the run, so a couple of retries here saves a manual restart in the morning. Pair it with a sane Timeout, since retries multiply the wait.
09

Top P

options.topP

What you see in n8n

Notes & use cases

Controls diversity through nucleus sampling: 0.5 means half of all likelihood-weighted options are considered. A lower value tells the model to ignore the less probable options.

Key parameters

  • Top P: a number between zero and one, the slice of probable options the completion draws from.
Use cases
tightening answers on a factual assistant. The usual advice is to change this or the temperature, not both, so pick one lever and leave the other alone.
Need help

Need help automating DeepSeek Chat Model with n8n?

A person reads every message.

FAQ

Common questions

01Is the n8n DeepSeek Chat Model node included with n8n Cloud and self-hosted?
Yes. It belongs to n8n's AI package, which ships with n8n, so there is nothing to install and nothing extra to pay on the n8n side. It behaves the same on n8n Cloud, the hosted offer run by n8n, and on a self-hosted instance installed with Docker or npm under the Community Edition and its Sustainable Use license. What does cost money is the model itself: DeepSeek bills the API calls you make under its own terms, exactly as it would outside n8n. The node adds no markup and no per-call fee of its own.
02What do you need for the DeepSeek Chat Model node to work?
A DeepSeek credential and a root node to attach to. You create the credential once in the Credentials menu and reuse it everywhere on that instance. The node has no Authentication selector, so the credential is the whole setup. Beyond that, remember this is a sub-node: it has no main input and never runs on its own, which means dropping it on the canvas and hitting execute does nothing. Connect it to the model port of an AI Agent, a chain or another root node, and the root node is what you run.
03What are the limits of this node?
It does one job: hand a DeepSeek model to a root node. It carries no prompt, no system message and no output of its own, because all of that belongs to the root node you attach it to. It has no main input, so items never flow through it. The page describes version 1 of the node, the highest one in the catalog; an older workflow may show an earlier version with fewer options. The model dropdown is loaded live from DeepSeek, so it shows only what your account can reach, and a broken credential looks exactly like an empty model list.
04What is the minimum to plug in, and which node should a beginner start with?
At minimum, one root node plus this model on its Model port, with a valid credential. For a first build, Basic LLM Chain is the gentler entry point: prompt in, text out, no tool selection to reason about. Once you need the model to choose between actions, move up to the AI Agent node and add a memory sub-node so the conversation holds together across messages. Tools come after that. Adding an agent, a memory and four tools on day one is the fastest way to lose track of which piece broke.
05n8n or Make for this kind of AI workflow?
It depends on where you want the workflow to live. n8n can be self-hosted with Docker or npm, or used on n8n Cloud, and the workflow is identical either way, which matters when prompts and customer data should stay on your own servers. Make is hosted only, with no self-hosting option, and bills per operation, so a chatty agent that loops through tool calls has a cost shape you should model before you commit. Both give you a visual canvas. The sub-node model described here, one model plugged into a root node, is specific to n8n.
Hack'celeration Lab

Get our weekly integration tips.

No spam. Unsubscribe anytime.

Same platform

Other docs you might need on this stack.

Browse all integrations