n8n Chat Model with Reasoning

A community node for n8n that adds extended reasoning / thinking support to AI agents and chains — something the built-in n8n nodes don't expose today.

Works with OpenRouter, DeepSeek, and any OpenAI-compatible API that supports reasoning models.

Why this node exists

When you use a reasoning model (like DeepSeek-R1 or Claude with extended thinking) through n8n's built-in OpenRouter or DeepSeek nodes, the reasoning parameter is never sent to the API. The model responds as a regular chat model — no chain-of-thought, no deeper reasoning.

This node fixes that by letting you configure and send the reasoning parameter directly in every API call.

Installation

In your n8n instance:

Settings → Community Nodes → Install

@bruno_groehs/n8n-nodes-chatmodel-reasoning

After installation, restart n8n (or hard-refresh the browser). The node will appear as "Chat Model with Reasoning" in the AI section.

Configuration

Credential — OpenAI Compatible API (Reasoning)

| Field | Description | Example | |---|---|---| | API Key | Your API key | sk-or-... | | Base URL | API endpoint | https://openrouter.ai/api/v1 |

Supported base URLs:

• OpenRouter: https://openrouter.ai/api/v1
• DeepSeek native: https://api.deepseek.com/v1
• Any OpenAI-compatible provider that supports reasoning

Node parameters

| Parameter | Description | |---|---| | Model | Dynamic list loaded from the API. Defaults to deepseek/deepseek-r1. | | Enable Reasoning | Sends the reasoning parameter to the API to activate extended thinking. | | Reasoning Effort | low / medium / high — controls how much the model thinks. Ignored when Max Reasoning Tokens > 0. | | Max Reasoning Tokens | Token budget for thinking (overrides Effort when > 0). | | Capture Reasoning | When on, the thinking text is returned from the API and stored in additional_kwargs._reasoning (visible in the execution log). When off, the model reasons silently — better quality answers, lower output cost. |

Options (collection)

| Option | Default | |---|---| | Sampling Temperature | 0.7 | | Maximum Number of Tokens | -1 (model default) | | Top P | 1 | | Frequency Penalty | 0 | | Presence Penalty | 0 | | Response Format | Text or JSON | | Timeout (ms) | 360,000 | | Max Retries | 2 |

Note for reasoning models: Many reasoning models (DeepSeek-R1, Claude 3.7 Sonnet) require temperature = 1. Set this in Options if you get errors.

How reasoning works

What gets sent to the API

When Enable Reasoning is on, the node adds this to the request body:

{
  "model": "deepseek/deepseek-r1",
  "messages": [...],
  "reasoning": { "effort": "high" }
}

Or with a token budget:

{
  "reasoning": { "max_tokens": 8000 }
}

Text vs. Thinking — separation

The final answer and the thinking content are always separate:

• $json.output on the AI Agent node → clean final answer (what goes to the client)
• additional_kwargs._reasoning → the thinking content (for debugging, visible in the node's execution log)

The content field is never polluted with <thinking> tags.

Supported response formats

The fetch interceptor normalises both API formats:

| Provider | Reasoning field | |---|---| | OpenRouter | choices[0].message.reasoning | | DeepSeek native | choices[0].message.reasoning_content |

Both are mapped to additional_kwargs._reasoning in the LangChain message.

Compatible models

Any model accessible via OpenRouter or a compatible API that supports the reasoning parameter:

| Model | Provider | Notes | |---|---|---| | deepseek/deepseek-r1 | OpenRouter | Native reasoning model | | deepseek/deepseek-r1-0528 | OpenRouter | Latest R1 snapshot | | deepseek-reasoner | DeepSeek native | R1 via official API | | anthropic/claude-sonnet-4-5 | OpenRouter | Via OpenRouter reasoning param | | anthropic/claude-opus-4-7 | OpenRouter | Adaptive thinking | | google/gemini-2.5-pro | OpenRouter | Thinking mode | | google/gemini-2.5-flash | OpenRouter | Fast thinking |

Workflow example

Webhook → Set Node → AI Agent (Secretária)
                         ├── Chat Model (Reasoning)   ← this node
                         ├── Memory
                         └── Tools

The node connects to the Chat Model slot of any AI Agent or AI Chain — exactly like the built-in OpenRouter or DeepSeek nodes.

Development

git clone https://github.com/your-username/n8n-nodes-chatmodel-reasoning
cd n8n-nodes-chatmodel-reasoning
npm install
npm run build

To watch for changes:

npm run dev

Project structure

src/
├── credentials/
│   └── OpenAiCompatibleReasoningApi.credentials.ts
├── nodes/
│   └── LmChatReasoning/
│       ├── LmChatReasoning.node.ts   ← main node
│       └── reasoning.png
└── index.ts

Publishing

npm run build
npm publish --access public

License

MIT