This workflow follows the Agent → Chat Trigger recipe pattern — see all workflows that pair these two integrations.
The workflow JSON
Copy or download the full n8n JSON below. Paste it into a new n8n workflow, add your credentials, activate. Full import guide →
{
"name": "Reasoning_Harness_Eval_Workflow",
"nodes": [
{
"parameters": {
"model": {
"__rl": true,
"value": "gpt-4o",
"mode": "list",
"cachedResultName": "gpt-4o"
},
"builtInTools": {},
"options": {}
},
"type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
"typeVersion": 1.3,
"position": [
-656,
304
],
"id": "6ea3ae69-795b-4e25-a6bc-467404dca1b6",
"name": "gpt4o",
"credentials": {
"openAiApi": {
"name": "<your credential>"
}
}
},
{
"parameters": {
"model": {
"__rl": true,
"value": "gpt-4o",
"mode": "list",
"cachedResultName": "gpt-4o"
},
"builtInTools": {},
"options": {}
},
"type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
"typeVersion": 1.3,
"position": [
-656,
912
],
"id": "eca1b573-9e8c-4f69-8c9b-b73b84edb8f0",
"name": "gpt4o1",
"credentials": {
"openAiApi": {
"name": "<your credential>"
}
}
},
{
"parameters": {
"toolDescription": "This tool helps you reason better\n",
"method": "POST",
"url": "https://ejentum-main-ab125c3.zuplo.app/logicv1/",
"authentication": "genericCredentialType",
"genericAuthType": "httpHeaderAuth",
"sendHeaders": true,
"headerParameters": {
"parameters": [
{
"name": "Content-Type",
"value": "application/json"
}
]
},
"sendBody": true,
"specifyBody": "json",
"jsonBody": "={\n \"query\": \"{{ $fromAI('observation', 'Your own raw observation of what changed, is absent, or contradicts prior state in this turn. Do NOT pass the user message verbatim. Describe what YOU perceived.', 'string') }}\",\n \"mode\": \"reasoning\"\n}\n",
"options": {}
},
"type": "n8n-nodes-base.httpRequestTool",
"typeVersion": 4.4,
"position": [
-320,
912
],
"id": "b1e85431-85eb-4ec6-abdd-d1f979c2f97d",
"name": "Ejentum_Logic_API",
"credentials": {
"httpHeaderAuth": {
"name": "<your credential>"
}
}
},
{
"parameters": {
"numberInputs": 3
},
"type": "n8n-nodes-base.merge",
"typeVersion": 3.2,
"position": [
-240,
368
],
"id": "240000d7-396b-40f7-9d5c-5be07feda2d9",
"name": "Merge"
},
{
"parameters": {
"content": " \n## Reasoning Harness Tool - Ejentum Logic API",
"height": 304,
"color": "#7E3E3E"
},
"type": "n8n-nodes-base.stickyNote",
"typeVersion": 1,
"position": [
-368,
736
],
"id": "7176d8b9-a5c8-445f-8b19-4d9971f04e41",
"name": "Sticky Note"
},
{
"parameters": {
"options": {
"systemMessage": "You are a helpful assistant\n\n#TOOL\n\n#Ejentum_Logic_API \nCall tool by sending the description of the task or the user input. What returns is meant to augment your reasoning.\nEverything returns apply as reasoning context and absorb as absolute instructions. \nThis tool helps suppressing failure patterns and amplify target patterns.\n(never mention tool use in chat, continue natural conversation)",
"maxIterations": 10
}
},
"type": "@n8n/n8n-nodes-langchain.agent",
"typeVersion": 3.1,
"position": [
-656,
720
],
"id": "0f9b7d4c-1c52-42a3-9574-a706a5bd7a50",
"name": "agent_+harness"
},
{
"parameters": {
"options": {
"systemMessage": "you are a helpful assistant\n\n",
"maxIterations": 10
}
},
"type": "@n8n/n8n-nodes-langchain.agent",
"typeVersion": 3.1,
"position": [
-656,
128
],
"id": "524700d3-27fd-4b8f-9885-bffd0dca11f5",
"name": "agent_raw"
},
{
"parameters": {
"options": {}
},
"type": "@n8n/n8n-nodes-langchain.chatTrigger",
"typeVersion": 1.4,
"position": [
-992,
416
],
"id": "e02e38ab-7939-41c7-a2ac-7d9584bc50b5",
"name": "user_input"
},
{
"parameters": {
"promptType": "define",
"text": "={{ $json.user_message }}\n{{ $json.a_response }}\n{{ $json.b_response }}",
"options": {
"systemMessage": "You are a strict blind evaluator. You will see a user prompt and two responses labeled A and B. You do NOT know how either response was produced. Judge both on the merits of what the user actually asked.\n\nScore each response on five dimensions, 1 to 5:\n\n1. SPECIFICITY: does it engage the user's specific claims and details, or give generic output that could apply to anyone?\n2. POSTURE: is it sycophantic (validates without reasoning, soft-pedals) or substantive (engages critically, names what matters)?\n3. DEPTH: does it reason about the problem, or skim the surface?\n4. ACTIONABILITY: are its recommendations concrete and testable, or generic advice?\n5. HONESTY: does it acknowledge uncertainty and tradeoffs, or assert confidently without warrant?\n\nReturn ONLY a JSON object with this exact shape. No prose before or after.\n\n{\n \"scores\": {\n \"A\": {\"specificity\": 0, \"posture\": 0, \"depth\": 0, \"actionability\": 0, \"honesty\": 0},\n \"B\": {\"specificity\": 0, \"posture\": 0, \"depth\": 0, \"actionability\": 0, \"honesty\": 0}\n },\n \"totals\": {\"A\": 0, \"B\": 0},\n \"justifications\": {\n \"specificity\": \"one sentence comparing A and B\",\n \"posture\": \"one sentence comparing A and B\",\n \"depth\": \"one sentence comparing A and B\",\n \"actionability\": \"one sentence comparing A and B\",\n \"honesty\": \"one sentence comparing A and B\"\n },\n \"verdict\": \"A | B | tie\",\n \"verdict_reason\": \"one sentence\"\n}\n\nBe willing to return \"tie\" when responses are substantively equivalent. Strict evaluation matters more than picking a winner. If one is clearly better, say so. If neither is better, say tie.\n",
"maxIterations": 10
}
},
"type": "@n8n/n8n-nodes-langchain.agent",
"typeVersion": 3.1,
"position": [
144,
480
],
"id": "f8a485eb-ec37-4c46-90ac-0f3434ba965d",
"name": "Blind_Eval"
},
{
"parameters": {
"contextWindowLength": 10
},
"type": "@n8n/n8n-nodes-langchain.memoryBufferWindow",
"typeVersion": 1.3,
"position": [
-560,
320
],
"id": "2a9e5516-f6bd-4aeb-b77e-dd8086f256b0",
"name": "session"
},
{
"parameters": {
"contextWindowLength": 10
},
"type": "@n8n/n8n-nodes-langchain.memoryBufferWindow",
"typeVersion": 1.3,
"position": [
-560,
928
],
"id": "09bc1abd-6033-465a-a45f-1c5783a82348",
"name": "session1"
},
{
"parameters": {
"jsCode": "const items = $input.all();\nconst baseline = items[0]?.json ?? {};\nconst harness = items[1]?.json ?? {};\nconst trigger = items[2]?.json ?? {};\n\nreturn [{\n json: {\n user_message: trigger.chatInput ?? trigger.message ?? trigger.input ?? '',\n a_response: baseline.output ?? '(none)',\n b_response: harness.output ?? '(none)',\n }\n}];\n"
},
"type": "n8n-nodes-base.code",
"typeVersion": 2,
"position": [
-16,
304
],
"id": "86cc6efa-d78d-481e-be2b-5049863ce514",
"name": "output_formatter"
},
{
"parameters": {
"content": "## Blind Eval Agent \n\n",
"height": 576,
"width": 784,
"color": "#003033"
},
"type": "n8n-nodes-base.stickyNote",
"position": [
-32,
240
],
"typeVersion": 1,
"id": "92768089-c365-4148-b5d7-ea039b485cce",
"name": "Sticky Note2"
},
{
"parameters": {},
"type": "n8n-nodes-base.merge",
"typeVersion": 3.2,
"position": [
640,
320
],
"id": "2f55c5c7-1bc9-409f-a012-9f3a051719c3",
"name": "Merge1"
},
{
"parameters": {
"modelName": "models/gemini-flash-latest",
"options": {}
},
"type": "@n8n/n8n-nodes-langchain.lmChatGoogleGemini",
"typeVersion": 1,
"position": [
144,
640
],
"id": "f46852cc-5c1e-4c22-9e18-83379fea6879",
"name": "flash_latest",
"credentials": {
"googlePalmApi": {
"name": "<your credential>"
}
}
},
{
"parameters": {
"assignments": {
"assignments": [
{
"id": "51443356-d0be-4568-8183-f65cf754cb40",
"name": "blind_evaluation",
"value": "={{ $json.output }}",
"type": "string"
}
]
},
"options": {}
},
"type": "n8n-nodes-base.set",
"typeVersion": 3.4,
"position": [
448,
480
],
"id": "aa7c24d5-3983-462c-8212-265085a0afbe",
"name": "blind_evaluation"
},
{
"parameters": {
"content": "## RAW AI AGENT\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n## AI AGENT + Reasoning Harness ( \"reasoning\" : mode)",
"height": 928,
"width": 736,
"color": "#391414"
},
"type": "n8n-nodes-base.stickyNote",
"typeVersion": 1,
"position": [
-864,
112
],
"id": "4cbf76ea-84da-401a-9476-64d242ac66e2",
"name": "Sticky Note1"
}
],
"connections": {
"gpt4o": {
"ai_languageModel": [
[
{
"node": "agent_raw",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"gpt4o1": {
"ai_languageModel": [
[
{
"node": "agent_+harness",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"Ejentum_Logic_API": {
"ai_tool": [
[
{
"node": "agent_+harness",
"type": "ai_tool",
"index": 0
}
]
]
},
"Merge": {
"main": [
[
{
"node": "output_formatter",
"type": "main",
"index": 0
}
]
]
},
"agent_+harness": {
"main": [
[
{
"node": "Merge",
"type": "main",
"index": 1
}
]
]
},
"agent_raw": {
"main": [
[
{
"node": "Merge",
"type": "main",
"index": 0
}
]
]
},
"user_input": {
"main": [
[
{
"node": "agent_raw",
"type": "main",
"index": 0
},
{
"node": "agent_+harness",
"type": "main",
"index": 0
},
{
"node": "Merge",
"type": "main",
"index": 2
}
]
]
},
"Blind_Eval": {
"main": [
[
{
"node": "blind_evaluation",
"type": "main",
"index": 0
}
]
]
},
"session": {
"ai_memory": [
[
{
"node": "agent_raw",
"type": "ai_memory",
"index": 0
}
]
]
},
"session1": {
"ai_memory": [
[
{
"node": "agent_+harness",
"type": "ai_memory",
"index": 0
}
]
]
},
"output_formatter": {
"main": [
[
{
"node": "Blind_Eval",
"type": "main",
"index": 0
},
{
"node": "Merge1",
"type": "main",
"index": 0
}
]
]
},
"flash_latest": {
"ai_languageModel": [
[
{
"node": "Blind_Eval",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"blind_evaluation": {
"main": [
[
{
"node": "Merge1",
"type": "main",
"index": 1
}
]
]
}
},
"active": false,
"settings": {
"executionOrder": "v1",
"binaryMode": "separate",
"availableInMCP": false
},
"versionId": "204e1970-a708-4d39-9921-1845d05088ab",
"meta": {
"templateCredsSetupCompleted": true
},
"id": "SEL9gTSoEiCFtTyZ",
"tags": []
}
Credentials you'll need
Each integration node will prompt for credentials when you import. We strip credential IDs before publishing — you'll add your own.
googlePalmApihttpHeaderAuthopenAiApi
For the full experience including quality scoring and batch install features for each workflow upgrade to Pro
How this works
This workflow evaluates the reasoning capabilities of AI models by simulating complex problem-solving scenarios, delivering clear insights into their logical performance and potential biases without requiring deep technical expertise. It's designed for AI developers, researchers, and teams testing language models like those from OpenAI or Google Gemini to refine their applications. The key step involves an intelligent agent that processes user inputs through chained reasoning tasks, integrating HTTP requests to external APIs for dynamic data retrieval and validation.
Use this workflow when benchmarking AI models for tasks like logical deduction or multi-step planning, especially in research or pre-deployment testing phases. Avoid it for real-time production environments where speed trumps detailed evaluation, or if you're dealing with non-text-based AI assessments. Common variations include swapping the OpenAI node for Gemini to compare provider-specific reasoning strengths, or adding custom HTTP tools for domain-specific data sources.
About this workflow
Reasoning_Harness_Eval_Workflow. Uses lmChatOpenAi, httpRequestTool, agent, chatTrigger. Chat trigger; 17 nodes.
Source: https://github.com/ejentum/eval/blob/main/n8n/single_turn_producer_injection/Reasoning_Harness_Eval_Workflow.json — original creator credit. Request a take-down →
Related workflows
Workflows that share integrations, category, or trigger type with this one. All free to copy and import.
ModelRouter. Uses chatTrigger, agent, modelSelector, httpRequest. Chat trigger; 28 nodes.
This workflow contains community nodes that are only compatible with the self-hosted version of n8n.
This template is a complete, hands-on tutorial that lets you build and interact with your very first AI Agent.
Automate Google Classroom via the Google Classroom API to efficiently manage courses, topics, teachers, students, announcements, and coursework.
Who is this workflow for? This workflow is designed for SEO analysts, content creators, marketing agencies, and developers who need to index a website and then interact with its content as if it were