Automatisation HTTP avec n8n : traitement de messages Chat
Ce workflow n8n a pour objectif d'automatiser le traitement des messages reçus via un chat, en utilisant un modèle de langage d'OpenAI. Il est particulièrement utile pour les entreprises qui souhaitent intégrer des réponses automatiques dans leurs systèmes de messagerie, améliorant ainsi l'efficacité de la communication avec les clients. Dans un contexte où la rapidité et la pertinence des réponses sont cruciales, ce workflow permet d'analyser et de répondre aux messages en temps réel. Étape 1 : le déclencheur 'On new manual Chat Message' active le workflow dès qu'un nouveau message est reçu. Étape 2 : le noeud 'OpenAI Chat Model' génère une réponse basée sur le contenu du message. Étape 3 : une requête HTTP est effectuée via le noeud 'HTTP Request' pour traiter les données. Étape 4 : le noeud 'Exctract HTML Body' extrait le corps de la réponse, suivi par des vérifications d'erreur avec le noeud 'Is error?'. Si une erreur est détectée, le message d'erreur est formaté et renvoyé. Enfin, le contenu est simplifié et converti en Markdown pour une meilleure lisibilité. Ce workflow apporte une valeur ajoutée significative en réduisant le temps de réponse et en améliorant l'expérience utilisateur, tout en permettant une gestion efficace des interactions client. Tags clés : automatisation, OpenAI, HTTP Request.
Vue d'ensemble du workflow n8n
Schéma des nœuds et connexions de ce workflow n8n, généré à partir du JSON n8n.
Détail des nœuds du workflow n8n
Inscris-toi pour voir l'intégralité du workflow
Inscription gratuite
S'inscrire gratuitementBesoin d'aide ?{
"id": "dsKnCFwysROIA4MT",
"meta": {
"instanceId": "03524270bab2c2dfd5b82778cd1355e56cdda3cf098bf2dfd865e18164c00485"
},
"name": "Agent with custom HTTP Request",
"tags": [],
"nodes": [
{
"id": "e7374976-f3c1-4f60-ae57-9eec65444216",
"name": "On new manual Chat Message",
"type": "@n8n/n8n-nodes-langchain.manualChatTrigger",
"position": [
763,
676
],
"parameters": {},
"typeVersion": 1
},
{
"id": "97e84a23-9536-43cd-94e9-b8166be8ed32",
"name": "OpenAI Chat Model",
"type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
"position": [
983,
896
],
"parameters": {
"model": "gpt-4-1106-preview",
"options": {
"timeout": 300000,
"temperature": 0.7,
"frequencyPenalty": 0.3
}
},
"credentials": {
"openAiApi": {
"id": "wPFAzp4ZHdLLwvkK",
"name": "OpenAi account"
}
},
"typeVersion": 1
},
{
"id": "63d98361-8978-4042-84e7-53a0e226f946",
"name": "HTTP Request",
"type": "n8n-nodes-base.httpRequest",
"onError": "continueRegularOutput",
"position": [
1360,
1200
],
"parameters": {
"url": "={{ encodeURI($json.query.url) }}",
"options": {
"response": {
"response": {
"neverError": true
}
},
"allowUnauthorizedCerts": true
}
},
"typeVersion": 4.1,
"alwaysOutputData": false
},
{
"id": "17d4b5ae-f5d3-4793-8419-d3c879f7f50d",
"name": "Exctract HTML Body",
"type": "n8n-nodes-base.set",
"position": [
1780,
1480
],
"parameters": {
"fields": {
"values": [
{
"name": "HTML",
"stringValue": "={{ $json?.data.match(/<body[^>]*>([\\s\\S]*?)<\\/body>/i)[1] }}"
}
]
},
"include": "selected",
"options": {},
"includeFields": "HTML"
},
"typeVersion": 3.2
},
{
"id": "36c38ee4-724c-4ba2-a59a-ac0bbc912e94",
"name": "Is error?",
"type": "n8n-nodes-base.if",
"position": [
1560,
1200
],
"parameters": {
"conditions": {
"boolean": [
{
"value1": "={{ $json.hasOwnProperty('error') }}",
"value2": true
}
]
}
},
"typeVersion": 1
},
{
"id": "4e4d97ce-14a9-4f4f-aa75-f218784d9ed9",
"name": "Stringify error message",
"type": "n8n-nodes-base.set",
"position": [
1780,
980
],
"parameters": {
"fields": {
"values": [
{
"name": "page_content",
"stringValue": "={{ $('QUERY_PARAMS').first()?.json?.query?.url == null ? \"INVALID action_input. This should be an HTTP query string like this: \\\"?url=VALIDURL&method=SELECTEDMETHOD\\\". Only a simple string value is accepted. JSON object as an action_input is NOT supported!\" : JSON.stringify($json.error) }}"
}
]
},
"include": "selected",
"options": {},
"includeFields": "HTML"
},
"typeVersion": 3.2
},
{
"id": "8452e5c4-aa29-4a02-9579-8d9da3727bcb",
"name": "Execute Workflow Trigger",
"type": "n8n-nodes-base.executeWorkflowTrigger",
"position": [
760,
1200
],
"parameters": {},
"typeVersion": 1
},
{
"id": "063220c2-fa4d-4d5e-9549-7712aaa72921",
"name": "Remove extra tags",
"type": "n8n-nodes-base.set",
"position": [
1980,
1480
],
"parameters": {
"fields": {
"values": [
{
"name": "HTML",
"stringValue": "={{ ($json.HTML || \"HTML BODY CONTENT FOR THIS SEARCH RESULT IS NOT AVAILABLE\").replace(/<script[^>]*>([\\s\\S]*?)<\\/script>|<style[^>]*>([\\s\\S]*?)<\\/style>|<noscript[^>]*>([\\s\\S]*?)<\\/noscript>|<!--[\\s\\S]*?-->|<iframe[^>]*>([\\s\\S]*?)<\\/iframe>|<object[^>]*>([\\s\\S]*?)<\\/object>|<embed[^>]*>([\\s\\S]*?)<\\/embed>|<video[^>]*>([\\s\\S]*?)<\\/video>|<audio[^>]*>([\\s\\S]*?)<\\/audio>|<svg[^>]*>([\\s\\S]*?)<\\/svg>/ig, '')}}"
}
]
},
"options": {}
},
"typeVersion": 3.2
},
{
"id": "036511d7-a4be-4bbf-b4bc-47ddfabfe76f",
"name": "Simplify output",
"type": "n8n-nodes-base.set",
"notes": "remove links and image URLs",
"position": [
2360,
1380
],
"parameters": {
"fields": {
"values": [
{
"name": "HTML",
"stringValue": "={{ $json.HTML.replace(/href\\s*=\\s*\"(.+?)\"/gi, 'href=\"NOURL\"').replace(/src\\s*=\\s*\"(.+?)\"/gi, 'src=\"NOIMG\"')}}"
}
]
},
"options": {}
},
"notesInFlow": true,
"typeVersion": 3.2
},
{
"id": "5e2b5383-adcf-4de0-a406-4f5d631b5e8a",
"name": "Simplify?",
"type": "n8n-nodes-base.if",
"position": [
2180,
1480
],
"parameters": {
"conditions": {
"string": [
{
"value1": "={{ $('CONFIG').first()?.json?.query?.method }}",
"value2": "simplif",
"operation": "contains"
}
]
}
},
"typeVersion": 1
},
{
"id": "a0fc004a-ab0f-4b31-94df-50f5eee69c86",
"name": "QUERY_PARAMS",
"type": "n8n-nodes-base.set",
"position": [
960,
1200
],
"parameters": {
"fields": {
"values": [
{
"name": "query",
"type": "objectValue",
"objectValue": "={{ $json.query.substring($json.query.indexOf('?') + 1).split('&').reduce((result, item) => (result[item.split('=')[0]] = decodeURIComponent(item.split('=')[1]), result), {}) }}"
}
]
},
"options": {}
},
"typeVersion": 3.2
},
{
"id": "3b6599d6-ce9a-4861-9b52-07156eb52539",
"name": "CONFIG",
"type": "n8n-nodes-base.set",
"position": [
1160,
1200
],
"parameters": {
"fields": {
"values": [
{
"name": "query.maxlimit",
"type": "numberValue",
"numberValue": "={{ $json?.query?.maxlimit == null ? 70000 : Number($json?.query?.maxlimit) }}"
}
]
},
"options": {}
},
"typeVersion": 3.2
},
{
"id": "14f683be-76f6-4034-9a0e-d785738b135f",
"name": "Sticky Note",
"type": "n8n-nodes-base.stickyNote",
"position": [
721,
1134
],
"parameters": {
"width": 556.25,
"height": 235.79999999999995,
"content": "### Convert the query string into JSON, apply the limit for a page length"
},
"typeVersion": 1
},
{
"id": "6deabcb7-a984-48ec-af2a-8c70b3a4e4bf",
"name": "Sticky Note1",
"type": "n8n-nodes-base.stickyNote",
"position": [
1720,
840
],
"parameters": {
"width": 491,
"height": 285.7,
"content": "## Send an error message:\n1. If query param was incorrect, return the instruction. AI Agent should pick up on this and adapt the query on the next iteration.\n2. If the query is OK and an error was during the HTTP Request, then send back the original error message."
},
"typeVersion": 1
},
{
"id": "df1e8d00-0e18-44fa-8f94-8a53c27f7c88",
"name": "Sticky Note2",
"type": "n8n-nodes-base.stickyNote",
"position": [
1720,
1160
],
"parameters": {
"width": 1200,
"height": 472.5,
"content": "## Post-processing of the HTML page:\n1. Keep only <BODY> content\n2. Remove inline <SCRIPT> tag entirely, as well as: NOSCRIPT, IFRAME, OBJECT, EMBED, VIDEO, AUDIO, SVG, and HTML comments.\n3. In case query parameter method=simplified, replace all page URLs (a href) and IMG (src) with NOURL / NOIMG - this may save up to 20% of the page length\n4. Convert the remaining HTML to Markdown. This step further reduces the length of the page: long HTML tags and styles are eliminated, but the markdown syntax keeps some page structure. This gives much better results compared to just a blank text.\n5. Finally, check the page length. If it's too long, send an \"ERROR: PAGE CONTENT TOO LONG\" instead of the actual page. Of course, you could split the page content in chunks, but sometimes long pages just don't have a needed content, so it makes little sense to burn tokens on them."
},
"typeVersion": 1
},
{
"id": "6afe96a0-0fba-4ae1-ab8f-f7da56d420b1",
"name": "Sticky Note3",
"type": "n8n-nodes-base.stickyNote",
"position": [
720,
540
],
"parameters": {
"width": 616.8597285067872,
"height": 483.0226244343891,
"content": "## Example ReAct AI Agent\n1. Agent Prompt is default\n2. Check the description of the HTTP_Request_Tool, it guides the agent to provide a query string with several parameters instead of a JSON object"
},
"typeVersion": 1
},
{
"id": "d5ff2114-1e74-43cf-9f3c-744c241988db",
"name": "ReAct AI Agent",
"type": "@n8n/n8n-nodes-langchain.agent",
"position": [
983,
676
],
"parameters": {
"agent": "reActAgent",
"options": {
"prefix": "Answer the following questions as best you can. You have access to the following tools:",
"suffix": "Begin!\n\n\tQuestion: {input}\n\tThought:{agent_scratchpad}",
"suffixChat": "Begin! Reminder to always use the exact characters `Final Answer` when responding.",
"humanMessageTemplate": "{input}\n\n{agent_scratchpad}"
}
},
"typeVersion": 1
},
{
"id": "cc7aef4a-a1fb-4a69-a670-1f200f9e9541",
"name": "Convert to Markdown",
"type": "n8n-nodes-base.markdown",
"position": [
2540,
1480
],
"parameters": {
"html": "={{ $json.HTML }}",
"options": {},
"destinationKey": "page_content"
},
"typeVersion": 1
},
{
"id": "11806e8c-5fc4-4d9d-8144-179356993aa7",
"name": "Send Page Content",
"type": "n8n-nodes-base.set",
"position": [
2740,
1480
],
"parameters": {
"fields": {
"values": [
{
"name": "page_content",
"stringValue": "={{ $json.page_content.length < $('CONFIG').first()?.json?.query?.maxlimit ? $json.page_content : \"ERROR: PAGE CONTENT TOO LONG\" }}"
},
{
"name": "page_length",
"type": "numberValue",
"numberValue": "={{ $json.page_content.length }}"
}
]
},
"include": "selected",
"options": {}
},
"typeVersion": 3.2
},
{
"id": "a3a6b199-517b-4987-8281-d7997a32f54b",
"name": "HTTP_Request_Tool",
"type": "@n8n/n8n-nodes-langchain.toolWorkflow",
"position": [
1103,
896
],
"parameters": {
"name": "HTTP_Request_Tool",
"workflowId": "={{ $workflow.id }}",
"description": "Call this tool to fetch a webpage content. The input should be a stringified HTTP query parameter like this: \"?url=VALIDURL&method=SELECTEDMETHOD\". \"url\" parameter should contain the valid URL string. \"method\" key can be either \"full\" or \"simplified\". method=full will fetch the whole webpage content in the Markdown format, including page links and image links. method=simplified will return the Markdown content of the page but remove urls and image links from the page content for simplicity. Before calling this tool, think strategically which \"method\" to call. Best of all to use method=simplified. However, if you anticipate that the page request is not final or if you need to extract links from the page, pick method=full.",
"responsePropertyName": "page_content"
},
"typeVersion": 1
}
],
"active": false,
"pinData": {},
"settings": {
"executionOrder": "v1"
},
"versionId": "9db853c5-3658-47c1-b98a-5858b1c184ec",
"connections": {
"CONFIG": {
"main": [
[
{
"node": "HTTP Request",
"type": "main",
"index": 0
}
]
]
},
"Is error?": {
"main": [
[
{
"node": "Stringify error message",
"type": "main",
"index": 0
}
],
[
{
"node": "Exctract HTML Body",
"type": "main",
"index": 0
}
]
]
},
"Simplify?": {
"main": [
[
{
"node": "Simplify output",
"type": "main",
"index": 0
}
],
[
{
"node": "Convert to Markdown",
"type": "main",
"index": 0
}
]
]
},
"HTTP Request": {
"main": [
[
{
"node": "Is error?",
"type": "main",
"index": 0
}
]
]
},
"QUERY_PARAMS": {
"main": [
[
{
"node": "CONFIG",
"type": "main",
"index": 0
}
]
]
},
"Simplify output": {
"main": [
[
{
"node": "Convert to Markdown",
"type": "main",
"index": 0
}
]
]
},
"HTTP_Request_Tool": {
"ai_tool": [
[
{
"node": "ReAct AI Agent",
"type": "ai_tool",
"index": 0
}
]
]
},
"OpenAI Chat Model": {
"ai_languageModel": [
[
{
"node": "ReAct AI Agent",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"Remove extra tags": {
"main": [
[
{
"node": "Simplify?",
"type": "main",
"index": 0
}
]
]
},
"Exctract HTML Body": {
"main": [
[
{
"node": "Remove extra tags",
"type": "main",
"index": 0
}
]
]
},
"Convert to Markdown": {
"main": [
[
{
"node": "Send Page Content",
"type": "main",
"index": 0
}
]
]
},
"Execute Workflow Trigger": {
"main": [
[
{
"node": "QUERY_PARAMS",
"type": "main",
"index": 0
}
]
]
},
"On new manual Chat Message": {
"main": [
[
{
"node": "ReAct AI Agent",
"type": "main",
"index": 0
}
]
]
}
}
}Pour qui est ce workflow ?
Ce workflow s'adresse aux entreprises et équipes techniques qui utilisent des systèmes de messagerie et souhaitent automatiser leurs réponses. Il est idéal pour les PME et les startups qui cherchent à améliorer leur service client tout en optimisant leurs ressources.
Problème résolu
Ce workflow résout le problème de la lenteur dans le traitement des messages entrants en automatisant les réponses. Il élimine les frustrations liées aux délais de réponse, réduit le risque d'erreurs humaines et garantit une communication fluide avec les clients. Grâce à cette automatisation, les utilisateurs peuvent s'attendre à une interaction plus rapide et efficace, ce qui améliore la satisfaction client.
Étapes du workflow
Étape 1 : le workflow est déclenché par un nouveau message dans le chat. Étape 2 : le modèle de langage d'OpenAI génère une réponse appropriée. Étape 3 : une requête HTTP est envoyée pour traiter les données. Étape 4 : le corps de la réponse est extrait et vérifié pour d'éventuelles erreurs. Étape 5 : si une erreur est détectée, un message d'erreur est formaté. Étape 6 : le contenu est simplifié et converti en Markdown pour une meilleure présentation.
Guide de personnalisation du workflow n8n
Pour personnaliser ce workflow, vous pouvez modifier l'URL dans le noeud 'HTTP Request' pour l'adapter à votre API. Ajustez également les paramètres du modèle OpenAI dans le noeud 'OpenAI Chat Model' pour affiner les réponses générées. Si vous souhaitez ajouter d'autres outils, vous pouvez intégrer des noeuds supplémentaires après le traitement des messages. Assurez-vous de sécuriser les données en configurant correctement les paramètres de sécurité dans chaque noeud.