Hi tsellhorn,
Thank you very much for your help.
This is my actual State:
{
"name": "SQL AI Agent - funktioniert fast",
"nodes": [
{
"parameters": {
"operation": "executeQuery",
"query": "USE datawarehouse;\n\nSELECT TABLE_NAME, COLUMN_NAME, DATA_TYPE\nFROM INFORMATION_SCHEMA.COLUMNS\nWHERE TABLE_SCHEMA = 'dbo'\nORDER BY TABLE_NAME, ORDINAL_POSITION;\n"
},
"id": "ba2700cc-b96d-4bfc-b04c-a91dea98709f",
"name": "Get DB Schema",
"type": "n8n-nodes-base.microsoftSql",
"typeVersion": 1,
"position": [
1500,
140
],
"alwaysOutputData": true,
"credentials": {
"microsoftSql": {
"id": "a643CpWdK4y4gPDH",
"name": "Microsoft SQL account"
}
}
},
{
"parameters": {
"jsCode": "const inputItems = $input.all(); // Alle Zeilen aus dem vorherigen Node\nconst tableMap = {}; // Objekt zur Gruppierung der Spalten pro Tabelle\n\nfor (const item of inputItems) {\n const { TABLE_NAME, COLUMN_NAME, DATA_TYPE } = item.json;\n\n if (!tableMap[TABLE_NAME]) {\n tableMap[TABLE_NAME] = [];\n }\n\n tableMap[TABLE_NAME].push(`- ${COLUMN_NAME}: ${DATA_TYPE}`);\n}\n\n// Formatiertes Schema erstellen\nlet promptSchema = 'Die Datenbank `datawarehouse` enthält folgende Tabellen:\\n\\n';\nfor (const [table, columns] of Object.entries(tableMap)) {\n promptSchema += `Tabelle: ${table}\\n${columns.join('\\n')}\\n\\n`;\n}\n\n// Prompt als neues JSON zurückgeben\nreturn [\n {\n json: {\n promptSchema\n }\n }\n];\n"
},
"id": "ad63946b-45dd-451b-8b81-31fb29d69891",
"name": "Format Schema Text",
"type": "n8n-nodes-base.code",
"typeVersion": 1,
"position": [
1740,
120
]
},
{
"parameters": {},
"id": "88d28e22-e4d4-4096-b9ed-a8343accc1fc",
"name": "Chat Trigger",
"type": "@n8n/n8n-nodes-langchain.chatTrigger",
"typeVersion": 1,
"position": [
1200,
140
],
"webhookId": "dynamic-sql-agent"
},
{
"parameters": {
"model": "mistral:7b",
"options": {}
},
"id": "aff1c0ca-566d-4352-a917-6d8dfcf71e91",
"name": "Ollama Model",
"type": "@n8n/n8n-nodes-langchain.lmChatOllama",
"typeVersion": 1,
"position": [
1920,
440
],
"credentials": {
"ollamaApi": {
"id": "b1F3VB0Aj4eGL5Jl",
"name": "Ollama account"
}
}
},
{
"parameters": {},
"id": "1412e4a5-464a-4f9b-a524-30e6929e2fa9",
"name": "Memory",
"type": "@n8n/n8n-nodes-langchain.memoryBufferWindow",
"typeVersion": 1,
"position": [
2120,
420
]
},
{
"parameters": {
"promptType": "define",
"text": "=Hier ist das Datenbankschema der `datawarehouse`-Datenbank:\n\n{{ $json.promptSchema }}\n\nBasierend auf der folgenden Benutzerfrage:\n\"{{ $json.userMessage }}\"\n\nFinde die am besten passende Tabelle und generiere eine SQL-Abfrage, um die Frage zu beantworten.\n\n- Nutze COUNT(*) wenn nach Anzahl gefragt wird\n- Gib nur die SQL-Abfrage zurück, ohne Erklärung\n- Nutze Tabellennamen und Spaltennamen exakt wie im Schema\n\nWenn die Tabelle \"Auftraege\" erwähnt wird, ist wahrscheinlich die Tabelle `auftraege` gemeint – nutze dein bestes Sprachverständnis, um passende Tabellen zuzuordnen.\n",
"options": {}
},
"id": "c2c16866-40ef-42b8-804e-82cdd45c70a2",
"name": "AI Agent",
"type": "@n8n/n8n-nodes-langchain.agent",
"typeVersion": 1.8,
"position": [
2000,
180
],
"alwaysOutputData": true
},
{
"parameters": {
"values": {
"string": [
{
"name": "antwort",
"value": "Ergebnis: {{ $json[0] || 'Keine Daten zurückgegeben' }}"
}
]
},
"options": {}
},
"id": "0a3f0ed2-3876-448e-b5ce-2496a65075c9",
"name": "Antwort formatieren",
"type": "n8n-nodes-base.set",
"typeVersion": 2,
"position": [
2780,
180
]
},
{
"parameters": {
"operation": "executeQuery"
},
"type": "n8n-nodes-base.microsoftSql",
"typeVersion": 1.1,
"position": [
2540,
180
],
"id": "e7327c46-c182-4e63-b06e-c48e3d7fd27c",
"name": "Microsoft SQL",
"credentials": {
"microsoftSql": {
"id": "a643CpWdK4y4gPDH",
"name": "Microsoft SQL account"
}
}
},
{
"parameters": {
"mode": "raw",
"jsonOutput": "{{ $json.output }}",
"options": {}
},
"type": "n8n-nodes-base.set",
"typeVersion": 3.4,
"position": [
2340,
180
],
"id": "bbad8b5e-7351-4a59-bed8-00480bbcb3d3",
"name": "Edit Fields"
}
],
"pinData": {},
"connections": {
"Chat Trigger": {
"main": [
[
{
"node": "Get DB Schema",
"type": "main",
"index": 0
}
]
]
},
"Get DB Schema": {
"main": [
[
{
"node": "Format Schema Text",
"type": "main",
"index": 0
}
]
]
},
"Format Schema Text": {
"main": [
[
{
"node": "AI Agent",
"type": "main",
"index": 0
}
]
]
},
"Ollama Model": {
"ai_languageModel": [
[
{
"node": "AI Agent",
"type": "ai_languageModel",
"index": 0
}
]
]
},
"Memory": {
"ai_memory": [
[
{
"node": "AI Agent",
"type": "ai_memory",
"index": 0
}
]
]
},
"AI Agent": {
"main": [
[
{
"node": "Edit Fields",
"type": "main",
"index": 0
}
]
]
},
"Antwort formatieren": {
"main": [
[]
]
},
"Microsoft SQL": {
"main": [
[
{
"node": "Antwort formatieren",
"type": "main",
"index": 0
}
]
]
},
"Edit Fields": {
"main": [
[
{
"node": "Microsoft SQL",
"type": "main",
"index": 0
}
]
]
}
},
"active": false,
"settings": {
"executionOrder": "v1"
},
"versionId": "b0b06bea-4002-412b-9848-a96fe2087066",
"meta": {
"templateCredsSetupCompleted": true,
"instanceId": "ad5ba5777f8d48f5b9b814aaa0019bfd1a2b000bb69aaf3c9d3245cf58e109be"
},
"id": "jrwGIbPNHkv1DlFO",
"tags": []
}```
The Workflow is running in all Modules (Ollama, SQL..).
My problem is that my AI agent doesn't really learn. A question “How many orders are there” is implemented correctly by the part.
( SELECT COUNT(*) FROM datawarehouse.dbo.orders;)
But if I ask another more detailed query, the correct statement does not come up.
Greetings from Germany
Private & Local Ollama Self-Hosted Al Assistant: