{
  "name": "Spintax AI authoring funnel — brief → LLM → validate → repair → variants",
  "nodes": [
    {
      "parameters": {},
      "id": "fx-trigger",
      "name": "Start the authoring funnel",
      "type": "n8n-nodes-base.manualTrigger",
      "typeVersion": 1,
      "position": [
        0,
        400
      ]
    },
    {
      "parameters": {
        "assignments": {
          "assignments": [
            {
              "id": "fx-brief",
              "name": "brief",
              "type": "string",
              "value": "Write a short cold-outreach email (3–5 sentences) for %first_name% at %company%, pitching a service that generates unique product descriptions for e-commerce catalogs. Friendly, concrete, one clear call to action."
            }
          ]
        }
      },
      "id": "fx-brief-node",
      "name": "Your brief",
      "type": "n8n-nodes-base.set",
      "typeVersion": 3.4,
      "position": [
        200,
        400
      ]
    },
    {
      "parameters": {
        "operation": "buildAuthoringPrompt",
        "brief": "={{ $json.brief }}",
        "useItemFieldNames": false,
        "allowedVariables": {
          "variable": [
            {
              "name": "first_name",
              "case": "",
              "note": "recipient's first name"
            },
            {
              "name": "company",
              "case": "",
              "note": "recipient's company"
            }
          ]
        },
        "channel": "email",
        "variationLevel": "balanced"
      },
      "id": "fx-authoring",
      "name": "Build authoring prompt",
      "type": "n8n-nodes-spintax.spintax",
      "typeVersion": 1,
      "position": [
        400,
        400
      ]
    },
    {
      "parameters": {
        "promptType": "define",
        "text": "={{ $json.userPrompt }}",
        "messages": {
          "messageValues": [
            {
              "message": "={{ $json.systemPrompt }}"
            }
          ]
        }
      },
      "id": "fx-llm",
      "name": "Your LLM",
      "type": "@n8n/n8n-nodes-langchain.chainLlm",
      "typeVersion": 1.5,
      "position": [
        620,
        400
      ]
    },
    {
      "parameters": {
        "model": "gpt-4o-mini",
        "options": {}
      },
      "id": "fx-model",
      "name": "Chat model for authoring and repair",
      "type": "@n8n/n8n-nodes-langchain.lmChatOpenAi",
      "typeVersion": 1,
      "position": [
        640,
        620
      ]
    },
    {
      "parameters": {
        "operation": "validate",
        "template": "={{ $json.text ?? $json.template }}",
        "cleanModelOutput": true,
        "locale": "",
        "spintaxMeta": "={{ $('Build authoring prompt').item.json.spintaxMeta }}"
      },
      "id": "fx-validate1",
      "name": "Validate draft",
      "type": "n8n-nodes-spintax.spintax",
      "typeVersion": 1,
      "position": [
        860,
        400
      ]
    },
    {
      "parameters": {
        "operation": "buildRepairPrompt",
        "template": "={{ $json.cleanedTemplate ?? $json.template }}",
        "diagnostics": "={{ $json.diagnostics }}",
        "spintaxMeta": "={{ $('Build authoring prompt').item.json.spintaxMeta }}"
      },
      "id": "fx-repair",
      "name": "Build repair prompt",
      "type": "n8n-nodes-spintax.spintax",
      "typeVersion": 1,
      "position": [
        1080,
        560
      ]
    },
    {
      "parameters": {
        "promptType": "define",
        "text": "={{ $json.userPrompt }}",
        "messages": {
          "messageValues": [
            {
              "message": "={{ $json.systemPrompt }}"
            }
          ]
        }
      },
      "id": "fx-llm2",
      "name": "LLM (repair)",
      "type": "@n8n/n8n-nodes-langchain.chainLlm",
      "typeVersion": 1.5,
      "position": [
        1300,
        560
      ]
    },
    {
      "parameters": {
        "operation": "validate",
        "template": "={{ $json.text ?? $json.template }}",
        "cleanModelOutput": true,
        "locale": "",
        "spintaxMeta": "={{ $('Build authoring prompt').item.json.spintaxMeta }}"
      },
      "id": "fx-validate2",
      "name": "Validate (attempt 2)",
      "type": "n8n-nodes-spintax.spintax",
      "typeVersion": 1,
      "position": [
        1520,
        560
      ]
    },
    {
      "parameters": {},
      "id": "fx-giveup",
      "name": "Still invalid — stop (loop is capped)",
      "type": "n8n-nodes-base.noOp",
      "typeVersion": 1,
      "position": [
        1740,
        700
      ]
    },
    {
      "parameters": {
        "operation": "renderMany",
        "template": "={{ $json.cleanedTemplate }}",
        "count": 5,
        "baseSeed": "demo",
        "useIncomingItem": false,
        "fixedVariables": {
          "pair": [
            {
              "name": "first_name",
              "value": "Ada",
              "neutralize": false
            },
            {
              "name": "company",
              "value": "Brightline Gear",
              "neutralize": false
            }
          ]
        }
      },
      "id": "fx-render",
      "name": "Render 5 variants",
      "type": "n8n-nodes-spintax.spintax",
      "typeVersion": 1,
      "position": [
        1740,
        400
      ]
    },
    {
      "parameters": {
        "content": "## Spintax AI authoring funnel — brief → LLM → validate → repair → variants\n\n### How it works\n\n1. Your brief becomes an authoring prompt, and your LLM writes ONE spintax template from it — the model runs once, not once per variant.\n2. The draft is validated before anything is rendered, so a broken template never reaches your output.\n3. An invalid draft gets exactly one repair round, aimed at the line, column and code of each diagnostic.\n4. A valid template renders five distinct variants and reports how many it actually produced.\n\n### Setup steps\n\n- [ ] Install the community node `n8n-nodes-spintax`.\n- [ ] Add credentials to **Chat model for authoring and repair** — one node feeds both LLM steps — or swap in your own.\n- [ ] Put your brief, locale and allowed variables in **Your brief**.\n- [ ] Run it once — a valid draft goes straight to variants.\n\n### Customization\n\nRaise the variant count on **Render 5 variants**; `produced` tells you when a template runs out of distinct combinations. Add a second repair round by copying the three repair nodes — the cap is the wiring, so it is yours to change.",
        "width": 480,
        "height": 784
      },
      "type": "n8n-nodes-base.stickyNote",
      "typeVersion": 1,
      "position": [
        -608,
        128
      ],
      "id": "c3d4e5f6-0000-4a11-9c01-5f1a2b3c4d00",
      "name": "Sticky Note"
    },
    {
      "parameters": {
        "content": "## Start and set your brief\n\nOne Set node holds the brief, the locale and the allowed variable names. They travel the whole funnel as `spintaxMeta`.",
        "width": 384,
        "height": 320,
        "color": 7
      },
      "type": "n8n-nodes-base.stickyNote",
      "typeVersion": 1,
      "position": [
        -48,
        240
      ],
      "id": "c3d4e5f6-0001-4a11-9c01-5f1a2b3c4d01",
      "name": "Sticky Note1"
    },
    {
      "parameters": {
        "content": "## Build the authoring prompt\n\nThe node emits a system/user prompt pair for your model — any model, OpenAI is only the one wired here. `Clean Model Output` strips the code fences models add whatever the prompt says.",
        "width": 432,
        "height": 560,
        "color": 7
      },
      "type": "n8n-nodes-base.stickyNote",
      "typeVersion": 1,
      "position": [
        352,
        224
      ],
      "id": "c3d4e5f6-0002-4a11-9c01-5f1a2b3c4d02",
      "name": "Sticky Note2"
    },
    {
      "parameters": {
        "content": "## Validate, then repair once\n\nValid means no error-level diagnostics; a valid draft goes straight right. The repair prompt carries the line, column and code of each problem, so the model is pointed at the offending token rather than asked to guess.",
        "width": 848,
        "height": 480,
        "color": 7
      },
      "type": "n8n-nodes-base.stickyNote",
      "typeVersion": 1,
      "position": [
        816,
        240
      ],
      "id": "c3d4e5f6-0003-4a11-9c01-5f1a2b3c4d03",
      "name": "Sticky Note3"
    },
    {
      "parameters": {
        "content": "## Where the funnel ends\n\nA good template renders five distinct variants and reports `produced` against `requested`. A draft that survives two rounds stops here — the loop is capped by wiring, not by a counter.",
        "width": 240,
        "height": 736,
        "color": 7
      },
      "type": "n8n-nodes-base.stickyNote",
      "typeVersion": 1,
      "position": [
        1696,
        128
      ],
      "id": "c3d4e5f6-0005-4a11-9c01-5f1a2b3c4d05",
      "name": "Sticky Note4"
    }
  ],
  "connections": {
    "Your brief": {
      "main": [
        [
          {
            "node": "Build authoring prompt",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Build authoring prompt": {
      "main": [
        [
          {
            "node": "Your LLM",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Your LLM": {
      "main": [
        [
          {
            "node": "Validate draft",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Validate draft": {
      "main": [
        [
          {
            "node": "Render 5 variants",
            "type": "main",
            "index": 0
          }
        ],
        [
          {
            "node": "Build repair prompt",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Build repair prompt": {
      "main": [
        [
          {
            "node": "LLM (repair)",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "LLM (repair)": {
      "main": [
        [
          {
            "node": "Validate (attempt 2)",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Validate (attempt 2)": {
      "main": [
        [
          {
            "node": "Render 5 variants",
            "type": "main",
            "index": 0
          }
        ],
        [
          {
            "node": "Still invalid — stop (loop is capped)",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Start the authoring funnel": {
      "main": [
        [
          {
            "node": "Your brief",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Chat model for authoring and repair": {
      "ai_languageModel": [
        [
          {
            "node": "Your LLM",
            "type": "ai_languageModel",
            "index": 0
          },
          {
            "node": "LLM (repair)",
            "type": "ai_languageModel",
            "index": 0
          }
        ]
      ]
    }
  },
  "settings": {},
  "pinData": {}
}