Turn Any PDF into a Clean Google Doc with Mistral OCR

{
  "id": "evPxNVkJ3vVScGsS",
  "meta": {
    "instanceId": "36fee986cc83112881fb12ec7cc2d0221d7bddd71c11715c196899b114e8b0d2",
    "templateCredsSetupCompleted": true
  },
  "name": "Turn Any PDF into a Clean Google Doc with Mistral OCR",
  "tags": [],
  "nodes": [
    {
      "id": "f5199976-f999-4552-9943-3a1a25d34eeb",
      "name": "On form submission",
      "type": "n8n-nodes-base.formTrigger",
      "position": [
        -1440,
        80
      ],
      "webhookId": "81573452-f871-4eb1-b13e-60eabde2e012",
      "parameters": {
        "options": {},
        "formTitle": "Document Scanner",
        "formFields": {
          "values": [
            {
              "fieldType": "file",
              "fieldLabel": "file",
              "requiredField": true,
              "acceptFileTypes": ".pdf"
            },
            {
              "fieldLabel": "Document Name",
              "placeholder": "NVIDIA Annual Report Doc",
              "requiredField": true
            }
          ]
        },
        "formDescription": "Insert the google drive url that stores your image PDFs"
      },
      "typeVersion": 2.2
    },
    {
      "id": "863d40f4-01a5-4989-99a5-ba2ba2e2ab92",
      "name": "Sticky Note",
      "type": "n8n-nodes-base.stickyNote",
      "position": [
        -2016,
        -144
      ],
      "parameters": {
        "width": 416,
        "height": 464,
        "content": "## Start here (required settings)\n• Upload your PDF and enter a document name in the form trigger.\n• In **Create • Google Doc**, paste the Google Drive folder ID.\n• Make sure your Mistral and Google Docs credentials are active.\n\n## What this workflow does\n• Uploads the PDF to Mistral for OCR.\n• Re-OCRs every inline image to catch small text.\n• Merges page and image text into one markdown file.\n• Cleans the markdown, stripping headers and noise.\n• Creates a Google Doc and inserts the clean text.\n\n## Customize\n• Edit the cleanup prompt for stricter or looser filtering.\n• Pipe the clean text to Gmail or Slack instead of Docs if preferred."
      },
      "typeVersion": 1
    },
    {
      "id": "b0f99739-aed6-45bc-a0f0-818a6a5523a0",
      "name": "Upload • PDF to Mistral",
      "type": "n8n-nodes-base.httpRequest",
      "position": [
        -1216,
        80
      ],
      "parameters": {
        "url": "https://api.mistral.ai/v1/files",
        "method": "POST",
        "options": {},
        "sendBody": true,
        "contentType": "multipart-form-data",
        "authentication": "predefinedCredentialType",
        "bodyParameters": {
          "parameters": [
            {
              "name": "purpose",
              "value": "ocr"
            },
            {
              "name": "file",
              "parameterType": "formBinaryData",
              "inputDataFieldName": "file"
            }
          ]
        },
        "nodeCredentialType": "mistralCloudApi"
      },
      "credentials": {
        "mistralCloudApi": {
          "id": "Xmws9BgP1rH4wzpK",
          "name": "Mistral Cloud account"
        }
      },
      "typeVersion": 4.2
    },
    {
      "id": "b508b895-52d6-4060-b409-97b6bb18aee0",
      "name": "Fetch • Signed URL",
      "type": "n8n-nodes-base.httpRequest",
      "position": [
        -992,
        80
      ],
      "parameters": {
        "url": "=https://api.mistral.ai/v1/files/{{ $json.id }}/url",
        "options": {},
        "sendQuery": true,
        "sendHeaders": true,
        "authentication": "predefinedCredentialType",
        "queryParameters": {
          "parameters": [
            {
              "name": "expiry",
              "value": "24"
            }
          ]
        },
        "headerParameters": {
          "parameters": [
            {
              "name": "Accept",
              "value": "application/json"
            }
          ]
        },
        "nodeCredentialType": "mistralCloudApi"
      },
      "credentials": {
        "mistralCloudApi": {
          "id": "Xmws9BgP1rH4wzpK",
          "name": "Mistral Cloud account"
        }
      },
      "typeVersion": 4.2
    },
    {
      "id": "ca574ffa-2e18-49b3-a4d5-a456ce3cc2bf",
      "name": "OCR • PDF via Mistral",
      "type": "n8n-nodes-base.httpRequest",
      "position": [
        -768,
        80
      ],
      "parameters": {
        "url": "https://api.mistral.ai/v1/ocr",
        "method": "POST",
        "options": {},
        "jsonBody": "={\n  \"model\": \"mistral-ocr-latest\",\n  \"document\": {\n    \"type\": \"document_url\",\n    \"document_url\": \"{{ $json.url }}\"\n  },\n  \"include_image_base64\": true\n}",
        "sendBody": true,
        "sendHeaders": true,
        "specifyBody": "json",
        "authentication": "predefinedCredentialType",
        "headerParameters": {
          "parameters": [
            {}
          ]
        },
        "nodeCredentialType": "mistralCloudApi"
      },
      "credentials": {
        "mistralCloudApi": {
          "id": "Xmws9BgP1rH4wzpK",
          "name": "Mistral Cloud account"
        }
      },
      "typeVersion": 4.2
    },
    {
      "id": "af8dcda4-7928-401f-acab-a58091564c2b",
      "name": "Extract • Image Placeholders",
      "type": "n8n-nodes-base.code",
      "position": [
        -544,
        16
      ],
      "parameters": {
        "jsCode": "// Input: one item with ocrResponse at items[0].json\nconst out = [];\nconst res = items[0].json;\n\nfor (const page of res.pages || []) {\n  const md = page.markdown || \"\";\n  const imgs = page.images || [];\n  // Build a quick lookup by id: \"img-0.jpeg\" -> image object\n  const byId = Object.fromEntries(imgs.map(im => [im.id, im]));\n  // Find placeholders in markdown\n  const re = /!\\[(.*?)\\]\\((img-\\d+\\.(?:jpe?g|png|webp|avif))\\)/g;\n  let m;\n  while ((m = re.exec(md)) !== null) {\n    const id = m[2];\n    const im = byId[id];\n    if (!im || !im.image_base64) continue;\n    out.push({\n      json: {\n        pageIndex: page.index,\n        imageId: id,\n        // This is already a full data URI like \"data:image/jpeg;base64,...\"\n        imageDataUri: im.image_base64\n      }\n    });\n  }\n}\nreturn out.length ? out : [{ json: { note: \"no-embedded-images\" } }];\n"
      },
      "typeVersion": 2
    },
    {
      "id": "7fdaedc6-bf1b-46cd-831f-db3298b96f15",
      "name": "OCR • Inline Images",
      "type": "n8n-nodes-base.httpRequest",
      "position": [
        -320,
        -64
      ],
      "parameters": {
        "url": "https://api.mistral.ai/v1/ocr",
        "method": "POST",
        "options": {
          "response": {}
        },
        "jsonBody": "={\n  \"model\": \"mistral-ocr-latest\",\n  \"document\": {\n    \"type\": \"image_url\",\n    \"image_url\": \"{{ $json.imageDataUri }}\"\n  },\n  \"include_image_base64\": true\n} ",
        "sendBody": true,
        "specifyBody": "json",
        "authentication": "predefinedCredentialType",
        "nodeCredentialType": "mistralCloudApi"
      },
      "credentials": {
        "mistralCloudApi": {
          "id": "Xmws9BgP1rH4wzpK",
          "name": "Mistral Cloud account"
        }
      },
      "typeVersion": 4.2
    },
    {
      "id": "8f7fa9a4-a328-4e02-bc99-01132c180873",
      "name": "Merge • OCR + Image Data",
      "type": "n8n-nodes-base.merge",
      "position": [
        -96,
        16
      ],
      "parameters": {
        "mode": "combine",
        "options": {},
        "combineBy": "combineByPosition"
      },
      "typeVersion": 3.2
    },
    {
      "id": "5e0e12d2-0985-48e3-89e6-e62824d7edb6",
      "name": "Set • Rename Columns",
      "type": "n8n-nodes-base.set",
      "position": [
        128,
        16
      ],
      "parameters": {
        "options": {},
        "assignments": {
          "assignments": [
            {
              "id": "558c0d83-7dfe-412f-927f-82c37551b910",
              "name": "pageIndex",
              "type": "number",
              "value": "={{ $json.pageIndex }}"
            },
            {
              "id": "a7df3b1e-a851-4b83-a811-bd58a2ad3253",
              "name": "imageId",
              "type": "string",
              "value": "={{ $json.imageId }}"
            },
            {
              "id": "2cba94bc-7d32-4098-ac1a-e1ee9d27e587",
              "name": "imageDataUri",
              "type": "string",
              "value": "={{ $json.imageDataUri }}"
            },
            {
              "id": "69b97aab-6822-4d88-ab57-75b4a5aa861e",
              "name": "imagePages",
              "type": "array",
              "value": "={{ $json.pages }}"
            }
          ]
        }
      },
      "typeVersion": 3.4
    },
    {
      "id": "1b6c4a20-276c-44b1-8dfe-bb9f9e5df91d",
      "name": "Aggregate • Image Text",
      "type": "n8n-nodes-base.aggregate",
      "position": [
        352,
        16
      ],
      "parameters": {
        "options": {},
        "fieldsToAggregate": {
          "fieldToAggregate": [
            {
              "fieldToAggregate": "imagePages[0].markdown"
            }
          ]
        }
      },
      "typeVersion": 1
    },
    {
      "id": "03f08eba-80a5-429c-8584-cf9f40dd4e7f",
      "name": "Merge • PDF & Images Markdown",
      "type": "n8n-nodes-base.merge",
      "position": [
        576,
        80
      ],
      "parameters": {
        "mode": "combine",
        "options": {},
        "combineBy": "combineByPosition"
      },
      "typeVersion": 3.2
    },
    {
      "id": "7e02c3cc-68e0-4bda-ad41-8a67b2aee2a5",
      "name": "Enrich • Replace Placeholders",
      "type": "n8n-nodes-base.code",
      "position": [
        800,
        80
      ],
      "parameters": {
        "jsCode": "// n8n Code node: merge main page markdown with OCR text for each img-N / imageN\nconst rows = $input.all().map(i => i.json);\n\n// ───────────────────────── config ─────────────────────────\nconst MERGE_MODE = 'replace'; // 'replace' or 'append'\nconst EXTS = '(?:jpeg|jpg|png|gif|webp|avif)';\n\n// ───────────────────────── helpers ────────────────────────\nconst asString = v => (v == null ? '' : String(v));\nconst isStringArray = a => Array.isArray(a) && a.every(x => typeof x === 'string');\n\n// Strip any image tokens from OCR text so we do not re-insert placeholders\nconst stripImageTokens = s =>\n  asString(s)\n    .replace(/!\\[[^\\]]*]\\([^)]+\\)/gi, '')          // remove markdown image tags\n    .replace(/\\b(?:img[-_]?|image[-_]?)\\d+\\.(?:jpeg|jpg|png|gif|webp|avif)\\b/gi, '')\n    .trim();\n\n// Try to locate the \"main\" document row (has page-level markdown)\nconst main =\n  rows.find(r => typeof r['pages[0].markdown'] === 'string' && r['pages[0].markdown'].length) ||\n  rows.find(r => Array.isArray(r.pages) && r.pages.length && typeof r.pages[0]?.markdown === 'string') ||\n  rows.find(r => typeof r.documentMarkdown === 'string' && r.documentMarkdown.length);\n\nif (!main) {\n  return [{ json: { error: 'No main document row with pages[].markdown or pages[0].markdown found', rows } }];\n}\n\n// Build the starting document markdown\nlet doc = main['pages[0].markdown'] ||\n          (Array.isArray(main.pages) ? main.pages.map(p => p?.markdown || '').join('\\n\\n') : '') ||\n          main.documentMarkdown ||\n          '';\n\n// ────────────────── build idx → text map ──────────────────\nconst idxToText = new Map();\n\n// 1) Per-image rows: derive index from imageId like \"img-12.jpg\" or \"image12.jpg\"\nfor (const r of rows) {\n  if (r === main) continue;\n\n  let imageId = r.imageId || (Array.isArray(r.images) && r.images[0]?.id) || '';\n  imageId = asString(imageId);\n\n  // Accept both \"img-12.jpg\" and \"image12.jpg\"\n  let m = imageId.match(/(?:img[-_]?|image[-_]?)(\\d+)\\./i);\n  if (!m) continue;\n  const idx = Number(m[1]);\n\n  // Prefer r.markdown[0], then r['markdown[0]'], then r.imagePages[0].markdown, then r.markdown (string)\n  let text = null;\n  if (Array.isArray(r.markdown) && r.markdown.length) text = r.markdown[0];\n  else if (typeof r['markdown[0]'] === 'string') text = r['markdown[0]'];\n  else if (Array.isArray(r.imagePages) && r.imagePages.length) text = r.imagePages[0]?.markdown || '';\n  else if (typeof r.markdown === 'string') text = r.markdown;\n\n  if (text != null && !idxToText.has(idx)) {\n    idxToText.set(idx, stripImageTokens(text));\n  }\n}\n\n// 2) Aggregated array case: r['imagePages[0].markdown'] is an array OR r.markdown is an array (your screenshot)\nfor (const r of rows) {\n  const agg =\n    (Array.isArray(r['imagePages[0].markdown']) && r['imagePages[0].markdown']) ||\n    (isStringArray(r.markdown) && r.markdown);\n\n  if (!agg) continue;\n\n  agg.forEach((t, i) => {\n    if (t == null) return;\n    if (!idxToText.has(i)) idxToText.set(i, stripImageTokens(t));\n  });\n}\n\n// If we still have no mapping, bail with debug\nif (idxToText.size === 0) {\n  return [{\n    json: {\n      ...main,\n      error: 'No OCR mapping built. Neither per-image rows nor aggregated markdown[] found.',\n      debug_keys: rows.map(r => Object.keys(r)),\n      sample_row: rows[0],\n    }\n  }];\n}\n\n// ────────────────── replace or append ─────────────────────\nlet placeholdersFound = 0;\nlet placeholdersReplaced = 0;\nconst missing = [];\n\nconst sorted = [...idxToText.entries()].sort((a, b) => a[0] - b[0]);\nfor (const [idx, textRaw] of sorted) {\n  const text = asString(textRaw).trim();\n  const nameGroup = `(?:img[-_]?${idx}|image[-_]?${idx})`;\n\n  const patterns = [\n    // markdown image: ![alt](img-0.jpg) or ![alt](image0.jpg)\n    new RegExp(`!\\\\[[^\\\\]]*\\\\]\\\\(\\\\s*${nameGroup}\\\\.${EXTS}\\\\s*\\\\)`, 'gi'),\n    // link style: [img-0.jpg](img-0.jpg)\n    new RegExp(`\\\\[\\\\s*${nameGroup}\\\\.${EXTS}\\\\s*\\\\]\\\\(\\\\s*${nameGroup}\\\\.${EXTS}\\\\s*\\\\)`, 'gi'),\n    // bare token somewhere in text\n    new RegExp(`\\\\b${nameGroup}\\\\.${EXTS}\\\\b`, 'gi'),\n  ];\n\n  let localFound = 0;\n  for (const rx of patterns) {\n    const matches = doc.match(rx);\n    if (!matches) continue;\n    localFound += matches.length;\n\n    if (MERGE_MODE === 'replace') {\n      doc = doc.replace(rx, () => text);\n    } else {\n      // append under the image/link\n      doc = doc.replace(rx, (m) => `${m}\\n\\n${text}\\n`);\n    }\n  }\n\n  placeholdersFound += localFound;\n  placeholdersReplaced += localFound;\n  if (localFound === 0) missing.push({ idx, note: 'no placeholder found for this index' });\n}\n\n// ───────────────────────── output ─────────────────────────\nreturn [{\n  json: {\n    ...main,\n    markdown: doc,\n    enrichment_debug: {\n      placeholdersFound,\n      placeholdersReplaced,\n      missing,\n      mappedCount: idxToText.size,\n    },\n  },\n}];\n"
      },
      "typeVersion": 2
    },
    {
      "id": "d8b064fb-ede2-47ea-b1a0-d964f04e6ab1",
      "name": "Clean • Markdown Text",
      "type": "@n8n/n8n-nodes-langchain.chainLlm",
      "position": [
        1024,
        80
      ],
      "parameters": {
        "text": "=You are a careful assistant trained to clean raw text extracted from any file type (docs, PDFs, web pages, OCR, spreadsheets, slides, logs).\n\nThe input may contain:\n\n* Placeholder or boilerplate text (“Lorem ipsum…”, template notes)\n* Broken or noisy formatting (excess asterisks, stray symbols, HTML/XML tags, Markdown artifacts)\n* OCR or export debris (headers, footers, page numbers, hyphenated line breaks)\n* Gibberish or data dumps (meaningless numbers, repeated years, fake tables)\n* Duplicates or irrelevant lines\n\nYour task:\n\n* **Remove** placeholders, boilerplate, headers/footers, page numbers, repeated dates or numbers with no context, broken tables, tracking strings, and formatting noise.\n* **Normalize** spacing, line breaks, and list structure. Fix obviously broken sentences or words split by line-wrap.\n* **Preserve** substantive content: real sentences, section headings, lists, quotes, citations, code blocks that appear intentional, figures or formulas that are meaningful, and the original tone.\n* **Do not rewrite or summarize** valid content. Only repair text that is clearly malformed.\n* Keep the **original language** of the input.\n* Return the result as clean **markdown-style text**.\n\nInput:\n\n```\n{{ $json[\"markdown\"] }}\n```\n\nOutput:\n\n* Only the cleaned, concise markdown text. No extra commentary.\n",
        "batching": {},
        "promptType": "define"
      },
      "typeVersion": 1.7
    },
    {
      "id": "8e461ca0-b9a0-4158-9b56-51885ee183c8",
      "name": "Create • Google Doc",
      "type": "n8n-nodes-base.googleDocs",
      "position": [
        1424,
        80
      ],
      "parameters": {
        "title": "={{ $('On form submission').item.json['Document Name'] }}",
        "folderId": "16ooKdLhm5GzzKSSZl1wMGTHJOfJ3-r13"
      },
      "credentials": {
        "googleDocsOAuth2Api": {
          "id": "FAyP7uzcbNGZpjFI",
          "name": "Google Docs account"
        }
      },
      "typeVersion": 2
    },
    {
      "id": "46219fff-1f07-4ac1-896e-81fccf5821ce",
      "name": "Insert • Clean Text into Doc",
      "type": "n8n-nodes-base.googleDocs",
      "position": [
        1648,
        80
      ],
      "parameters": {
        "actionsUi": {
          "actionFields": [
            {
              "text": "={{ $('Clean • Markdown Text').item.json.text }}",
              "action": "insert"
            }
          ]
        },
        "operation": "update",
        "documentURL": "={{ $json.id }}"
      },
      "credentials": {
        "googleDocsOAuth2Api": {
          "id": "FAyP7uzcbNGZpjFI",
          "name": "Google Docs account"
        }
      },
      "typeVersion": 2
    },
    {
      "id": "0a8010b7-c2bd-41bf-93f5-d39221721836",
      "name": "LLM Model • GPT-4.1-mini",
      "type": "@n8n/n8n-nodes-langchain.lmChatOpenRouter",
      "position": [
        1120,
        304
      ],
      "parameters": {
        "options": {}
      },
      "credentials": {
        "openRouterApi": {
          "id": "9H6tnWvfT2T0cL3J",
          "name": "OpenRouter account (Augra)"
        }
      },
      "typeVersion": 1
    },
    {
      "id": "f9deb8d7-51c4-495d-a895-d89821a96af5",
      "name": "Sticky Note8",
      "type": "n8n-nodes-base.stickyNote",
      "position": [
        -2016,
        336
      ],
      "parameters": {
        "color": 5,
        "width": 419,
        "height": 422,
        "content": "## Start here: Step-by Step Youtube Tutorial :star:\n\n[![Building an AI Personal Assistant](https://img.youtube.com/vi/VjIoAqvRPGs/sddefault.jpg)](https://youtu.be/VjIoAqvRPGs)\n"
      },
      "typeVersion": 1
    }
  ],
  "active": false,
  "pinData": {},
  "settings": {
    "executionOrder": "v1"
  },
  "versionId": "3cba0804-300f-4b73-b62b-1da49d68031b",
  "connections": {
    "On form submission": {
      "main": [
        [
          {
            "node": "Upload • PDF to Mistral",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Fetch • Signed URL": {
      "main": [
        [
          {
            "node": "OCR • PDF via Mistral",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Create • Google Doc": {
      "main": [
        [
          {
            "node": "Insert • Clean Text into Doc",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "OCR • Inline Images": {
      "main": [
        [
          {
            "node": "Merge • OCR + Image Data",
            "type": "main",
            "index": 1
          }
        ]
      ]
    },
    "Set • Rename Columns": {
      "main": [
        [
          {
            "node": "Aggregate • Image Text",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Clean • Markdown Text": {
      "main": [
        [
          {
            "node": "Create • Google Doc",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "OCR • PDF via Mistral": {
      "main": [
        [
          {
            "node": "Extract • Image Placeholders",
            "type": "main",
            "index": 0
          },
          {
            "node": "Merge • PDF & Images Markdown",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Aggregate • Image Text": {
      "main": [
        [
          {
            "node": "Merge • PDF & Images Markdown",
            "type": "main",
            "index": 1
          }
        ]
      ]
    },
    "Upload • PDF to Mistral": {
      "main": [
        [
          {
            "node": "Fetch • Signed URL",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "LLM Model • GPT-4.1-mini": {
      "ai_languageModel": [
        [
          {
            "node": "Clean • Markdown Text",
            "type": "ai_languageModel",
            "index": 0
          }
        ]
      ]
    },
    "Merge • OCR + Image Data": {
      "main": [
        [
          {
            "node": "Set • Rename Columns",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Extract • Image Placeholders": {
      "main": [
        [
          {
            "node": "OCR • Inline Images",
            "type": "main",
            "index": 0
          },
          {
            "node": "Merge • OCR + Image Data",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Enrich • Replace Placeholders": {
      "main": [
        [
          {
            "node": "Clean • Markdown Text",
            "type": "main",
            "index": 0
          }
        ]
      ]
    },
    "Merge • PDF & Images Markdown": {
      "main": [
        [
          {
            "node": "Enrich • Replace Placeholders",
            "type": "main",
            "index": 0
          }
        ]
      ]
    }
  }
}