{ "edges": [ { "to": "6496bb73-0007-4b60-84cc-7e1268e85146", "from": "99aea42e-4e2f-476c-96b0-de382b83922e", "toPort": "0", "fromPort": "0" }, { "to": "9e25bad8-5f56-49b8-aa69-42fa0faaf489", "from": "ab0ce8ce-5e7c-4ea1-884c-04e29e4085c0", "toPort": "0", "fromPort": "ai_languageModel" }, { "to": "f0ff7c52-ab08-4179-a606-7c26dac2be67", "from": "9e25bad8-5f56-49b8-aa69-42fa0faaf489", "toPort": "0", "fromPort": "0" }, { "to": "1c626123-f1d6-45fd-8519-ac5342b4b38c", "from": "36935d91-24c5-4338-b766-5f58d092053e", "toPort": "0", "fromPort": "0" }, { "to": "36935d91-24c5-4338-b766-5f58d092053e", "from": "f0ff7c52-ab08-4179-a606-7c26dac2be67", "toPort": "0", "fromPort": "0" }, { "to": "99aea42e-4e2f-476c-96b0-de382b83922e", "from": "1c626123-f1d6-45fd-8519-ac5342b4b38c", "toPort": "0", "fromPort": "0" }, { "to": "9e25bad8-5f56-49b8-aa69-42fa0faaf489", "from": "6ee8c2c2-ae3e-4b3e-9c3c-64c22b75f575", "toPort": "0", "fromPort": "0" } ], "nodes": [ { "id": "1c626123-f1d6-45fd-8519-ac5342b4b38c", "name": "Retrieve an existing dartboard", "type": "integration", "params": { "id": "K7jRC0JC2Wxz", "service": "dart", "resource": "Dartboard", "operation": "Get Dartboard", "__originalType": "n8n-nodes-dart.dart", "requestOptions": {} }, "position": [ 880, -96 ] }, { "id": "99aea42e-4e2f-476c-96b0-de382b83922e", "name": "Create a new task", "type": "integration", "params": { "item": "={\n \"title\": \"{{ $('Launch workflow on email receive').item.json.subject }}\",\n \"priority\": \"{{ $('Parsing stage for logic result').item.json.priority }}\",\n \"tags\": [\"{{ $('Parsing stage for logic result').item.json.category }}\"],\n\n \"description\": \"Rationale: {{ $('Parsing stage for logic result').item.json.rationale }}\\n\\n Confidence lvl: {{ $('Parsing stage for logic result').item.json.confidence }}\\n\\n Summary: {{ $('Parsing stage for logic result').item.json.summary }}\\n \\n Full email: \\n \\n {{ $('Cleaning up full text of email').item.json.cleanText }}\"\n}\n", "service": "dart", "resource": "Task", "operation": "Create Task", "__originalType": "n8n-nodes-dart.dart", "requestOptions": {} }, "position": [ 1312, 32 ] }, { "id": "6496bb73-0007-4b60-84cc-7e1268e85146", "name": "Create a new comment", "type": "integration", "params": { "item": "={ \"taskId\": \"{{ $json.item.id }}\",\n \"text\":\"Sender: {{ $('Launch workflow on email receive').item.json.from.value[0].name }} - {{ $('Launch workflow on email receive').item.json.from.value[0].address }}\"\n}", "service": "dart", "resource": "Comment", "operation": "Add Task Comment", "__originalType": "n8n-nodes-dart.dart", "requestOptions": {} }, "position": [ 1536, 32 ] }, { "id": "36935d91-24c5-4338-b766-5f58d092053e", "name": "Cleaning up full text of email", "type": "code", "params": { "jsCode": "// Safely get the Gmail HTML body\nconst html = $('Launch workflow on email receive').first().json.textAsHtml || \"\";\n\n// --- CLEAN TEXT FROM HTML ---\n\n// Step 1: Normalize line breaks for

,
, and

tags\nlet clean = html\n .replace(/<(?:br|div)\\s*\\/?>/gi, '\\n')\n .replace(/<\\/p>/gi, '\\n');\n\n// Step 2: Remove anchor tags but keep inner text\nclean = clean.replace(/]*>(.*?)<\\/a>/gi, '$1');\n\n// Step 3: Remove scripts, styles, and comments\nclean = clean\n .replace(//gi, '')\n .replace(//gi, '')\n .replace(//g, '');\n\n// Step 4: Remove tags and Gmail-style [image: ...] placeholders\nclean = clean\n .replace(/]*>/gi, '')\n .replace(/\\[image:\\s*.*?\\]/gi, '');\n\n// Step 5: Strip remaining HTML tags\nclean = clean.replace(/<[^>]+>/g, ' ');\n\n// Step 6: Remove plain text URLs\nclean = clean.replace(/\\b(?:https?:\\/\\/|ftp:\\/\\/|www\\.|mailto:)\\S+\\b/gi, '');\n\n// Step 7: Decode basic HTML entities\nclean = clean\n .replace(/ /gi, ' ')\n .replace(/&/gi, '&')\n .replace(/</gi, '<')\n .replace(/>/gi, '>')\n .replace(/"/gi, '\"')\n .replace(/'/gi, \"'\");\n\n// Step 8: Clean up whitespace\nclean = clean\n .replace(/\\r/g, '')\n .replace(/[ \\t]+\\n/g, '\\n')\n .replace(/\\n{3,}/g, '\\n\\n')\n .replace(/\\s+/g, ' ')\n .trim();\n\n// Step 9: Remove potentially problematic symbols and quotes\nclean = clean\n .replace(/['\"`]/g, '') // remove single, double, and back quotes\n .replace(/[<>\\\\]/g, '') // remove angle brackets and backslashes\n .replace(/[{}[\\]]/g, '') // remove braces and brackets\n .replace(/[^\\x20-\\x7E\\n]/g, ''); // remove non-printable characters (safe ASCII only)\n\n// --- RETURN CLEAN TEXT AS OUTPUT ---\n\nreturn [\n {\n json: {\n cleanText: clean,\n },\n },\n];\n" }, "position": [ 464, -80 ] }, { "id": "9e25bad8-5f56-49b8-aa69-42fa0faaf489", "name": "Email triage assistant logic", "type": "llm", "params": { "text": "=[System instruction]\nYou are a concise email triage assistant whose sole job is to read a user email and pick the best single category from {Billing, Feature Request, Support Help Request, App Feedback, Bug Report, Task Updates, Unrelated}. Use the definitions and output schema provided below. Output only valid JSON that matches the schema exactly.\n\nDefinitions\n- Billing: Questions or disputes about charges, invoices, subscriptions, refunds, payment failures, billing address, tax, pricing plan changes, or account credits.\n- Feature Request: Suggestions for new features, changes to product behavior, wishlist items, or requests to add/change functionality.\n- Support Help Request: User reports of problems, errors, account access issues, setup/installation help, crashes, or anything asking for a fix or step-by-step assistance.\n- App Feedback: General feedback about experience, UI/UX suggestions not asking for a new feature, praise, complaints about design/tone/performance that do not request a fix or new functionality.\n- Bug Report: Reports describing reproducible software defects, unexpected behavior, incorrect outputs, or developer-facing error traces that require debugging and may need code fixes.\n- Task Updates: Notifications or status updates related to tasks, assignments, or workflow progress—especially from Dart or similar systems.\n- Unrelated: Emails that do not fit any of the above categories, such as promotions, marketing material, job applications, newsletters, or personal messages.\n\nRequired JSON output schema (must be followed exactly)\n{\n \"category\": \"\",\n \"confidence\": ,\n \"rationale\": \"\",\n \"summary\": \"\",\n \"priority\": \"\"\n}\n\n\n\nExample 1\nEmail: \"I was charged twice this month for my subscription. Please refund the duplicate charge and confirm when it’s processed.\"\nCategory: Billing\nConfidence: 0.96\nRationale: \"Mentions duplicate charge and explicitly requests a refund.\"\nSummary: \"User reports duplicate charge and requests refund confirmation.\"\nPriority: high\n\nExample 2\nEmail: \"It would be great if we could export project timelines as CSV and include custom fields.\"\nCategory: Feature Request\nConfidence: 0.92\nRationale: \"User asks for new export feature and specific fields to include.\"\nSummary: \"Request to add CSV export for timelines with custom fields.\"\nPriority: low\n\nExample 3\nEmail: \"App crashes every time I click Save on long tasks; I need help getting my data back.\"\nCategory: Support Help Request\nConfidence: 0.98\nRationale: \"Reports a reproducible crash and asks for help retrieving data.\"\nSummary: \"App crashes on Save for long tasks; user needs data recovery help.\"\nPriority: high\n\nExample 4\nEmail: \"Love the new color theme but the contrast on small text feels low; overall experience is much improved.\"\nCategory: App Feedback\nConfidence: 0.85\nRationale: \"Gives UI feedback and praise without requesting changes or fixes.\"\nSummary: \"Positive feedback on theme with a note about low contrast on small text.\"\nPriority: low\n\nExample 5\nEmail: \"When running the Code node, I get a stack trace: Error: Code doesn't return items properly at validateRunCodeAllItems ... Please advise.\"\nCategory: Bug Report\nConfidence: 0.95\nRationale: \"Provides a developer-facing stack trace and describes a reproducible failure in node execution.\"\nSummary: \"Code node throws validateRunCodeAllItems error with full stack trace.\"\nPriority: high\n\nExample 6\nEmail: \"Task #3421 has been marked complete. Please review the final notes and close the loop.\"\nCategory: Task Updates\nConfidence: 0.93\nRationale: \"Mentions a specific task ID and status update, typical of automated workflow notifications.\"\nSummary: \"Task #3421 marked complete; user asked to review and finalize.\"\nPriority: medium\n\nExample 7\nEmail: \"We’re hiring! Check out our open roles and apply today to join our team.\"\nCategory: Unrelated\nConfidence: 0.90\nRationale: \"Promotional job application email not related to product, support, or user feedback.\"\nSummary: \"Marketing email promoting open job roles.\"\nPriority: low\n\nInstructions\n- Read the email and map it to exactly one category.\n\n- If the email contains both billing and a technical problem, prioritize Billing if a payment dispute or refund is explicitly requested; otherwise prioritize Support Help Request.\n\n- Confidence Calibration\n0.95–1.00: Explicit intent (e.g., “refund,” “add feature,” “crash”).\n0.85–0.94: Strongly implied intent with supporting context.\n0.75–0.84: Somewhat implied or ambiguous.\n<0.75: Weak or uncertain match.\n\n- Priority Rules\ncritical – Use only when the issue completely prevents the user (or all users) from accessing core features, causes total data loss, or exposes a severe security or privacy risk. Typically urgent and business-blocking.\nExamples – “The app won’t load at all for any of our users.” / “All my project data disappeared after the update.” / “Sensitive user info is visible to others.”\n\nhigh – The issue directly affects core functionality, payments, or productivity but does not fully block all access. Includes reproducible crashes, failed payments, or broken workflows with financial or time impact.\nExamples – “I was charged twice this month for my subscription.” / “The app crashes whenever I click Save.” / “Checkout form gives an error and I can’t complete my purchase.”\n\nmedium – The issue or update affects secondary features, task progress, or non-critical operations. The user can still perform main tasks but with inconvenience or delay. Also applies to workflow notifications and status updates.\nExamples – “Task #2341 marked complete — please review.” / “The report export takes longer than usual.” / “Notifications stopped appearing, but the app still works.”\n\nlow – Minor or non-urgent matters that don’t affect functionality or access. Includes suggestions, praise, design opinions, and promotional or unrelated content.\nExamples – “Love the new theme, but text contrast is low.” / “It would be great if you could add dark mode.” / “We’re hiring! Apply today.”\n\n- Output only the JSON object and nothing else, matching the schema exactly.\n\nEmail: {{ $json.text }}", "prompt": "=[System instruction]\nYou are a concise email triage assistant whose sole job is to read a user email and pick the best single category from {Billing, Feature Request, Support Help Request, App Feedback, Bug Report, Task Updates, Unrelated}. Use the definitions and output schema provided below. Output only valid JSON that matches the schema exactly.\n\nDefinitions\n- Billing: Questions or disputes about charges, invoices, subscriptions, refunds, payment failures, billing address, tax, pricing plan changes, or account credits.\n- Feature Request: Suggestions for new features, changes to product behavior, wishlist items, or requests to add/change functionality.\n- Support Help Request: User reports of problems, errors, account access issues, setup/installation help, crashes, or anything asking for a fix or step-by-step assistance.\n- App Feedback: General feedback about experience, UI/UX suggestions not asking for a new feature, praise, complaints about design/tone/performance that do not request a fix or new functionality.\n- Bug Report: Reports describing reproducible software defects, unexpected behavior, incorrect outputs, or developer-facing error traces that require debugging and may need code fixes.\n- Task Updates: Notifications or status updates related to tasks, assignments, or workflow progress—especially from Dart or similar systems.\n- Unrelated: Emails that do not fit any of the above categories, such as promotions, marketing material, job applications, newsletters, or personal messages.\n\nRequired JSON output schema (must be followed exactly)\n{\n \"category\": \"\",\n \"confidence\": ,\n \"rationale\": \"\",\n \"summary\": \"\",\n \"priority\": \"\"\n}\n\n\n\nExample 1\nEmail: \"I was charged twice this month for my subscription. Please refund the duplicate charge and confirm when it’s processed.\"\nCategory: Billing\nConfidence: 0.96\nRationale: \"Mentions duplicate charge and explicitly requests a refund.\"\nSummary: \"User reports duplicate charge and requests refund confirmation.\"\nPriority: high\n\nExample 2\nEmail: \"It would be great if we could export project timelines as CSV and include custom fields.\"\nCategory: Feature Request\nConfidence: 0.92\nRationale: \"User asks for new export feature and specific fields to include.\"\nSummary: \"Request to add CSV export for timelines with custom fields.\"\nPriority: low\n\nExample 3\nEmail: \"App crashes every time I click Save on long tasks; I need help getting my data back.\"\nCategory: Support Help Request\nConfidence: 0.98\nRationale: \"Reports a reproducible crash and asks for help retrieving data.\"\nSummary: \"App crashes on Save for long tasks; user needs data recovery help.\"\nPriority: high\n\nExample 4\nEmail: \"Love the new color theme but the contrast on small text feels low; overall experience is much improved.\"\nCategory: App Feedback\nConfidence: 0.85\nRationale: \"Gives UI feedback and praise without requesting changes or fixes.\"\nSummary: \"Positive feedback on theme with a note about low contrast on small text.\"\nPriority: low\n\nExample 5\nEmail: \"When running the Code node, I get a stack trace: Error: Code doesn't return items properly at validateRunCodeAllItems ... Please advise.\"\nCategory: Bug Report\nConfidence: 0.95\nRationale: \"Provides a developer-facing stack trace and describes a reproducible failure in node execution.\"\nSummary: \"Code node throws validateRunCodeAllItems error with full stack trace.\"\nPriority: high\n\nExample 6\nEmail: \"Task #3421 has been marked complete. Please review the final notes and close the loop.\"\nCategory: Task Updates\nConfidence: 0.93\nRationale: \"Mentions a specific task ID and status update, typical of automated workflow notifications.\"\nSummary: \"Task #3421 marked complete; user asked to review and finalize.\"\nPriority: medium\n\nExample 7\nEmail: \"We’re hiring! Check out our open roles and apply today to join our team.\"\nCategory: Unrelated\nConfidence: 0.90\nRationale: \"Promotional job application email not related to product, support, or user feedback.\"\nSummary: \"Marketing email promoting open job roles.\"\nPriority: low\n\nInstructions\n- Read the email and map it to exactly one category.\n\n- If the email contains both billing and a technical problem, prioritize Billing if a payment dispute or refund is explicitly requested; otherwise prioritize Support Help Request.\n\n- Confidence Calibration\n0.95–1.00: Explicit intent (e.g., “refund,” “add feature,” “crash”).\n0.85–0.94: Strongly implied intent with supporting context.\n0.75–0.84: Somewhat implied or ambiguous.\n<0.75: Weak or uncertain match.\n\n- Priority Rules\ncritical – Use only when the issue completely prevents the user (or all users) from accessing core features, causes total data loss, or exposes a severe security or privacy risk. Typically urgent and business-blocking.\nExamples – “The app won’t load at all for any of our users.” / “All my project data disappeared after the update.” / “Sensitive user info is visible to others.”\n\nhigh – The issue directly affects core functionality, payments, or productivity but does not fully block all access. Includes reproducible crashes, failed payments, or broken workflows with financial or time impact.\nExamples – “I was charged twice this month for my subscription.” / “The app crashes whenever I click Save.” / “Checkout form gives an error and I can’t complete my purchase.”\n\nmedium – The issue or update affects secondary features, task progress, or non-critical operations. The user can still perform main tasks but with inconvenience or delay. Also applies to workflow notifications and status updates.\nExamples – “Task #2341 marked complete — please review.” / “The report export takes longer than usual.” / “Notifications stopped appearing, but the app still works.”\n\nlow – Minor or non-urgent matters that don’t affect functionality or access. Includes suggestions, praise, design opinions, and promotional or unrelated content.\nExamples – “Love the new theme, but text contrast is low.” / “It would be great if you could add dark mode.” / “We’re hiring! Apply today.”\n\n- Output only the JSON object and nothing else, matching the schema exactly.\n\nEmail: {{ $json.text }}", "options": {}, "provider": "openai", "promptType": "define", "approximated": true }, "position": [ -352, 96 ] }, { "id": "f0ff7c52-ab08-4179-a606-7c26dac2be67", "name": "Parsing stage for logic result", "type": "code", "params": { "jsCode": "// Get raw input\nlet rawString = $input.first().json.output;\n\n// Clean up markdown formatting if present\nrawString = rawString\n .trim()\n .replace(/^```(json)?/i, '') // remove leading ``` or ```json\n .replace(/```$/i, '') // remove trailing ```\n .trim();\n\n// Try parsing the cleaned string\nlet parsed;\ntry {\n parsed = JSON.parse(rawString);\n} catch (error) {\n throw new Error(`Failed to parse JSON. Cleaned input was:\\n${rawString}\\n\\nError: ${error.message}`);\n}\n\nreturn parsed;" }, "position": [ 80, -80 ] }, { "id": "6ee8c2c2-ae3e-4b3e-9c3c-64c22b75f575", "name": "Launch workflow on email receive", "type": "trigger", "params": { "kind": "n8n-nodes-base.gmailTrigger", "simple": false, "filters": {}, "options": {}, "pollTimes": { "item": [ { "mode": "everyMinute" } ] } }, "position": [ -528, 96 ] }, { "id": "ab0ce8ce-5e7c-4ea1-884c-04e29e4085c0", "name": "OpenAI Chat Model", "type": "llm", "params": { "model": "gpt-4.1-mini", "options": {}, "provider": "openai" }, "position": [ -384, 272 ] } ] }

Categorize support emails with AI and create tasks in Dart