autoclass

Bulk classify support tickets with AI. Categories are automatically generated from your data - no manual taxonomy setup required.

How it works

1. Upload CSV - Drop your file with ticket text
2. Auto-generate categories - AI analyzes a sample and suggests categories specific to your tickets
3. Review & edit - Add, remove, or rename categories before classifying
4. Classify all - Process your entire dataset with consistent categorization
5. Download results - Get your classified CSV

Key features

• Dynamic categories - Categories are generated from YOUR data, not hardcoded templates
• BYOK - Bring your own OpenAI API key. Key stays in your browser
• No file size limits - Process thousands of tickets
• Editable taxonomy - Review and customize categories before classification
• Privacy first - No data stored. No accounts. No tracking

Deploy

Or run locally:

git clone https://github.com/debgotwired/autoclass.git
cd autoclass
npm install
npm run dev

CSV format

Your CSV needs a text column. Accepts: text, ticket_text, message, content, description, body

text
"I want to cancel my order"
"Where is my package?"
"I was charged twice"

Output

text,category
"I want to cancel my order","order_cancellation"
"Where is my package?","shipping_tracking"
"I was charged twice","billing_dispute"

Categories are generated based on your specific tickets, not predetermined templates.

How category generation works

1. Samples up to 50 tickets from your dataset (evenly distributed)
2. Asks GPT-4o-mini to analyze patterns and create 5-15 categories
3. Returns categories with descriptions
4. You can edit before classifying

Cost

Uses gpt-4o-mini:

• Category generation: ~1 API call
• Classification: 1 call per ticket (batched for speed)

Typical cost: ~$0.01 per 100 tickets

Privacy

• API key stored in browser localStorage
• Requests proxied through our API route (never logged) then sent to OpenAI
• No database, no analytics, no cookies, no tracking
• Open source - audit the code

Sample data

sample_tickets_hard.csv contains 1000 challenging test tickets with typos, ambiguous cases, and multi-issue tickets.

Tech stack

• Next.js 14
• OpenAI API (gpt-4o-mini)
• TypeScript
• No database

License

MIT