AI Analysis
Scitor’s AI analysis automatically processes every inbound email and form submission on the Pro plan. It provides instant insights to help your team prioritize and organize support requests.
What gets analyzed
When an email or form submission arrives, Scitor’s AI engine produces four outputs:
Sentiment detection
Determines the emotional tone of the message:
| Label | Description | Color |
|---|---|---|
sentiment:positive |
Customer is happy or grateful | 🟢 Green |
sentiment:neutral |
Neutral or informational tone | 🔵 Light blue |
sentiment:negative |
Customer is frustrated or upset | 🔴 Red |
Category classification
Classifies the message into a support category:
| Label | Description | Color |
|---|---|---|
category:bug-report |
Bug or error report | 🔴 Dark red |
category:feature-request |
Feature or enhancement request | 🔵 Cyan |
category:question |
General question | 🟣 Purple |
category:account |
Account-related inquiry | 🟠 Peach |
category:billing |
Billing or payment issue | 🟡 Yellow |
category:other |
Doesn’t fit other categories | ⚪ Gray |
Summary
A brief 1-2 sentence summary (max 50 words) of the email content. This appears in the issue body below the email content:
🤖 AI analysis — positive · question
Summary: Customer is asking about the availability of the API
documentation and whether there are code samples for the webhook
integration.
Priority assignment
The AI assigns a priority level based on urgency, impact, and content:
| Label | Description |
|---|---|
priority:urgent |
System down, security breach, data loss |
priority:high |
Major feature broken, blocking issue |
priority:medium |
Bugs with workarounds, partial issues |
priority:low |
Questions, feature requests, general inquiries |
See Ticket Priority for full details on priority configuration and the /priority command.
How it works
Scitor uses Cloudflare Workers AI with the Llama 3.1 8B model to analyze emails. The analysis runs within Cloudflare’s infrastructure — your email content is not stored or used for model training.
Privacy
- Email content is only used for the current analysis request
- Content is truncated to 16,000 characters before analysis
- Control characters and potential prompt injections are sanitized
- No data is sent to third-party AI services
Enabling AI analysis
AI analysis requires:
- Pro plan — Available through the GitHub Marketplace
- AI enabled in configuration — Enabled by default, but can be toggled:
# .github/scitor.yaml
ai: true # or false to disable
Using AI labels for automation
Since AI analysis applies standard GitHub labels, you can use them with GitHub’s built-in features:
Filtering issues
label:sentiment:negative label:category:bug-report
GitHub Actions automation
on:
issues:
types: [labeled]
jobs:
escalate:
if: contains(github.event.label.name, 'sentiment:negative')
runs-on: ubuntu-latest
steps:
- name: Notify team
uses: slackapi/slack-github-action@v2
with:
payload: |
{"text": "⚠️ Negative sentiment detected on ${{ github.event.issue.html_url }}"}
Project board rules
Use GitHub Projects’ built-in automation to automatically move issues to specific columns based on their labels.
Tip
Combine sentiment and category labels for powerful filtering. For example, sentiment:negative category:billing surfaces frustrated customers with billing issues — a high-priority combination.