Instrumentation Tags
Add metadata to every LLM call using simple inline comments. No SDK required.
How It Works
Add /// KEY value comments at the beginning of your prompts. Demeterics extracts these tags before forwarding to the LLM provider, giving you granular analytics without affecting model behavior.
/// APP customer-support
/// FLOW ticket-response
/// MARKET enterprise
/// VARIANT v3-empathetic
You are a helpful customer support agent.
Respond to the following ticket with empathy...
Key benefits:
- Tags are stripped before reaching the LLM (zero extra tokens)
- Works in any tool: n8n, Zapier, Make, REST, SDKs
- No code changes required—just edit your prompts
Available Tags
| Tag | Description | Example |
|---|---|---|
APP |
Application or service name | /// APP marketing-bot |
FLOW |
Business workflow or process | /// FLOW lead-qualification |
MARKET |
Geographic region or segment | /// MARKET latam-spanish |
VARIANT |
A/B test variant or experiment | /// VARIANT friendly-v2 |
VERSION |
Prompt template version | /// VERSION 2.3.1 |
UNIT |
Team, department, or division | /// UNIT sales-team |
COMPANY |
Customer or client identifier | /// COMPANY acme-corp |
SESSION |
Conversation or session ID | /// SESSION abc123 |
USER |
User identifier | /// USER user_456 |
PRODUCT |
Product line or feature | /// PRODUCT premium-tier |
PROJECT |
Project or campaign identifier | /// PROJECT q4-campaign |
ENVIRONMENT |
Deployment environment | /// ENVIRONMENT production |
PRIORITY |
Request priority level | /// PRIORITY high |
TAGS |
Comma-separated flexible labels | /// TAGS urgent,vip,escalated |
Tag Syntax
Demeterics supports multiple syntax formats for flexibility:
/// APP my-app # Space-separated (recommended)
/// APP: my-app # Colon-separated
/// APP = my-app # Equals-separated
/// APP [my-app] # Bracket notation
/// APP "my value" # Quoted values (for spaces)
All formats are equivalent. Use whichever fits your workflow.
Best Practices
1. Be Consistent
Use the same tag values across your organization. /// MARKET us-west and /// MARKET US_West will be tracked as different segments.
2. Start Simple
Begin with 2-3 tags (APP, FLOW, VARIANT) and add more as needed.
3. Use Hierarchical Names
Structure values for easier filtering:
/// APP myproduct-chatbot
/// FLOW myproduct-chatbot-onboarding
4. Version Your Prompts
Track prompt changes over time:
/// APP email-generator
/// VERSION 2.3.1
5. Tag A/B Tests
Compare variants directly in your dashboard:
/// FLOW checkout-upsell
/// VARIANT control # or "treatment-a", "treatment-b"
Example Use Cases
Multi-Tenant Cost Tracking
Track LLM costs per customer for accurate billing:
/// COMPANY acme-corp
/// UNIT engineering
/// FLOW code-review
A/B Testing Prompts
Compare conversion rates between prompt strategies:
/// FLOW onboarding-welcome
/// VARIANT friendly-casual
Geographic Analysis
Optimize prompts for different markets:
/// MARKET eu-german
/// FLOW product-recommendation
Prompt Version Control
Measure regressions after prompt updates:
/// APP support-agent
/// VERSION 3.1.0
Integration Examples
Python (OpenAI SDK)
from openai import OpenAI
client = OpenAI(
base_url="https://api.demeterics.com/v1",
default_headers={"X-Demeterics-Key": "YOUR_KEY"}
)
response = client.chat.completions.create(
model="gpt-4o",
messages=[{
"role": "user",
"content": """/// APP my-app
/// FLOW summarization
/// MARKET enterprise
Summarize the following document..."""
}]
)
n8n
In your OpenAI node, add tags to the prompt field:
/// APP n8n-workflow
/// FLOW {{ $json.workflow_name }}
{{ $json.user_prompt }}
curl
curl https://api.demeterics.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_KEY" \
-H "X-Demeterics-Key: $DEMETERICS_KEY" \
-d '{
"model": "gpt-4o",
"messages": [{
"role": "user",
"content": "/// APP my-app\n/// FLOW test\n\nHello, world!"
}]
}'
Querying Tagged Data
Once tagged, filter your dashboard by any dimension:
- By App: See all costs for
customer-supportapp - By Flow: Compare
ticket-responsevsrefund-requestflows - By Variant: A/B test results for
friendly-casualvsformal - By Company: Export monthly spend per customer for invoicing
Use the REST API to query tagged data programmatically.
FAQ
Do tags affect the LLM output?
No. Tags are stripped before the request reaches the LLM provider. The model never sees them.
Do tags add tokens to my bill?
No. Tags are removed before token counting. You only pay for your actual prompt content.
Can I add custom tags?
Use the TAGS field for flexible labeling:
/// TAGS urgent,vip,high-value
What if I forget to add tags?
Requests without tags are still tracked, but grouped under "untagged" in analytics. We recommend always adding at least APP and FLOW.
Next Steps
- Getting Started - Set up your first integration
- API Reference - Query your data programmatically
- n8n Integration - Use tags in n8n workflows