Tags

Tags are arbitrary key-value metadata attached to any LLM call. They appear in your Weflayr dashboard and let you slice cost, latency, and error analytics by any dimension you care about.

How to pass tags

Add a tags keyword argument to any instrumented call. The SDK strips it before forwarding to the upstream provider — no change to your prompt or request payload.

client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[...],
    tags={
        "feature": "support-bot",
        "customer_id": "cust_abc123",
        "experiment": "prompt-v3",
    },
)

Common patterns

By feature

feature: onboarding feature: support-bot feature: search feature: summarization

By environment

env: production env: staging env: development

By customer or tenant

customer_id: cust_123 tenant: acme-corp plan: enterprise

By experiment or version

experiment: prompt-v3 model_variant: gpt-4o ab_group: B

By pipeline step

pipeline: rag step: retrieval step: generation

Rules

• Keys and values must be strings
• Any number of tags per call
• Tags are stored as-is — they are not forwarded to the LLM provider
• Omitting tags entirely is equivalent to tags={}