How do I monitor spend by app or project for LLM usage?

To monitor LLM spend by app or project, the best approach is to combine request-level metadata with usage/cost aggregation.

1) Tag every request with app/project identifiers

When you send a request to the model API, include metadata such as:

• app_name
• project_name
• environment (prod/staging/dev)
• team
• user_id or tenant_id if relevant

If your provider supports custom metadata fields or request tags, use those. If not, add the tags in your own application logs.

2) Log usage per request

For each LLM call, capture:

• timestamp
• model name
• input token count
• output token count
• total token count
• request ID
• app/project tags

Example log record:

{
  "timestamp": "2026-10-10T12:34:56Z",
  "app_name": "support-bot",
  "project_name": "customer-care",
  "model": "gpt-4.1",
  "input_tokens": 1200,
  "output_tokens": 300,
  "request_id": "req_123",
  "environment": "prod"
}

3) Convert tokens into cost

Multiply tokens by the model’s pricing:

• input token rate
• output token rate

Then aggregate by:

• app
• project
• team
• environment
• day/week/month

4) Use dashboards and cost reports

Push these logs into a monitoring/analytics tool such as:

• Datadog
• Grafana / Prometheus
• BigQuery / Snowflake
• Elasticsearch / OpenSearch
• OpenTelemetry-based observability pipelines

Build charts for:

• spend by app
• spend by project
• spend by model
• token usage trend
• average cost per request
• top expensive endpoints

5) Set budgets and alerts

Create alerts for:

• daily spend over threshold
• unexpected spikes in token usage
• high-cost model usage in non-prod
• unusually long prompts or responses

6) If your provider offers billing tags or organization usage reports

Some LLM providers expose usage data grouped by API key, project, or org. If available, use:

• project-scoped API keys
• separate keys per app/environment
• billing exports
• usage APIs
• cost allocation tags

This makes reporting much easier.

Recommended setup

If you want something practical and reliable:

1. Assign a unique API key per app/project/environment
2. Log every request with app/project tags
3. Store token counts and model name
4. Join logs with pricing data
5. Build a daily cost dashboard
6. Add alerts for unusual spend

If you want, I can also give you:

• a sample database schema for tracking LLM spend,
• a Python/Node middleware example that logs cost per request,
• or a dashboard design for app/project spend monitoring.