View Usage

View Usage

Query daily activity and spend data from a live LiteLLM proxy.

Setup

Ask for these if not already known:

LITELLM_BASE_URL  — e.g. https://my-proxy.example.com
LITELLM_API_KEY   — proxy admin key

API reference: https://docs.litellm.ai/docs/proxy/users#get-user-spend

Ask the user

1. View by — overall / user / team / org / tag / job (default: overall)
2. Date range — default to current month if not given
3. Filter by model? (optional)
4. Job tag(s)? (optional) — for job cost attribution, ask which request

tag identifies the job, for example job:nightly-eval or job=batch-import.

Job cost attribution

LiteLLM attributes per-request costs through request tags. For LLM jobs, prefer tagging requests with a stable job label such as job: and then query tag APIs:

• Use /tag/daily/activity?tags= for daily spend, tokens, request count,

and model/provider breakdowns for one or more job tags.

• Use /global/spend/tags?tags= for a top-level spend total by tag over a

date range.

• If the user asks "which jobs cost the most?", call /global/spend/tags

without a tags filter, sort by spend descending, and present the top tags that look like job labels.

Endpoints

Overall spend (across all users)

curl -s "$BASE/user/daily/activity?start_date=YYYY-MM-DD&end_date=YYYY-MM-DD&page_size=30" \
  -H "Authorization: Bearer $KEY"

Overall request and token volume

curl -s "$BASE/global/activity?start_date=YYYY-MM-DD&end_date=YYYY-MM-DD" \
  -H "Authorization: Bearer $KEY"

By team

curl -s "$BASE/team/daily/activity?team_ids=&start_date=YYYY-MM-DD&end_date=YYYY-MM-DD" \
  -H "Authorization: Bearer $KEY"

By org

curl -s "$BASE/organization/daily/activity?organization_ids=&start_date=YYYY-MM-DD&end_date=YYYY-MM-DD" \
  -H "Authorization: Bearer $KEY"

By user

curl -s "$BASE/user/daily/activity?user_id=&start_date=YYYY-MM-DD&end_date=YYYY-MM-DD" \
  -H "Authorization: Bearer $KEY"

By tag or job

curl -s "$BASE/tag/daily/activity?tags=&start_date=YYYY-MM-DD&end_date=YYYY-MM-DD&page_size=30" \
  -H "Authorization: Bearer $KEY"

For multiple tags, pass a comma-separated list:

curl -s "$BASE/tag/daily/activity?tags=job:nightly-eval,job:batch-import&start_date=YYYY-MM-DD&end_date=YYYY-MM-DD&page_size=30" \
  -H "Authorization: Bearer $KEY"

Top tag spend

curl -s "$BASE/global/spend/tags?start_date=YYYY-MM-DD&end_date=YYYY-MM-DD" \
  -H "Authorization: Bearer $KEY"

Filter to a specific job tag:

curl -s "$BASE/global/spend/tags?tags=&start_date=YYYY-MM-DD&end_date=YYYY-MM-DD" \
  -H "Authorization: Bearer $KEY"

Response shape

{
  "results": [
    {
      "date": "2026-03-14",
      "metrics": {
        "spend": 1.23,
        "prompt_tokens": 45000,
        "completion_tokens": 12000,
        "total_tokens": 57000,
        "api_requests": 120,
        "successful_requests": 118,
        "failed_requests": 2
      },
      "breakdown": {
        "models": { "gpt-4o": { "metrics": { "spend": 1.23, ... } } }
      }
    }
  ],
  "metadata": { "page": 1, "page_size": 10, "total_count": 31 }
}

Note: top-level key is results (not data).

Summarize with python3

curl -s "$BASE/user/daily/activity?start_date=YYYY-MM-DD&end_date=YYYY-MM-DD&page_size=30" \
  -H "Authorization: Bearer $KEY" | python3 -c "
import sys, json
d = json.load(sys.stdin)
rows = d.get('results', [])
print('{:10} {:>12} {:>10}'.format('Date', 'Requests', 'Tokens', 'Spend'))
print('-' * 46)
total_spend = 0
for r in rows:
    m = r.get('metrics', {})
    print('{:10} {:>12} ${:>9.4f}'.format(
        r.get('date', ''),
        m.get('api_requests', 0),
        m.get('total_tokens', 0),
        m.get('spend', 0),
    ))
    total_spend += m.get('spend', 0)
print('-' * 46)
print('{:10} {:>12} ${:>9.4f}'.format('TOTAL', '', '', total_spend))
"

Error handling

Before processing results, check the HTTP status:

• 401/403 — invalid or expired LITELLM_API_KEY; ask the user to verify
• 404 — endpoint not available; check LiteLLM proxy version supports activity endpoints
• Empty results — no activity in the given date range; confirm dates are correct

Instructions

1. Ask for date range — default to current month.
2. Run the appropriate endpoint. For job attribution, prefer tag endpoints and

ask for the job tag if it was not provided.

1. Print a table: Date | Requests | Tokens | Spend.
2. Show totals row at the bottom.
3. Highlight any days with failed_requests > 0.
4. If metadata.total_pages > 1, offer to fetch remaining pages.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: BerriAI
• Source: BerriAI/litellm-skills
• License: MIT
• Homepage: https://github.com/BerriAI/litellm-skills

Install and usage instructions live in the source repository linked above.