Usage & analytics

See what you spend, the mix of request types your traffic generates, and which application made each call — with no client setup beyond an optional header.

Usage & spend

The Overview summarizes requests, input and output tokens, and spend over a time window, each with a trend against the prior period. When caching is active it adds a savings strip and a cache timeline.

Analytics by request type

DI Analytics breaks your traffic down by the request type each call was classified as — the same vocabulary returned per request in the X-DI-Request-Type header. The model behind each type stays private; the type itself is yours to analyze.

vision document long code json reason flash pro

Per request vs. in aggregate

Read X-DI-Request-Type on a single response (see Response headers) to know how one call was handled; use DI Analytics to see the distribution across all of them. Full definitions are in Request types.

Traces

Traces lists individual requests — timestamp, endpoint, the model id you sent, request type, tokens, cost, cache breakdown, and latency — and each row expands to the raw request and response. Consistent with the rest of the platform, a trace shows your call and how it was classified, never the backend model that served it.

Per-application attribution

Usage is automatically grouped by application so one key can power several apps and still report separately. The label comes from X-Title if you send it, otherwise the request’s Referer host (browser and web apps send this automatically), falling back to the API key’s name — so attribution works even with zero setup.

Python

client = OpenAI(
    api_key="llm_live_...",
    base_url="https://api.directinference.com/di/v1",
    default_headers={"X-Title": "billing-worker"},   # names this app in analytics
)

TypeScript

const client = new OpenAI({
  apiKey: "llm_live_...",
  baseURL: "https://api.directinference.com/di/v1",
  defaultHeaders: { "X-Title": "billing-worker" }, // names this app in analytics
});

curl

curl https://api.directinference.com/di/v1/chat/completions \
  -H "Authorization: Bearer llm_live_..." \
  -H "X-Title: billing-worker" \
  -H "Content-Type: application/json" \
  -d '{ "model": "gpt-5.5-mini", "messages": [{ "role": "user", "content": "..." }] }'

HTTPie

echo '{ "model": "gpt-5.5-mini", "messages": [{ "role": "user", "content": "..." }] }' \
  | https POST api.directinference.com/di/v1/chat/completions \
      Authorization:'Bearer llm_live_...' \
      X-Title:billing-worker \
      Content-Type:application/json

Name your apps to split a shared key

Set a stable X-Title per service (billing-worker, support-bot) and each shows up as its own line in Applications, deep-linkable to a scoped usage view.

REST API

The same usage and per-request cost the dashboard shows are also available programmatically under your normal llm_live_... key — no portal session required. Both endpoints are read-only, scoped to your account, and exist on the root and …/di surfaces. For your balance, see Spend & limits.

Aggregate usage — GET /v1/usage

Spend and token counts bucketed by day (default), request_type, application, or api_key, optionally bounded by start_date / end_date (each YYYY-MM-DD or RFC 3339; a date-only end_date is inclusive). cost is your charge in USD.

GET /v1/usage

curl "https://api.directinference.com/v1/usage?group_by=day&start_date=2026-06-01" \
  -H "Authorization: Bearer llm_live_..."

200

{
  "object": "usage",
  "group_by": "day",
  "start_date": "2026-06-01T00:00:00Z",
  "data": [
    {
      "bucket": "2026-06-01",
      "requests": 1234,
      "input_tokens": 90000,
      "output_tokens": 30000,
      "total_tokens": 120000,
      "cached_input_tokens": 4096,
      "cost": "1.840000"
    }
  ]
}

The request_type axis uses the same vocabulary as X-DI-Request-Type; the backend model is never part of a bucket. An unknown group_by returns 400.

Per-request cost — GET /v1/generation?id=

Look one call up by the id it returned in its response body — useful for attributing cost to a specific request after the fact.

GET /v1/generation

curl "https://api.directinference.com/v1/generation?id=chatcmpl-..." \
  -H "Authorization: Bearer llm_live_..."

200

{
  "id": "chatcmpl-...",
  "request_type": "code",
  "created_at": "2026-06-14T22:55:00Z",
  "streamed": false,
  "finish_reason": "stop",
  "total_cost": "0.001230",
  "usage": {
    "prompt_tokens": 100,
    "completion_tokens": 50,
    "total_tokens": 150,
    "cached_prompt_tokens": 0,
    "reasoning_tokens": 0
  },
  "latency_ms": 200,
  "generation_time_ms": 1500
}

A missing id returns 400; an unknown or another account’s id returns 404. As everywhere, total_cost is your charge — no provider, model, or upstream cost is included.