API Reference

Base URL: http://localhost:8080

Authentication

When API keys are configured, all /v1/* endpoints require a Bearer token:

Authorization: Bearer your-api-key-here

When no keys are configured (default), authentication is disabled. /health and /metrics are always public.

POST /v1/events

Ingest a single event. Waits for durable storage. Times out after 10 seconds. Supports Idempotency-Key header to prevent duplicate creation on retries.

Request body

Response 201 Created

{
  "id": "01J5XQKR4M2E3N8V7P6Y1WDCBA",
  "cost_nanodollars": 6250000,
  "model": "gpt-4o",
  "provider": "openai"
}

POST /v1/events/batch

Ingest up to 10,000 events. Fire-and-forget by default. Set X-Keplor-Durable: true header to await flush confirmation for each event.

Request

{"events": [{"model": "gpt-4o", "provider": "openai", ...}, ...]}

Response 201 or 207 Multi-Status

{
  "results": [
    {"id": "01J5X...", "cost_nanodollars": 100, ...},
    {"error": "validation: model must not be empty"}
  ],
  "accepted": 1,
  "rejected": 1
}

GET /v1/events

Query events with filtering and cursor-based pagination.

Query parameters

Response 200 OK

{
  "events": [{
    "id": "01J5XQKR...",
    "timestamp": 1700000000000000000,
    "model": "gpt-4o",
    "provider": "openai",
    "usage": {"input_tokens": 500, "output_tokens": 200, ...},
    "cost_nanodollars": 6250000,
    "latency_total_ms": 1200,
    "streaming": false
  }],
  "cursor": 1700000000000000000,
  "has_more": false
}

GET /v1/quota

Real-time cost and event count from the event table. At least one of user_id or api_key_id is required.

Query parameters

* At least one of user_id or api_key_id must be provided.

Response 200 OK

{
  "cost_nanodollars": 48250000,
  "event_count": 137
}

GET /v1/rollups

Pre-aggregated daily rollup rows broken down by provider and model. The background rollup task refreshes every 60 seconds.

Query parameters

Response 200 OK

{
  "rollups": [{
    "day": 1700006400,
    "user_id": "user-123",
    "api_key_id": "key-abc",
    "provider": "openai",
    "model": "gpt-4o",
    "event_count": 42,
    "error_count": 1,
    "input_tokens": 21000,
    "output_tokens": 8400,
    "cache_read_input_tokens": 5000,
    "cache_creation_input_tokens": 0,
    "cost_nanodollars": 18750000
  }],
  "has_more": false
}

GET /v1/stats

Aggregated period statistics summed from daily rollups. Optionally group by model for per-model cost breakdowns.

Query parameters

Response 200 OK

{
  "stats": [{
    "provider": "openai",
    "model": "gpt-4o",
    "event_count": 137,
    "error_count": 3,
    "input_tokens": 68500,
    "output_tokens": 27400,
    "cache_read_input_tokens": 12000,
    "cache_creation_input_tokens": 0,
    "cost_nanodollars": 48250000
  }],
  "has_more": false
}

DELETE /v1/events/:id

Delete a single event by ID.

Returns 204 No Content if deleted, 404 Not Found if the event does not exist.

DELETE /v1/events?older_than_days=N

Bulk delete events older than N days. Equivalent to keplor gc via HTTP. older_than_days must be greater than 0.

Response 200 OK

{
  "events_deleted": 1234,
  "blobs_deleted": 56
}

GET /v1/events/export

Stream all matching events as JSON Lines (application/x-ndjson). Accepts the same filter parameters as GET /v1/events but with no result-set limit. Each line is one JSON event object.

GET /health

Liveness probe. Returns 200 when healthy, 503 when degraded.

{
  "status": "ok",
  "version": "0.1.0",
  "db": "connected",
  "queue_depth": 0,
  "queue_capacity": 32768,
  "queue_utilization_pct": 0
}

GET /metrics

Prometheus text exposition format.

# TYPE keplor_events_ingested_total counter
keplor_events_ingested_total{provider="openai"} 42

Request/Response headers

Error responses

{"error": "validation: model must not be empty"}