Management API

The Management API gives your organization programmatic, read-only access to its own gateway activity. It is authenticated by a dedicated management key — a separate key type that cannot run inference.

Management keys

Each organization gets one management key, issued by Consus on request (contact sales@consusindustries.co). It uses the same x-api-key header and rate limits as your other keys, but:

It is read-only: GET /v1/logs is the only endpoint it can call. Every other endpoint — including all inference endpoints (/v1/chat/completions, /v1/messages, /v1/messages/count_tokens, /v1/responses) and GET /v1/models — returns 403 management_key_not_permitted. Use your standard API keys for model requests and for listing models.
It reads org-wide: GET /v1/logs returns request logs for every API key in your organization, not just one key.
It never spends and is not subject to monthly spend caps — it keeps working even when your organization has hit its budget, so you can always see why.

Treat the key value like any credential: it is shown once at issuance and can be rotated or revoked by Consus on request.

Get request logs

GET /v1/logs Returns your organization’s per-request usage logs, newest first. Logs are metadata only. The gateway never stores prompt or response content (zero data retention), so log rows contain token counts, cost, latency, and status — never message text.

Headers

Header	Required	Description
`x-api-key`	Yes	Your organization’s management key. Standard keys receive `403`.

Query parameters

Parameter	Type	Default	Description
`start`	ISO 8601	30 days before `end`	Window start (UTC). A bare date (`2026-08-01`) means midnight UTC.
`end`	ISO 8601	now	Window end (UTC). A bare date includes that whole day. Max window: 90 days.
`status`	string	`all`	`all`, `success` (HTTP 200 only), or `errors` (non-200 only).
`api_key_id`	string	—	Scope results to one of your organization’s key IDs.
`limit`	integer	100	Rows per page, 1–1000.
`cursor`	string	—	Opaque pagination token from a previous response. Treat it as opaque — do not construct or modify it.

Response

{
  "object": "list",
  "data": [
    {
      "request_id": "9c5b2f1a-8d3e-4b6a-9f21-0e7c44aa1b55",
      "timestamp": "2026-08-04T17:22:31.481923+00:00",
      "api_key_id": "a1b2c3d4e5",
      "model": "claude-sonnet-4-5:il5",
      "compliance": "il5",
      "reasoning_effort": "high",
      "input_tokens": 1204,
      "output_tokens": 356,
      "cost": 0.00897,
      "latency_ms": 2841,
      "status": "200"
    }
  ],
  "has_more": true,
  "next_cursor": "eyJ2IjoxLCJ3bSI6..."
}

Field	Description
`request_id`	The gateway request ID (also returned to the caller as `X-Request-ID`).
`timestamp`	When the request completed (UTC).
`api_key_id`	Which of your organization’s keys made the request.
`model`	The composite model ID as requested.
`compliance`	The compliance level from the model ID (`il5`, `fedramp-high`, `il5+itar`, …); `null` when the request never resolved a model.
`reasoning_effort`	The reasoning level used, when one applied; otherwise `null`.
`input_tokens` / `output_tokens`	Billed token counts.
`cost`	Cost of the request in USD.
`latency_ms`	End-to-end gateway latency.
`status`	HTTP status returned to the caller — error rows (`"4xx"`/`"5xx"`) appear alongside successes.

Pagination

When has_more is true, pass next_cursor back as cursor to fetch the next (older) page. The cursor freezes the query window, so results are stable even while new traffic arrives — new requests appear when you start a fresh query, never in the middle of one. Keep the other parameters identical between pages; changing status or the window invalidates the cursor. Always drive the loop from has_more, not from the size of data: when a narrow filter matches nothing in a stretch of history, a page can come back empty while has_more is still true. Stop only when has_more is false.

curl "https://api.consus.io/v1/logs?limit=200" \
  -H "x-api-key: $CONSUS_MANAGEMENT_KEY"

import requests

rows, cursor = [], None
while True:
    params = {"limit": 200}
    if cursor:
        params["cursor"] = cursor
    page = requests.get(
        "https://api.consus.io/v1/logs",
        headers={"x-api-key": MANAGEMENT_KEY},
        params=params,
        timeout=30,
    ).json()
    rows.extend(page["data"])
    if not page["has_more"]:
        break
    cursor = page["next_cursor"]

What’s not in logs

A request is logged once it reaches a model-serving endpoint, so the log includes provider errors and malformed-request rejections (4xx/5xx rows) alongside successful calls. Two classes of rejection are not recorded, because they are refused before the request is ever metered:

Rejected at the edge — invalid or missing API keys, WAF blocks, and rate limiting. These are visible only in Consus’s edge audit logs.
Rejected by key policy — a key over its monthly spend cap (402), an expired trial (403), or a management key attempting inference (403). These are refused before the request is billed or recorded, so an absence of rows during an outage can itself be the signal: if a key stopped producing log entries entirely, check its spend cap and expiry.

Errors

Status	Type	Meaning
`400`	`invalid_request_error`	Malformed parameter or an invalid/expired cursor.
`403`	`permission_error`	The key is not a management key.
`404`	`not_found`	`api_key_id` is not a key in your organization.
`503`	`server_error`	Log store temporarily unreachable — retry shortly.

Product

Getting Started

API Reference

Integrations

Management keys

Get request logs

Headers

Query parameters

Response

What’s not in logs

Errors

​Management keys

​Get request logs

​Headers

​Query parameters

​Response

​Pagination

​What’s not in logs

​Errors

Management keys

Get request logs

Headers

Query parameters

Response

Pagination

What’s not in logs

Errors