API Request Logs

Retrieve detailed logs of every API call made across your organization, inspect provider attempts and fallbacks, download CSV exports, and monitor LLM latency and reliability metrics. While the aggregate Monitoring endpoints report bucketed spend and call totals over time, the request log endpoints provide granular per-request auditing for debugging, cost attribution, and compliance.

Authentication

Request log endpoints belong to the Management API. They require a management key (mgmt-eden-…) with the manage:read scope:

Authorization: Bearer <management_key>

Inference keys (sk-eden-…) cannot call management endpoints and return 401.

Endpoints

Endpoint	Method	Scope	Description
`/v3/manage/usage/requests/`	`GET`	`manage:read`	List organization request logs with pagination and filters.
`/v3/manage/usage/requests/{request_id}/`	`GET`	`manage:read`	Retrieve metadata and outcome for a single request.
`/v3/manage/usage/requests/{request_id}/attempts/`	`GET`	`manage:read`	List provider attempts and fallbacks for a request.
`/v3/manage/usage/requests/volume/`	`GET`	`manage:read`	Request count time series bucketed by hour or day.
`/v3/manage/usage/requests/export/`	`GET`	`manage:read`	Export matching request logs as a CSV file.
`/v3/manage/usage/facets/`	`GET`	`manage:read`	List available filter values over a date window.
`/v3/manage/usage/llm/`	`GET`	`manage:read`	Aggregate LLM latency, cache hit rate, and error metrics.
`/v3/manage/usage/llm/models/`	`GET`	`manage:read`	LLM performance and reliability metrics broken down by model.

Query Parameters and Filtering

The request list, volume, and CSV export endpoints share a common set of query filters. Unknown query parameters are rejected with 400 rather than silently ignored.

Date Window

Parameter	Type	Description
`begin`	ISO 8601 datetime	Window start, inclusive (for example, `2026-11-01T00:00:00Z`). Defaults to 7 days ago.
`end`	ISO 8601 datetime	Window end, exclusive. Maximum window length is 366 days.

Filters

Filter parameters accept multiple comma-separated values to match any of the listed options:

Parameter	Type	Description	Example
`feature`	string	Feature category	`llm`, `ocr`, `image`
`subfeature`	string	Specific subfeature	`chat`, `embeddings`
`provider`	string	AI provider involved in the call	`openai`, `anthropic`
`api_key`	string	API key ID. Use `0` for personal keys, or `0:<user_id>` for a specific member’s key.	`12`, `0`
`user`	UUID	Member ID who initiated the request	`3fa85f64-5717-4562-b3fc-2c963f66afa6`
`outcome`	string	Request outcome: `any` (default), `success`, `error`, `provider_fault`, `caller_fault`, or `rate_limit`	`error`
`streaming`	boolean	Filter by streaming requests (`true` or `false`)	`true`
`search`	string	Free-text search across errors, models, and request paths	`timeout`
`tag.<key>`	string	Filter by caller tag value. Multiple values are comma-separated.	`tag.env=production,staging`
`exclude_tag.<key>`	string	Exclude requests with the specified tag value	`exclude_tag.env=test`
`exclude_<facet>`	string	Exclude specific values for any filter facet	`exclude_provider=mock`

List Requests

GET /v3/manage/usage/requests/ returns requests ordered newest first by default:

Python

import requests

headers = {"Authorization": "Bearer YOUR_MANAGEMENT_KEY"}

response = requests.get(
    "https://api.edenai.run/v3/manage/usage/requests/",
    headers=headers,
    params={
        "limit": 20,
        "feature": "llm",
        "outcome": "error"
    }
)
response.raise_for_status()

data = response.json()
print(f"Total matching requests: {data['total_count']}")
for req in data["results"]:
    print(f"{req['created_at']} | {req['feature']}/{req['subfeature']} | status {req['status_code']} | cost: ${req['cost']:.4f}")

cURL

curl -X GET "https://api.edenai.run/v3/manage/usage/requests/?limit=20&feature=llm&outcome=error" \
  -H "Authorization: Bearer YOUR_MANAGEMENT_KEY"

Pagination

The request list uses offset pagination with standard envelope links:

limit: Number of records to return, from 1 to 200 (default 50).
offset: Starting row offset, from 0 up to 10000. Offsets past 10000 return 400 to prevent query degradation; narrow your begin and end timestamps to inspect deeper historical logs.
order: Sort direction, either newest (default) or oldest.

{
  "results": [
    {
      "request_id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
      "created_at": "2026-11-12T14:32:01.120000Z",
      "org_id": 42,
      "user_id": "9b1deb4d-3b7d-4bad-9bdd-2b0d7b3dcb6d",
      "user_email": "dev@example.com",
      "user_former": false,
      "custom_token_id": 15,
      "custom_token_name": "prod-backend",
      "custom_token_masked": "sk-eden-…8a90",
      "custom_token_revoked": false,
      "custom_token_expired": false,
      "call_type": "sync",
      "url": "/v3/llms/chat/completions",
      "method": "POST",
      "api_version": "v3",
      "feature": "llm",
      "subfeature": "chat",
      "job_id": null,
      "status_code": 200,
      "error_kind": "none",
      "error_message": "",
      "providers": ["openai"],
      "attempt_count": 1,
      "is_streaming": true,
      "is_cached": false,
      "requested_model": "openai/gpt-4o",
      "routing_strategy": "",
      "routing_mode": "",
      "cost": 0.0042,
      "latency_ns": 482000000,
      "tags": {
        "client": "acme",
        "env": "production"
      }
    }
  ],
  "limit": 20,
  "offset": 0,
  "total_count": 1,
  "next": null,
  "previous": null
}

To safeguard data privacy, stored request and response bodies are excluded from the Management API. Management keys inspect audit metadata, performance, and spending without exposing user payload content.

Inspect a Request and Provider Attempts

To inspect an individual request, pass its request_id:

cURL

curl -X GET "https://api.edenai.run/v3/manage/usage/requests/3fa85f64-5717-4562-b3fc-2c963f66afa6/" \
  -H "Authorization: Bearer YOUR_MANAGEMENT_KEY"

To see every provider call Eden AI made for that request, including automatic fallback steps:

cURL

curl -X GET "https://api.edenai.run/v3/manage/usage/requests/3fa85f64-5717-4562-b3fc-2c963f66afa6/attempts/" \
  -H "Authorization: Bearer YOUR_MANAGEMENT_KEY"

Each attempt records the provider, model, latency, outcome, and prompt and completion token counts.

When querying an individual request or its attempts, pass the created_at timestamp returned in the list (for example, ?created_at=2026-11-12T14:32:01.120000Z) to make ClickHouse partition pruning faster.

Request Volume Time Series

GET /v3/manage/usage/requests/volume/ aggregates call counts into time buckets using the same filters as the request list:

cURL

curl -X GET "https://api.edenai.run/v3/manage/usage/requests/volume/?interval=hour&feature=llm" \
  -H "Authorization: Bearer YOUR_MANAGEMENT_KEY"

Set interval to hour or day (default day).

Export Requests as CSV

GET /v3/manage/usage/requests/export/ streams matching requests as a CSV file, ordered newest first:

cURL

curl -X GET "https://api.edenai.run/v3/manage/usage/requests/export/?feature=llm" \
  -H "Authorization: Bearer YOUR_MANAGEMENT_KEY" \
  -o request_logs.csv

CSV export details:

Row limit: A single export streams up to 100,000 rows. When truncated, the response includes the header X-EdenAI-Export-Truncated: true.
Tag columns: Up to 50 distinct caller tag keys appear as individual columns prefixed with tag:<key>. Any tags beyond the first 50 are aggregated into an other_tags column.
Throttling: To protect database stability, CSV exports are capped at 20 exports per hour per management key (manage_usage_export). Ordinary API reads are not subject to this export rate limit.

GET /v3/manage/usage/facets/ returns all distinct values present in your organization’s traffic over a date window:

cURL

curl -X GET "https://api.edenai.run/v3/manage/usage/facets/?begin=2026-11-01&end=2026-11-08" \
  -H "Authorization: Bearer YOUR_MANAGEMENT_KEY"

The response includes lists of features, providers, members, API keys, and tag keys seen during that period. Pass tag_key=<key> to list the observed values for a specific tag.

LLM Performance and Reliability Metrics

Track cache efficiency, latency percentiles, and error rates across all LLM requests:

GET /v3/manage/usage/llm/: Time series of LLM metrics bucketed by interval (hour or day). Reports cache hit rate, time to first token (TTFT) p50, provider latency percentiles, fallback rates, and token mix.
GET /v3/manage/usage/llm/models/: Metrics per provider and model. Sort by sort (attempts, cost, cache_hit_rate, error_rate, ttft_p50_ns, or latency_p95_ns) and order (desc or asc).

cURL

curl -X GET "https://api.edenai.run/v3/manage/usage/llm/models/?sort=cost&order=desc&limit=10" \
  -H "Authorization: Bearer YOUR_MANAGEMENT_KEY"

Rate Limits

Operation	Limit	Scope
Request log reads	60 requests per minute	Per management key (`manage_usage`)
CSV export	20 exports per hour	Per management key (`manage_usage_export`)

Next Steps

Monitoring

Track aggregated spend and consumption time series

Usage Statements

Download monthly PDF and CSV reports of billed usage

Management API

Manage keys, members, and organization policies

V3 Documentation

Quick Start

Overview

LLMs

Expert Models

General

Organization

Data Governance

Integrations

Authentication

Endpoints

Query Parameters and Filtering

Date Window

Filters

List Requests

Inspect a Request and Provider Attempts

Request Volume Time Series

Export Requests as CSV

Filter Facets

LLM Performance and Reliability Metrics

Rate Limits

Next Steps

Monitoring

Usage Statements

Tags

Management API

​Authentication

​Endpoints

​Query Parameters and Filtering

​Date Window

​Filters

​List Requests

​Pagination

​Inspect a Request and Provider Attempts

​Request Volume Time Series

​Export Requests as CSV

​Filter Facets

​LLM Performance and Reliability Metrics

​Rate Limits

​Next Steps

Monitoring

Usage Statements

Tags

Management API

Authentication

Endpoints

Query Parameters and Filtering

Date Window

Filters

List Requests

Pagination

Inspect a Request and Provider Attempts

Request Volume Time Series

Export Requests as CSV

Filter Facets

LLM Performance and Reliability Metrics

Rate Limits

Next Steps