Telemetry API
Telemetry reports observed reachability and inference timing, not guaranteed provider availability. Requests do not trigger paid inference probes. The worker checks health hourly at minute 00 and schedules latency probes at 06:15 and 18:15 UTC, subject to credentials, provider policy and daily limits.
Missing measurements are null, not zero. Clients should display No data (or a localized equivalent), exclude unavailable values from best-value markers, and sort unknown values last. A measured 0% uptime remains distinct from null.
See the generated OpenAPI specification for current response schemas.
Provider Health
GET /api/providers/:id/healthReturns current synthetic reachability and rolling 24-hour/7-day percentages. HTTP 401/403 can count as reachable: healthy does not guarantee model access or successful authenticated inference.
Path Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
id | string | Yes | The provider’s unique identifier (e.g. openai, linkrra) |
Query Parameters
This endpoint has no period or limit query parameters.
Example Request
curl "https://api.openmodels.run/api/providers/linkrra/health"Example Response
{
"provider_id": "linkrra",
"status": "unknown",
"uptime_24h": null,
"uptime_7d": null,
"last_checked_at": null
}Response Fields
| Field | Type | Description |
|---|---|---|
provider_id | string | The requested provider’s identifier |
status | string | healthy, unhealthy, or unknown |
uptime_24h | number or null | Percentage of successful usable checks in the last 24 hours |
uptime_7d | number or null | Percentage of successful usable checks in the last 7 days |
last_checked_at | string or null | Timestamp of the latest usable check, even if it has become stale |
The example shows the response shape without usable observations, not live Linkrra measurements. If the latest usable check is older than 90 minutes, status becomes unknown and both percentages become null. Local TLS verification failures are excluded rather than counted as outages; certificate verification stays enabled.
Error Responses
| Status | Type | Description |
|---|---|---|
404 | NOT_FOUND | Provider with the given ID does not exist |
Provider Latency
GET /api/providers/:id/latencyAggregates positive TTFT and total-time samples across a provider’s models over the last 24 hours. model_id is an empty string for this provider-wide aggregate. All numeric metrics are nullable.
Path Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
id | string | Yes | The provider’s unique identifier (e.g. linkrra, deepseek) |
Query Parameters
This endpoint has no period, limit or model_id query parameters.
Example Request
curl "https://api.openmodels.run/api/providers/linkrra/latency"Example Response
{
"provider_id": "linkrra",
"model_id": "",
"avg_ttft_ms": null,
"avg_total_time_ms": null,
"p50_ms": null,
"p95_ms": null,
"p99_ms": null
}Response Fields
| Field | Type | Description |
|---|---|---|
provider_id | string | The requested provider’s identifier |
model_id | string | Empty for this provider-wide aggregate |
avg_ttft_ms | number or null | Mean time to first token in milliseconds |
avg_total_time_ms | number or null | Mean total response time in milliseconds |
p50_ms | number or null | 50th percentile TTFT |
p95_ms | number or null | 95th percentile TTFT |
p99_ms | number or null | 99th percentile TTFT |
The example shows the shape without usable samples. Percentiles describe TTFT, not total response time. Values can be fractional; absent data must not display as 0 ms. For model-specific timing, use GET /api/models/:id/providers or GET /api/models/:id/compare, which filter by provider and model ID. Comparison rows expose nullable uptime_24h and avg_ttft_ms.
Error Responses
| Status | Type | Description |
|---|---|---|
404 | NOT_FOUND | Provider with the given ID does not exist |
Ranked Providers
GET /api/models/:id/providers/rankedReturns an array of provider mappings, each with nested health and latency, provider_name, a 1-based rank and numeric score. An existing model with no mappings returns [].
Path Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
id | string | Yes | The model’s unique identifier (e.g. qwen3-coder-30b-a3b) |
Query Parameters
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
min_uptime | number | No | none | Minimum observed 24-hour uptime percentage (0-100); unknown uptime is excluded when this filter is specified |
Example Request
curl "https://api.openmodels.run/api/models/qwen3-coder-30b-a3b/providers/ranked?min_uptime=95"Example Response
[]Response Fields
| Field | Type | Description |
|---|---|---|
[].rank | integer | Position in the ranking (1-indexed) |
[].score | number | Composite score (0-100) |
[].provider_id | string | Provider identifier |
[].provider_name | string | Display name |
[].pricing | object | Mapping prices per million tokens |
[].health | object | Provider health metrics described above |
[].latency | object | Provider-wide latency metrics described above |
The empty-array example applies when no mappings match. The score weights 24-hour uptime at 40%, average TTFT at 35%, and mean input/output price at 25%. Missing health/latency earns no credit for that factor, without converting the returned measurement to zero. This route currently uses provider-wide latency, not model-specific latency; use model comparison for that distinction. The old period and limit parameters are not implemented.
Error Responses
| Status | Type | Description |
|---|---|---|
404 | NOT_FOUND | Model with the given ID does not exist |
Provider Status History
GET /api/providers/status?days=30Returns providers, summary and generated_at, including current status, nullable rolling uptime and daily observations. days defaults to 30 and is clamped to 7-90 days. Days without checks are not fabricated as outages. TLS verification failures are excluded from the aggregates.
curl "https://api.openmodels.run/api/providers/status?days=7"Database/cache failures can produce a server error and are not measured provider outages. All routes remain subject to public API rate limiting (429).