Retrieval latency
Measured p50 and p95 for the two retrieval endpoints over the last 30 days, from production telemetry. The method is below the numbers, and the same numbers are available as JSON at https://api.mindgraph.cloud
/v1/status/latency. Last computed Thu, 22 Oct 2026 21:53:37 GMT.
POST /retrieve/context
616 ms p50 · 1444 ms p95 · 356 requests over 4 days (sample-weighted mean of daily percentiles)
Daily table
| Day (UTC) | Requests | p50 | p95 |
|---|---|---|---|
| 2026-10-22 | 7 | insufficient | insufficient |
| 2026-10-21 | 7 | insufficient | insufficient |
| 2026-10-20 | 21 | insufficient | insufficient |
| 2026-10-19 | 12 | insufficient | insufficient |
| 2026-10-18 | 5 | insufficient | insufficient |
| 2026-10-17 | 2 | insufficient | insufficient |
| 2026-10-16 | 12 | insufficient | insufficient |
| 2026-10-15 | 9 | insufficient | insufficient |
| 2026-10-14 | 4 | insufficient | insufficient |
| 2026-10-13 | 12 | insufficient | insufficient |
| 2026-10-12 | 1 | insufficient | insufficient |
| 2026-10-10 | 5 | insufficient | insufficient |
| 2026-10-09 | 4 | insufficient | insufficient |
| 2026-10-08 | 9 | insufficient | insufficient |
| 2026-10-07 | 26 | insufficient | insufficient |
| 2026-10-06 | 21 | insufficient | insufficient |
| 2026-10-05 | 36 | insufficient | insufficient |
| 2026-10-04 | 12 | insufficient | insufficient |
| 2026-10-03 | 43 | insufficient | insufficient |
| 2026-10-02 | 27 | insufficient | insufficient |
| 2026-10-01 | 19 | insufficient | insufficient |
| 2026-09-30 | 33 | insufficient | insufficient |
| 2026-09-29 | 39 | insufficient | insufficient |
| 2026-09-28 | 32 | insufficient | insufficient |
| 2026-09-27 | 63 | 557 ms | 1395 ms |
| 2026-09-26 | 65 | 654 ms | 1325 ms |
| 2026-09-25 | 91 | 622 ms | 1540 ms |
| 2026-09-24 | 137 | 621 ms | 1460 ms |
| 2026-09-23 | 36 | insufficient | insufficient |
POST /retrieve
Not enough samples yet in this window.
Daily table
| Day (UTC) | Requests | p50 | p95 |
|---|---|---|---|
| 2026-10-02 | 36 | insufficient | insufficient |
Method
What a number on this page means. Read this before comparing it with anyone else's.
- Source
- Every successful (2xx) POST /retrieve and POST /retrieve/context on api.mindgraph.cloud, all tenants, production only.
- Excluded
- Structured /retrieve actions (active_goals, open_questions, weak_claims, unresolved_contradictions, recent), ontology queries, errors and timeouts — tails are therefore understated.
- Timer
- Starts inside the cloud proxy after the request body is buffered and stops after the response body is read: includes graph execution and response serialization; excludes TLS, network, authentication, graph loading and usage middleware.
- Aggregation
- percentile_cont(0.5) and percentile_cont(0.95) per UTC day per endpoint.
- Minimum samples
- Days with fewer than 50 requests are shown as insufficient and excluded from the headline.
- Headline
- The per-endpoint headline is a sample-weighted mean of the daily percentiles, not a percentile of the pooled population; daily values are the primary numbers.
- Refresh
- Recomputed at most once an hour per server instance.
- Privacy
- No per-tenant rows; the aggregate is grouped by day and endpoint only.
Want the request-level shape? Every retrieval response also carries per-phase timings when you pass explain: true; see the ingestion and retrieval docs.