Observed incident recordDetected by RELIASTRA probes

Official status page failure window - Together AI

Incident cdd238b9-ba08-4bf1-badd-ca8c13ce6c52 - endpoint status.together.ai, measured from us-east. Confirming rule: 2 consecutive failed observations counted 2 of them.

ResolvedSeverity: MajorStatus: ResolvedAttribution: OBSERVED by RELIASTRA

Observed window

Every timestamp is a stored field of this record, printed in UTC. The window begins at the first failure the rule counted and ends, for a resolved record, at the observation run whose consecutive successes closed it.

Opened: 22 Oct 2026 20:50:50 UTC
Confirmed: 22 Oct 2026 20:55:39 UTC - the run that made the failure count reach the threshold; this record did not exist before it
Recovered: 22 Oct 2026 21:03:25 UTC
Duration: 12m (from the stored window)

What was measured

The record is a summary of an observation run set. The referenced observation ids resolve to stored probe entries inside the retention window; the detector keeps no second, narrative version of them.

Endpoint observed: Official status page - https://status.together.ai
Measured from: us-east - the only observation region named in this record; nothing about other regions is claimed or inferred
Failed observations: 2, consecutive through the trigger set
Failure shape: probe timeouts
Observation references: first ec274c26-3c3d-4ff3-9a40-c465d61f5ec9 last 0df9c33e-72f9-4852-a009-87cd8163d1fd

The claim, in full

Between 2026-10-22T20:50:50.075727+00:00 and 2026-10-22T20:55:39.704727+00:00, RELIASTRA probes in us-east recorded 2 consecutive failed observations against https://status.together.ai (the observed Together AI target). Failure class: timeout; observed status: no HTTP response. This record describes the observed endpoint only and is not a statement about the vendor's services as a whole. That is all of it. Nothing about the vendor as a whole, nothing about other regions, nothing about your requests - the record you are reading is the measurement, not a narration of it.

Evidence artifact

A frozen, hashed document of this record's window: the claim, the detection rule, and every observation row it rests on. Deterministic, reproducible from stored data, immutable once written.

Artifact: /api/v1/public/incidents/cdd238b9-ba08-4bf1-badd-ca8c13ce6c52/evidence - JSON, served byte-for-byte as stored
Content hash: sha256: 73d0c8143c0304bd89bf101ca1f039da63fb498da86382beefc17356bf48c62f
Freeze: version 2, taken while the incident was resolved (22 Oct 2026 21:08:19 UTC); later freezes supersede, never edit, and each stays at its own versioned URL
Coverage: 5 observations in the window
Verification: Remove the "verification" key from the document, re-serialise with sorted keys and compact separators, and SHA-256 the UTF-8 bytes: the digest must equal the hash above. The recipe ships inside every artifact.

What this record does not establish

The boundaries are part of the record, not a disclaimer under it.

Causation: Not established. A 5xx can be the vendor, a shared middle path, or the probe's own egress - the record reports what came back, not why.
Vendor acknowledgement: Together AI may have recorded this window differently, later, or not at all. RELIASTRA does not ingest vendor-reported status, so no reconciliation is offered either way.
Global reach: This window is named for us-east. A record from one observation region never settles the question in another.
Customer impact: Whether your requests were affected is your measurement to make - this record carries only RELIASTRA's own probes.

Where this fits

Together AI - current recordLive observatory
All observed incidents - togetherIncident search
Cross-vendor observed incident searchIncident search
How RELIASTRA measures vendor reliabilityMethodology