Together AI Status Page Guide: Endpoints, Components and What It Won't Tell You

Every team integrating Together AI eventually wires the status page into something — a health check, a dashboard tile, a Slack alert. Most of those integrations are subtly broken, because "the status page" is not one standard thing. Together AI runs on a Better Stack status page, and that choice decides which endpoints exist, what shape the payload is, and what the page is capable of telling you at all.

This guide is the practical version: the exact paths on status.together.ai, what each one returns today, the ones that return a success code with a useless body, and the gap between what the page reports and what your users are experiencing.

The Endpoints on status.together.ai

Verified against the live page. Note the rows marked with a warning — those are the paths that look like they work and do not, and they are the reason so many Together AI health checks report green through an incident.

How Granular Is the Component List?

Together AI runs on Better Stack rather than Atlassian Statuspage, which changes both the URL you poll and how much detail you get back. /index.json returns a JSON:API-shaped document whose data.attributes block carries company_name, timezone, announcement, and the field that actually matters: aggregate_state.

You get one rolled-up state, not a per-model breakdown. That is coarser than Groq or Cohere, so pair it with your own latency and error-rate metrics on api.together.xyz — a single degraded model will not move the aggregate.

Every Statuspage URL You Know Redirects Here

Because Together AI is not on Atlassian Statuspage, /api/v2/status.json and /api/v2/summary.json return 301 to the page root. A poller with redirect: 'follow' — which is the fetch default — will therefore receive HTTP 200 and an HTML body, and any code doing await res.json() throws a parse error that reads like a network blip rather than a wrong URL. Point at /index.json and the problem disappears.

Polling It Correctly

Roughly 40 lines, no dependencies beyond a fetch, and it fails loudly rather than silently when the page shape changes.

// Together AI runs on Better Stack, not Atlassian Statuspage.
// /api/v2/status.json 301s to the page root, so the Statuspage poller
// you use elsewhere will silently parse HTML. The real endpoint:
const TOGETHER_STATUS = 'https://status.together.ai/index.json';

async function togetherHealth() {
  const res = await fetch(TOGETHER_STATUS, {
    redirect: 'error',            // fail loudly if the URL ever moves
    signal: AbortSignal.timeout(5000),
  });
  if (!res.ok) throw new Error(`Together status returned ${res.status}`);

  const { data } = await res.json();
  const attrs = data?.attributes ?? {};

  return {
    state: attrs.aggregate_state,       // rolled up across all components
    announcement: attrs.announcement,   // maintenance banners land here
    updatedAt: attrs.updated_at,
  };
}

Two habits worth keeping regardless of provider: put a hard timeout on the request — a status page that hangs should never hang your health check — and assert on a parsed field rather than on the HTTP status code. Every trap in the table above passes an res.ok check.

Why the Status Page Is the Last to Know

A status page is a communications artifact, not a monitoring system. The sequence during a real Together AI incident is almost always the same: requests start failing, internal alerts fire, an engineer confirms scope, someone with page access writes an update, and only then does the indicator move. That chain has humans in it, and it runs in minutes rather than seconds.

The second structural problem is aggregation. A page-level indicator answers "is Together AI broadly healthy", which is not the question you have. Your question is whether your requests, to your model, from yourregion, are succeeding — and partial degradation that hits a slice of traffic routinely never moves the indicator at all.

1. Detect on your own metrics. Error rate and p99 latency against https://api.together.xyz/v1 move within seconds.
2. Attribute with the status page. Once you know something is wrong, the page tells you whose problem it is.
3. Check the component, not the headline. The indicator can read operational while the component you depend on does not.
4. Confirm with a second key. If a different key on a different account succeeds, you are looking at quota, not an outage.
5. Fail over on the second consecutive failure. One error is noise; two in a row is an incident.

For the full outage playbook see our Is Together AI Down? Outage Checking Guide, and for the error-code side of the same question, the Together AI API Error Codes Explained.

Subscribing to Together AI Incidents

Three channels, in ascending order of how quickly they reach you:

Where to Send Traffic During a Together AI Incident

Fallbacks only help if they are configured before the incident, and if the fallback does not share a status page with the thing that just broke. Each of these runs independently:

Frequently Asked Questions

Related Together AI Guides

• Is Together AI Down? Outage Checking Guide
• Together AI API Error Codes Explained
• Together AI API Status Alerts
• Together AI API Monitoring Guide
• Together AI API Outage History
• Together AI Status Guide
• Best Status Page Tools 2026
• API Monitoring Tools for Startups