Support Assistant

The Support Assistant lets your code ask Zernio support a question and get the answer back, without opening a chat. Send a question about a post, a connected account or a profile in your workspace: Ana, the Zernio support agent, looks into it and answers, usually within one to three minutes. When the problem is on Zernio's side, she hands it to the Zernio team and a teammate follows up by email.

What you can use it for

• Explain a failed post. When your backend receives post.failed, ask why with the postId and show your user the answer instead of a raw platform error.
• Diagnose a connected account. Ask why an account stopped publishing or needs reconnecting, with its accountId.
• Check a profile's setup. Ask about the accounts and settings under a profileId.
• Add a help button to your product. Forward your users' questions with the ids involved and show the answer in your own UI.
• Give your AI agent a support line. Let an agent or automation ask when it hits something it cannot explain.

First question

You need an API key with full access, a usage-based plan and a card on file (payment method). This asks why a post failed and waits for the answer. In production, receive the answer by webhook instead of polling.

import Zernio from '@zernio/node';

const zernio = new Zernio();

const { data: run } = await zernio.supportassistant.askSupportQuestion({
  headers: { 'Idempotency-Key': 'why-failed-64f0a1b2c3d4e5f6a7b8c9d0' },
  body: {
    message: 'Why did my Instagram post fail to publish?',
    context: { postId: '64f0a1b2c3d4e5f6a7b8c9d0' },
  },
});

let { data: result } = await zernio.supportassistant.getSupportAnswer({ path: { runId: run.runId } });
while (result.status === 'queued' || result.status === 'running') {
  await new Promise((resolve) => setTimeout(resolve, (result.pollAfterSeconds ?? 5) * 1000));
  ({ data: result } = await zernio.supportassistant.getSupportAnswer({ path: { runId: run.runId } }));
}
console.log(result.status, result.answer);

The first call answers 202 right away with the ids to follow:

{ "runId": "64f0a1b2c3d4e5f6a7b8c9d1", "threadId": "64f0a1b2c3d4e5f6a7b8c9d2", "status": "queued", "pollAfterSeconds": 5 }

A finished run carries the answer and what it cost:

{
  "runId": "64f0a1b2c3d4e5f6a7b8c9d1",
  "threadId": "64f0a1b2c3d4e5f6a7b8c9d2",
  "status": "completed",
  "stopReason": "answered",
  "answer": "Instagram could not download the image: its URL answered 403 to Meta's servers. Host it on a public URL and retry the post.",
  "usage": { "inputTokens": 12, "outputTokens": 3111, "cacheReadTokens": 82793, "cacheWriteTokens": 24625 },
  "costUsd": 0.327766,
  "maxCostUsd": 3,
  "createdAt": "2026-10-08T13:20:12.477Z",
  "startedAt": "2026-10-08T13:20:13.102Z",
  "finishedAt": "2026-10-08T13:21:27.514Z",
  "supportTicket": null
}

Pricing

You pay per question, once it finishes. There is no subscription and no fee for reading answers. What a question costs depends on how much looking into it takes, and the exact amount charged is the run's costUsd: the question above cost $0.33.

• maxCostUsd caps each question. Near the cap the run stops investigating and answers with what it found so far (stopReason: cost_cap), and it never charges more than the cap. If it has nothing to answer yet, it ends as failed and costs nothing. Lower it for simple questions.
• The monthly limit counts questions still running at their cap. A question is accepted only if its maxCostUsd fits in what is left of the $100 for the month; otherwise it gets 429 support_monthly_cap_exceeded.
• Charges appear as "Support API" on your invoice, billed with the rest of your usage (how billing works).

Integrate it

An integration that never loses an answer and never pays twice for one question takes five steps.

1. Subscribe a webhook

Subscribe an endpoint to support.run.completed and support.run.failed with POST /v1/webhooks/settings (Support Assistant webhooks). Use an endpoint with no profileIds or accountIds filter: an answer can cover your whole team, so scoped endpoints never receive these events. Deduplicate deliveries on the event id.

2. Ask with an Idempotency-Key

Send one Idempotency-Key per question and reuse it when you retry. A retry with the same key and body returns the first response instead of starting, and billing, a second run. Point Ana at what the question is about with context (postId, accountId or profileId); each id must belong to your workspace.

3. Handle the answer by status

4. Poll only as a fallback

If a delivery is missed, read the run with Get a support answer, waiting pollAfterSeconds between calls. It returns the same object the webhook carries.

5. Ask follow-ups in the same thread

Send the threadId back with the next question to continue the conversation with the same context. A thread answers one question at a time: asking while one is queued or running returns 409 support_run_in_progress. List a thread's support answers returns every run in a thread, oldest first. Runs are kept for 90 days.

Limits

If it fails

Related

• Ask a support question: every request field and response.
• Get a support answer: read one run.
• List a thread's support answers: every run in a thread.
• Support Assistant webhooks: the support.run.* payloads.
• Pricing: every usage meter on your invoice.