Ask AI

Turn on the assistant and readers get an "Ask AI" panel that answers questions using your published docs as its source. It's grounded in your content, so answers stay accurate and cite the pages they came from.

Enable it

Add an assistant block to docs.json:

{
  "assistant": {
    "enabled": true
  }
}

Once published, the assistant reads your live site's content — there's nothing to train or sync.

Starter questions

Seed the empty state with up to three suggested questions so readers know what to ask:

{
  "assistant": {
    "enabled": true,
    "starterQuestions": [
      "How do I authenticate?",
      "What are the rate limits?",
      "How do I handle webhooks?"
    ]
  }
}

Support escalation

Give the assistant a fallback so it can point readers to a human when it can't help:

{
  "assistant": {
    "enabled": true,
    "supportEmail": "support@acme.com"
  }
}

Skills

Skills are short Markdown playbooks that steer the assistant on specific topics — a migration guide, a troubleshooting flow, a getting-started path. They load on demand: when a reader's question matches a skill, the assistant follows that skill's guidance instead of relying on plain search.

Put one Markdown file per skill under content/assistant/skills/, each with a name and a description in its frontmatter. The description is what the assistant matches against, so make it specific about when the skill applies:

---
name: Troubleshooting failed requests
description: Diagnose failing or rejected API requests — auth errors, 401/403 responses, rate limits, invalid parameters.
---

When a request is failing, diagnose in this order:

1. **401 Unauthorized** — the API key is missing or revoked...
2. **429 Too Many Requests** — a rate limit was hit; retry with backoff...

At most three matching skills load per answer, and each is trimmed to keep the context focused. Skills take effect on your next publish. A companion content/assistant/instructions.md adds standing guidance (tone, terminology) to every answer.