What do people ask while reading your docs?

Docs readers are often partway through building something. They know what they want to do but not which page explains it, or they’ve found the page and want to check they’ve understood it:

  • “Which endpoint lists the invoices for one customer?”
  • “What’s the difference between a test key and a live key?”
  • “Do webhooks retry if my server is down?”
  • “Is there a rate limit on exports?”

Which docs questions suit a voice agent?

Questions about concepts, choices and where things are. A spoken answer works well for “which endpoint should I use” or “what does this setting change”, and the reader can turn on live captions to see names written down.

Code is different. A JSON payload read aloud is hard to follow, so tell the agent in its instructions to name the page and section that hold the example, and let the reader copy it from there.

How do you give the agent your docs from GitHub?

Import each .md, .mdx or .txt file into the support agent’s knowledge by repository and path, one file per source and up to 50,000 characters each. Public repositories need no token. Ask us before relying on a private one, because access is set up on our side for each workspace.

Nothing syncs when you push, and a re-synced file waits for approval again, so make re-syncing part of your docs release. The docs-as-code guide covers the rest: how the import reads MDX, how file names help the agent find a page and a release workflow that re-syncs changed files for someone to review.

Where do docs readers still need the page or your team?

The agent explains what your docs say. These cases need more than that:

  • Their own setup. The agent can’t see a reader’s account, keys or logs, so it can’t debug their integration.
  • Pages that aren’t Markdown files, such as API reference generated from a spec at build time. Add the parts readers ask about as manual sources.
  • Other languages. The agent greets readers in English, and there’s no language setting.
  • A person. Handover is an early release that rings one verified UK or US number for the support agent, so verify one your developer support team answers before you turn it on. Whoever picks up hears a private briefing first. Customer routing is still being tested, so also name your usual developer support route in the agent’s instructions.

Common questions

Can the voice agent read our docs straight from GitHub?

Yes, one file at a time. Kawl Now imports .md, .mdx and .txt files of up to 50,000 characters into support knowledge, from public repositories without a token. Each import needs approval before the agent uses it.

Is a voice agent a good fit for code questions?

For choosing an endpoint, a setting or an approach, yes. For the code itself, less so, because a payload read aloud is hard to follow. Tell Kawl Now’s agent to name the page that holds the example, and let readers copy the code from there.

Is Kawl Now a documentation chatbot?

It answers from your documentation, but by voice. A reader talks to the agent in their browser and can turn on live captions. Kawl Now has no typed chat mode, so keep your docs search for readers who want to type.