Website to Voice Agent Knowledge Pack

Turn a public business website into a compact, evidence-linked knowledge pack for AI receptionists and voice agents.

What you receive

• Business name, phone numbers, email addresses, locations, hours, services, and FAQs with source URLs.
• A Markdown knowledge brief ready for review and import into an agent platform.
• A list of missing operating rules—such as emergency escalation, cancellations, service coverage, and after-hours behavior—that must be confirmed before launch.
• Crawl diagnostics so failed or robots-blocked pages are visible instead of silently ignored.

Why this is different

Generic crawlers return page text. This Actor produces a voice-agent setup artifact: caller-facing facts, cited evidence, and the operational gaps most likely to cause a bad call. It uses deterministic extraction, not an external large-language-model API, so there is no second API key and no invented answer presented as fact.

Input

{
  "websiteUrl": "https://example.com",
  "maxPages": 8,
  "timeoutSeconds": 20
}

Output

One dataset item per website. The item includes structured facts, agentSetupGaps, sourcePages, and knowledgeBaseMarkdown.

Safety and limitations

• Public HTTP/HTTPS business pages only. Local networks, private IPs, login walls, and paywalls are rejected.
• The crawler respects robots.txt, stays on the submitted hostname, limits response size, and visits at most 20 pages.
• It follows only high-signal business links exposed by the submitted landing page, which prevents deep-link drift into unrelated blogs, catalogs, or marketplace listings.
• Website content can be wrong or outdated. The output explicitly separates found evidence from facts the business must verify.
• The Actor does not deploy, call, edit, or connect to a live voice agent.
• JavaScript-only pages may return limited content in version 0.1 because this build intentionally avoids a browser and its added compute cost.

Suggested marketplace pricing

Use pay-per-event pricing with a custom completed-pack event. The event is attached only after a completed website pack is produced; failed crawl diagnostics are not charged. Initial validation price: $0.19 per completed website pack. At that price, 987 paid results per month produce $187.53 in gross event revenue; Apify's published creator formula is 80% of event revenue minus platform costs. This is a target calculation, not a revenue forecast.

Local validation

$python3 -m unittest discover -s tests -v

No account, paid API, or external write is needed for the unit tests.

Version 0.2 also passed a full Docker run against a seven-page public-site sample in 5.59 seconds. At the minimum 128 MB allocation, that duration implies roughly 0.000194 compute units, or about $0.00004 using Apify's published $0.20/CU reference cost. Data transfer and storage are additional. Version 0.3 changes billing behavior so only the custom completed-pack event can charge; a new bounded cloud test is required before the paid variant starts.