Analyze, review, apply
Analyze website saves a background job. Run web and worker together; progress remains available across chapters, navigation and refresh. Official crawling and search discovery start together, with up to four page workers. Chat summarizes up to eight matched excerpts into concise editable company fields. Source matching validates provenance, not truth. Findings prefill empty or unchanged fields; later manual edits are protected. Review and Save persist the profile.
Changing the website clears the previous profile fields in the draft, updates the setup identity, and is saved when Analyze starts. Old-domain research is excluded from retrieval. A blocked official site can use matched third-party evidence marked provisional. Unknown facts remain blank; company size needs explicit staff-count evidence. Source restrictions are respected. Reports separate crawling, chat and indexing results; readable text survives model failure. Vision/video understanding is not part of this flow.
Free crawler
The bounded crawler follows same-site links and sitemaps, prioritizes business pages and can use Playwright for sparse JavaScript-rendered pages. It attempts 5, 15 or 30 pages within an approximate 100-second budget. Robots restrictions, login walls and challenges are reported rather than bypassed. This does not guarantee every page or directory entry.
Optional APIs and directories
Firecrawl can be used for scraping when configured, with free crawling as the automatic fallback. Google Custom Search can assist discovery for eligible existing customers. Directory URLs may be supplied or discovered from public links; LinkedIn and directory content can be incomplete or unavailable without access. Third-party facts remain unverified until a person confirms them.
Saved evidence
Reports are stored as Website research knowledge with URLs, timestamps, coverage and failures, and can be downloaded as JSON. See research setup for optional keys and boundaries.
The report distinguishes complete, partial and failed attempts. Zero-page attempts are retained for diagnosis but do not produce searchable evidence chunks. A report from a different website is flagged before applying fields. The crawler uses the Playwright browser library directly, not a Playwright MCP server. Rendering waits for the document and a bounded content check rather than waiting indefinitely for network silence.
Brand candidates
The Brand step shows color, font and logo-link candidates from website metadata, inline styles and rendered body/button styles when available. Review the source and extraction method before applying colors or fonts. Logo links are not imported automatically; upload an approved asset. Color customization is optional, and the preview protects text contrast. These settings guide company outputs and do not replace the application's shared visual theme.