The agent assembled a comprehensive understanding of Shiori.sh by combining the official homepage/docs fetches with third-party reviews and the creator's public statements on X, since the website itself does not publish pricing or positioning details in a traditional, fetched page structure. The site is minimally agent-ready: it renders as a single-page app with sparse visible content, forcing heavy reliance on web search results and off-site sources (blog posts, tweets, browser extensions) to answer 'what it is,' 'who it's for,' 'how it's priced,' and 'how it differs from alternatives.'

• ›The official shiori.sh homepage [9] and docs pages [18, 19] returned HTML but did not expose pricing tiers, business positioning, or competitive comparisons in a machine-readable form—agent had to visit external sources.
• ›Pricing and product positioning were primarily sourced from off-site third-party content: Yury Molodtsov's blog review [12] explicitly described features and positioning; Brian Lovin's X thread [20, 21] disclosed pricing shifts and free-tier changes; Firefox extension page [13] provided platform scope. The official site did not surface these details when fetched.
• ›Traditional route attempts failed (shiori.sh/pricing → 404, shiori.sh/about → 404) [1, 2], forcing the agent to rely on 50% web search and external artifacts (42%) rather than direct site navigation.
• ›Naming confusion persisted: the agent correctly identified shiori.sh (Brian Lovin's AI-integrated cloud bookmarker) vs. go-shiori/shiori (self-hosted open-source) vs. shiori.ai (unrelated AI chat tool), but this disambiguation required search results and external context, not site structure.
• ›The site's tagline 'Personal memory for you and your AI' [0, 9] is present in metadata but the meaning and differentiation are not unpacked on the site itself—external blog posts and the creator's tweets were the primary source for explaining the AI/MCP integration angle.