The agent gathered a partial but incomplete understanding of NUSOMI's core value proposition (private AI deployment in your own cloud with OpenAI-compatible endpoints) by extracting information from the homepage and an OpenAPI spec, but could not fulfill the pricing component of the task—a critical omission. The site is poorly navigable for comprehensive product evaluation: most standard pages (pricing, docs, blog, about, FAQ) return 404s, and the homepage lacks detailed feature breakdowns, supported cloud providers, or commercial terms.

• ›The homepage ([1], [7]) was the only content-bearing fetch and contained the core positioning (meta description: 'policy-aware control plane for open models... Prompts never leave your boundary') plus vague workflow references ('voice approval', AWS deployment). This single source had to carry the entire answer.
• ›The OpenAPI spec ([19]) confirmed a waitlist endpoint but revealed nothing about pricing, models, or features—it only validated that the product is in early/beta stages and collects contact info via forms.
• ›Standard navigation paths (pricing, blog, docs, about, FAQ) all returned 404s, forcing the agent to rely on the homepage alone and make inferences. The agent correctly identified this as a 'confusing/missing' element but had no alternative way to retrieve pricing tiers, support levels, or cloud provider coverage.
• ›Web searches ([4], [5], [8], [9], [10], [15], [16], [17], [18], [20], [21], [22]) yielded no results specific to NUSOMI—they returned unrelated products (Nusii proposal tool, Nozomio, etc.) or general AI deployment guides. The homepage copy explicitly states 'Everything lives on the homepage,' which the agent interpreted as an invitation to find all details there, but critical business information is actually absent.
• ›The agent had to infer enterprise/custom pricing from the 'Book a demo' CTA and the presence of a waitlist—reasonable deductions but not grounded in any fetched content.
• ›Prior knowledge was used for 44% of the run (primarily initial guesses at /pricing, /blog, /docs, /about, /faq), all of which failed, demonstrating the site's minimal public footprint. The agent relied heavily on prior knowledge to construct a coherent narrative around 'control plane' and 'open models' rather than explicit site content.