I built Whilio after years in ads/sales where dashboards showed what happened but not why. Even with tagging, surveys, and A/B tests, it was still hard to see why users churn or convert. I wanted something that would ask questions in the moment and surface real themes from conversations.

What it does:

- Auto onboarding: crawl a domain, extract structure (products / docs / tone etc.), set its own tasks. Optional docs for overrides.

- Agentic RAG: per-page + global context, planner decides when to answer/ask/defer. Answerability checks reduce hallucinations.

- Data extraction: builds a lightweight knowledge graph, normalizes attributes/CTAs, tracks diffs.

- Insight triage: clusters language into Threats, Strengths, Opportunities -> real themes at scale.

- Environment-aware prompting: agent monitors page context, nav, messages, tasks -> decides when to engage.

What didn't work (and what changed):

- Vanilla RAG -> added answerability + planning.

- Noisy clustering -> switched to density-based + weak supervision.

- Manual curation -> replaced with automatic site learning + doc overrides.

Status: Open beta. Free demo w/o signup. Launching on PH.

Ask HN:

1. How to benchmark insight quality beyond pretty clusters?

2. Would you accept on-site "why" prompts at key moments?

3. Is self-hosting/on-prem a must in your context?

4. How important is to have SoC / ISO27001 for now?