Research that shows its work.

§ 01 · What it is

A research engine that refuses to guess.

inquiry is a grounded deep-research engine. Give it a question and it discovers a population, derives a typed ontology for the topic, extracts every source against it, and grounds each value in the text that supports it. The ontology is versioned and reusable — a durable schema you re-query, not a throwaway prompt. It is governed so it won’t fabricate.

§ 02 · Why it’s different

One-shot AI gives an answer. inquiry gives an asset.

A search API hands back results and leaves the work to you. A chat model answers from its training data and asks you to trust it. inquiry answers from a population it actually built and canonicalized — and hands you the schema, the records, and the provenance, so you can re-run it, re-query it, and audit every line. What you keep is a dataset, not a transcript: the research an analyst would have done by hand, at the size of the web.

§ 03 · The method

Acquire. Understand. Improve. Deliver.

1. 01

   Acquire

   probe crawls and renders the open web — and your own sources — into content-addressed, replayable captures. inquiry never crawls; it reads the lake.

2. 02

   Understand

   Derive a per-topic ontology from sample captures, extract every source against it, ground each value to the text, and govern the result gold-free.

3. 03

   Improve

   A judge scores the schema against the evidence it produced and emits typed deficits; a repair loop refines it, escalating to a human when it can’t.

4. 04

   Deliver

   A typed, cited dataset plus a grounded narrative — queryable, versioned, and exposed to any agent over MCP.

• Grounding gate
• Gold-free governance
• Content-hash versioning
• Drift refresh
• On-prem local models

§ 04 · Where it earns its keep

Research you run more than once.

inquiry is built for the domains a business researches repeatedly — where the valuable artifact is a living, governed dataset, not a one-off answer. Each run is cheaper and better than the last, and re-running it tells you what changed.

MCP-native — drop inquiry into any agent. On-prem and private: it runs local models, no data egress. Configurable per stage, with pluggable search backends.

§ 05 · What’s live

What ships today, and what’s next.

The grounded core is live: capture → derive ontology → extract → ground → govern → persist, proven in production on coffee.church. Reasoning runs on-prem on a local model, so page text never leaves the cluster. The self-improving loop — a deficit-emitting judge, a repair planner, and a human-in-the-loop review gate — is a designed roadmap, not a finished claim. We say which is which.