About this product

Five bodies of knowledge already exist and none of them talk to each other. Product discovery lives in AI directories. Implementation lives in repositories and model registries. Governance lives with regulators, in compliance directories and in vendors’ own legal documentation. Model infrastructure lives in the registries and provider docs. The people who have deployed these systems live in forums and job markets.

A firm with a real problem has to cross all five to get one decision. This system crosses them for you: it reads your question into a structured deployment brief, queries those sources live, stores what it retrieved with the date and a hash of the contents, and writes an answer where every statement carries its class — fact, assessment or recommendation — and, for a fact, the document it came from.

The starting scope is jurisdictions where privacy, sovereignty, regulation and vendor restrictions actually decide the answer: Hong Kong, Mainland China, Japan, South Korea and Taiwan in North Asia; the European Union and the United Kingdom in Europe.

One answer runs through four questions, in this order. Skipping any of them produces the kind of recommendation that survives a demonstration and not a deployment.

1. 01

   Discover

   Is there AI for this?

   Live search across repositories, model registries, directories and the open web.

2. 02

   Deploy

   How do I actually implement it?

   A stack, an architecture, hardware sizing, cost and a step-by-step recipe.

3. 03

   Govern

   Can I safely and legally deploy it?

   Issue-spotting per jurisdiction, cited to the regulator and the vendor’s own terms.

4. 04

   People

   Who can deploy it, and how do I learn to?

   The skills it needs, and the engineers who have shipped it before.

The chain ends at a decision, not a list: what you should do, what it costs, what it requires of you, and where it is risky.

Five readers, with different questions about the same deployment. The answer is one document because they have to be able to read each other’s sections.

1. A

   SME operator

   “We have 30 accountants and want internal AI search.”

   Wants a recommended solution · an estimated cost · whether it is safe · whether data leaves their infrastructure · implementation difficulty · someone to implement it

2. B

   IT, CTO or technical team

   “We want a local RAG stack for confidential documents.”

   Wants repositories · model requirements · architecture · deployment instructions · licences · hardware · scaling considerations · alternatives

3. C

   Legal, compliance or DPO

   “Can our German HR team use this AI tool for recruitment?”

   Wants the applicable jurisdiction · AI Act issues · GDPR implications · DPA availability · data residency · training and retention terms · subprocessors · source evidence · the last verification date

4. D

   Forward-deployed engineer

   Wants implementation recipes · emerging tools · community knowledge · technical skills · deployment case studies · jobs · potential client work

5. E

   A company looking for an engineer

   Wants someone who understands their problem · the relevant technology · their regulatory environment · a concrete implementation scope

A product is also what it declines to do. These are the declines, and what happens instead of each.

Not a directory of AI tools.

Directories already exist and are good at what they do. This system queries them and cites them; it does not copy them.

Not a competitor to GitHub, Hugging Face or OpenRouter.

Those are the primary registries for repositories and models. They are read through their official APIs and named as the source of every figure taken from them.

Not an AI lawyer.

Compliance output is issue-spotting against cited primary sources — the ordinance, the regulator’s guidance, the vendor’s own terms — for a human to act on.

Not a ranking by popularity.

Options are scored against the brief you described, on weighted dimensions that are printed with their weights. Sponsorship data is not an input to ranking.

Not a generic forum.

Contributions attach to a specific claim, recipe or entity, are reviewed, and change the record when accepted.

Every displayed statement carries a class, as a word rather than a colour. A FACT is externally verifiable and points at a document that was fetched, dated and hashed. An ASSESSMENT is our interpretation and shows the inputs it rests on. A RECOMMENDATION is a conclusion about your brief and names the attributes it depends on.

A fact that cannot point at a retrieved document is removed from the answer rather than softened. Where something could not be verified, the answer says so and the gap appears in its warnings. How answers are produced describes the pipeline and the rules it enforces; the source registry lists every source this deployment can reach and how it reaches it.