Model Registry

Read your registered models and their champion through whichever registry engine this workspace has.

A registry is where a trained model becomes a version somebody can serve, and which version an alias points at is the deploy signal. This hub carries no registry of its own: it reads one through a declared engine contract, so the backend is a choice rather than a definition. Install the MLflow engine and this panel reads MLflow; write another engine and it reads that, with no change here.

What you can do

• See every registered model — one row per model with the version champion points at and the newest version, so the two are never confused.
• Open one model's versions — each with its registration status, its aliases, and the run that produced it, which is the link an evidence package walks back to the training data.
• Know which engine answers — the active engine, whether its backend can be reached right now, and when it cannot, that engine's own sentence for why. A server that is down, an address that is wrong and a host that cannot make an HTTP request are three different fixes.
• Ask programmatically — modelRegistry.engines reports the installed engines and their status. The models themselves are already a capability the engine declares.

What it does not do

It does not promote. Moving an alias is the one governed moment in this family: a person is asked, the answer is recorded as a single-use grant, and the call is dispatched carrying the trace of the run that trained that version — without which the promotion falls outside the evidence package that answers "who approved this". Resolving that trace means knowing the backend, so the promote action opens the engine's own panel and the write keeps exactly one implementation.

The engine contract

An engine is a set of declared capability ids. The hub finds it by reading the capability catalogue once; nothing is imported across extensions and nothing is registered by hand.

Id	Required	Meaning
mlops.<engine>.modelList	yes	every registered model with its alias map
mlops.<engine>.modelGet	no	one model with its complete version list
mlops.<engine>.engine	no	which backend, whether it is reachable, what it can do
mlops.<engine>.reveal	no	the engine's own panel, where a promotion is approved

An engine that declares only the first is listed and read from; the panel says what it cannot do rather than offering a click that fails. The full contract, including what an engine has to be able to do to qualify, is punica-editor/docs/model-registry-engines.md.

Requirements

A registry engine. extension-experiments 2.19.0 or newer ships the MLflow one and needs a reachable MLflow tracking server, which it can run for you.

Getting started

1. Install this extension and a registry engine.
2. Open Model Registry from the activity bar.
3. If the panel says the registry cannot be reached, the sentence under the engine row is the engine's own and names the fix.

License

MIT