What identifies a namespace
A namespace is identified by three values together: its name, its embedding model, and its dimension count. All three are required on every publish, search and subscribe, and Semantik guesses none of them. The reason is that one name can carry more than one embedding configuration, and vectors produced by different models are not comparable. If the model were inferred, a query could be answered in the wrong space and return confidently wrong results.Creating a namespace
- Go to the Semantik dashboard.
- In the namespace table, select New namespace.
- Enter a name and save.
Namespace creation requires an active subscription. If you land on a subscription-required page, complete billing setup under Settings → Billing first.
The shared global namespace
global is a well-known namespace available to every account, provisioned for the Qwen3-Embedding-4B model at 1024 dimensions. Other model and dimension combinations under that name are not provisioned and are refused.
Targeting a namespace
Every publish, search and subscribe scopes itself with thenamespace field in the request body, alongside model and dimensions:
NAMESPACE clause, which restates that scope so it can be checked:
"namespace": "acme-staging" is refused with 400 invalid_request rather than answered from staging, which is what makes it useful for a query that travels between environments.
One request reaches one namespace. Selecting several from a query is not supported yet: multiple names,
NOT exclusions and ALL are rejected rather than ignored. To cover several namespaces, issue one request each.
