create_workspace

A workspace's access-control setting, chosen once at creation.

labels = [] is the default and means no filtering at all — no classification pass at ingestion, no filter clause at retrieval, and Coverage reports nothing. A workspace that does not need document-level access control should not pay for a classification that always returns the same answer.

Coverage reporting is always strict: whenever the gate withholds candidates, the response says so in counts and label names. A silently narrowed answer is the failure mode this platform exists to prevent, so there is no opt-out.

A workspace's knowledge-graph build policy.

The defaults keep the whole corpus in scope and build the graph after every ingestion, so fact search and deep search have it to navigate by. Set auto_build=false for a workspace that builds only on request. exclude_path_prefixes and exclude_categories are the workspace-wide skip rules a file's graph_inclusion can override — a file matching either is left out of the graph corpus unless it is explicitly include\ d.

How this workspace's documents should be read where they leave a currency open.

Extraction never converts an amount and never invents a currency: one whose currency the source does not state is recorded as ambiguous with no value, so it matches no amount filter. New workspaces default to DEFAULT_WORKSPACE_CURRENCY (USD), so a corpus that writes every total as a bare $900.00 and names a currency nowhere is filterable out of the box. Set default to another ISO code to change the reading, or to null to assume nothing — an amount whose currency the source never states then stays ambiguous rather than being read as USD.

A workspace's ledger-understanding policy.

On by default in its cheapest form: auto costs a workspace with no ledger in it nothing at all, because the test for "is there a ledger here" is the census's own pure predicate over sidecars that ingestion already wrote.

A workspace's access-control setting, chosen once at creation.

labels = [] is the default and means no filtering at all — no classification pass at ingestion, no filter clause at retrieval, and Coverage reports nothing. A workspace that does not need document-level access control should not pay for a classification that always returns the same answer.

Coverage reporting is always strict: whenever the gate withholds candidates, the response says so in counts and label names. A silently narrowed answer is the failure mode this platform exists to prevent, so there is no opt-out.

A workspace's knowledge-graph build policy.

The defaults keep the whole corpus in scope and build the graph after every ingestion, so fact search and deep search have it to navigate by. Set auto_build=false for a workspace that builds only on request. exclude_path_prefixes and exclude_categories are the workspace-wide skip rules a file's graph_inclusion can override — a file matching either is left out of the graph corpus unless it is explicitly include\ d.

How this workspace's documents should be read where they leave a currency open.

Extraction never converts an amount and never invents a currency: one whose currency the source does not state is recorded as ambiguous with no value, so it matches no amount filter. New workspaces default to DEFAULT_WORKSPACE_CURRENCY (USD), so a corpus that writes every total as a bare $900.00 and names a currency nowhere is filterable out of the box. Set default to another ISO code to change the reading, or to null to assume nothing — an amount whose currency the source never states then stays ambiguous rather than being read as USD.

A workspace's ledger-understanding policy.

On by default in its cheapest form: auto costs a workspace with no ledger in it nothing at all, because the test for "is there a ledger here" is the census's own pure predicate over sidecars that ingestion already wrote.