API key — A long-lived credential (alt_…) minted for a service user, so a script or scheduled job can call the API without a human signing in. Binding — The link between a concept and the semantic model that realizes it — how “Customer” resolves to an actual table and columns. Capability (driver / federation / mcp) — What an integration is allowed to do once connected. driver lets a query engine read its data directly; federation lets other catalogs be queried alongside it; mcp exposes it to MCP tools. One integration can hold more than one. Catalog — A named collection of schemas and tables a query engine can see, usually one per connected integration. Catalogs pending / apply — After you change what an engine is connected to, or a new workspace database starts serving, its new catalog set is “pending” until you apply it. Applying restarts the engine and interrupts queries running on it at that moment. Certification — Marking a semantic model or concept version as the definition your tenant trusts. Only a data_admin can certify or deprecate one; anyone who can edit a draft can propose a new version, but never bless their own. Column mask — A rule that rewrites a column’s value (redacting or hashing it) for a given role, enforced at query time. Concept — A real-world thing your tenant defines once — what “Customer” means, its properties, and its relationships to other concepts — independent of which table implements it. Consent — Your explicit approval of what a connecting client (an AI agent, dip, a data app) may do on your behalf, shown as a list of scopes before you approve. Crawl — A metadata scan of a query engine’s catalogs that populates search and lineage. Costs engine time and, for models with embeddings, indexing spend. Data app — A packaged application that runs inside your data plane and queries it on your users’ behalf, installed from a manifest. Data plane — The execution boundary where your compute and data actually live: your query engines, data apps and workspace databases all run inside one. dip — Dippr’s command-line tool, for connecting to your tenant, running queries, and managing engines, integrations and apps from a terminal or a CI job. DuckDB — The embedded database engine behind a workspace database; you write to it directly and query engines read it live as a catalog, which its owner can also write with Trino SQL. Engine size — The rung of compute a query engine runs at (xs through xl, or custom on Enterprise), fixed at creation and changeable later. Explore — The console’s page for browsing catalogs and running SQL against your query engines. Grant — One rule attached to a role: this role may do this action on this kind of thing (or, for a deny, may explicitly not). Installation — One instance of a data app running in your data plane, with its own consent and its own wake/park state. Integration — A connection Dippr holds to an external system — a database, warehouse or file store — including its credentials and what the platform may do with it. Lineage — The graph of what produced what: which tables and queries a given table or column depends on, and what depends on it. Listing — One ready-made definition in the Marketplace, published at a named version, that an install copies into your tenant as drafts you own. A runnable listing (a pipeline, an agent or an app) is an image you add to its registry instead. Manifest — The file that declares a data app: its name, what it needs, and how to run it. Marketplace — A curated catalog of ready-made semantic models and ontologies you can install into your semantic layer as drafts, and pipelines, agents and apps you can add to your tenant. Included in the standard tenant bundle. See Marketplace. MCP (Model Context Protocol) — The protocol an AI agent uses to connect to your tenant and call a fixed set of tools (running SQL, searching metadata, compiling a metric) on your behalf. Member — A person who belongs to your tenant. Metadata search / index — Search across everything a crawl found — tables, columns, queries — from one search box, backed by an index the crawl builds. Metric — A named, governed calculation defined in a semantic model — what “revenue” or “active users” means and how to compute it. Ontology — Your tenant’s set of business concepts and the relationships between them, which can be realized by bindings to semantic models. An unbound concept is vocabulary with no data behind it. See Ontology concepts. Plane registry — The image registry a data app’s image is pushed to before it’s installed, namespaced to your tenant. Query engine — A managed Trino cluster that runs SQL against your connected catalogs. Query history — The record of every query your tenant’s engines have run, shared across everyone who can read it. Question — A natural-language question your tenant has agreed how to answer, paired with the verified SQL that answers it. Relationship — A declared connection between two concepts (a Customer has Orders), part of the business vocabulary. Review queue — Where a suggested change to a semantic model or concept waits for a certifier to accept or reject it. Role — A named set of grants. Every member holds one or more roles, and a role can inherit another’s grants. See the built-in roles. Row filter — A rule that restricts which rows of a table a given role can see. Row-level security — The industry term for what Dippr calls a row filter. Scope — A named capability an OAuth token carries — what a connected client is allowed to ask for, independent of your own roles. See OAuth scopes. Semantic model — A governed definition of what a table means: its metrics, dimensions, and which table they bind to. Service user — A non-human account for scripts, jobs and integrations, held to roles and grants exactly like a person. Shared plane — A data plane whose owner allows other tenants to bind to it, rather than being restricted to the tenant that registered it — typically a platform-operated regional plane you join rather than run yourself. SQL workbench — The query tab in Explore where you write and run SQL directly against a query engine. SSH tunnel — A secure channel an integration’s connector uses to reach a source database that isn’t directly reachable from the data plane, by hopping through a bastion host you control. Suspend / wake / park — The power states of a query engine or app installation: suspended (stopped, no cost, wakes on the next request), waking (starting back up), parked (deliberately kept off until you wake it again). Tenant — Another name for a workspace in Dippr. Both terms refer to the same space with its own members, roles, integrations, engines and data. Tool — One callable action an MCP client can invoke — running SQL, searching metadata, compiling a metric — each requiring the scope that covers it. Trino — The open-source distributed SQL engine that powers a Dippr query engine. Workspace — A workspace is an isolated space in Dippr with its own address, members, roles, integrations, engines and data. An account can belong to several workspaces and hold a different role in each. You can create up to 3 workspaces that still exist; workspaces you join by invitation do not count toward that limit. A workspace is also called a tenant. Workspace database — A DuckDB database you own inside your data plane, personal by default or shared with specific members, that you write to directly and query engines read live as a catalog.

Roles and permissions matrix

The built-in roles Dippr ships with.

OAuth scopes

Every scope Dippr issues, and what it lets a client do.