Review the metric once
Describe the dataset, metric expression, dimensions and relationships in an Apache Ossie model. Your team can review the definition before publishing it to agents or applications.
DOSI + Databricks SQL
A semantic layer for Databricks data and AI agents.
Keep your data and compute in Databricks. Define metrics in a team-reviewed Apache Ossie model, then let DOSI plan the query and send native SQL to your Databricks SQL warehouse.
Existing SQL + business rulesTeam reviewApache Ossie model
Agents + apps · MCP / API / CLIDOSINative SQL via Statement Execution API
Databricks SQL warehouse
Results return through DOSI to your tools
Definitions live outside the warehouse; the configured Databricks SQL warehouse executes the query.
Apache Ossie definitionsDatabricks SQL warehouseMCP · API · CLI
Where Datus fits
Two ways to give agents business context.
Genie Ontology brings governed Unity Catalog semantics together with context inferred from Databricks assets. DOSI takes a different route: a team-reviewed Apache Ossie model sits outside the warehouse, and DOSI compiles metric requests into Databricks SQL.
Databricks + Genie Ontology
Unity Catalog Semantics + inferred context from queries, dashboards and agents
Genie Ontology
Genie One / Code / Agents · external agents via MCP
Databricks SQL warehouse
Context is curated and inferred within the Databricks platform.
Databricks + Datus / DOSI
Team-reviewed Apache Ossie model
DOSI semantic layer
Agents and apps · MCP / API / CLI
Databricks SQL warehouse
The semantic model is managed separately and can be reused with supported SQL backends.
Use the tools your team already has
DOSI exposes compatible definitions through MCP, API and CLI interfaces. It does not require you to move the underlying Databricks tables into a new warehouse.
See the warehouse query
DOSI compiles the selected metric request to Databricks SQL and sends it to the SQL warehouse. Inspect the model, generated SQL and results in your workflow.
Choosing a Databricks semantic layer
Define shared metrics in DOSI. Reuse them across BI.
Keep your Databricks data foundation. Define shared metrics in an Apache Ossie model, let DOSI compile and execute them, and serve results to BI workflows through an API integration.
Gold tables or semantic definitions?
Keep data preparation and reusable tables in your existing pipelines. A semantic model describes how to calculate and group business metrics over those outputs. With DOSI, reviewed Apache Ossie definitions reference your Databricks data; Databricks still runs the resulting SQL.
Explore semantic model examplesOne metric definition for Power BI, Tableau and other BI tools
Define the calculation once in the model used by DOSI. DOSI compiles the requested metrics, dimensions and filters into Databricks SQL. An integration layer maps BI requests to the DOSI API, which executes the query on Databricks and returns results. Keep metric calculations in DOSI while Power BI, Tableau and other BI tools present the data. This approach requires a BI-specific adapter.
Explore DOSI interfacesMaintain shared KPIs in one reviewed model
Your data team owns the metrics in an Apache Ossie model; DOSI uses those definitions to plan and execute requests. Review changes centrally and have consuming workflows request the same metrics. This gives dashboards and agents a shared calculation source, with fewer separately maintained formulas.
Review the semantic layer approachOne shared model. Metrics served through DOSI.
Apache Ossie (incubating) defines the shared model. A BI adapter sends metric requests to DOSI; DOSI compiles compatible definitions into SQL, executes it on Databricks and returns results.
Apache OssieShared metrics · dimensions · relationships
BI & reportingPower BI
Tableau
Tableau
BI integration adapterMap metrics, dimensions & filtersREST API
DOSIValidate · compile · execute
DatabricksExecute native SQL← Results through DOSI and the adapter
Integration architecture: the BI adapter must be configured or developed for the target tool, including authentication, request mapping and refresh. This is not a built-in Power BI or Tableau connector.
Explore the open semantic layerA concrete workflow
From a reviewed metric to a warehouse query.
Configure a SQL warehouse, validate your Apache Ossie model, then request a metric such as revenue by order status.
dosi-connections.yaml
datasources:
databricks:
type: databricks
host: ${DATABRICKS_HOST}
warehouse: ${DATABRICKS_WAREHOUSE_ID}
password: ${DATABRICKS_TOKEN}
catalog: analytics
schema: reportingDOSI CLI · example request
dosi query --model model.yaml \
--metrics revenue --group-by orders.status \
--execute --connection databricksThe command requires your own validated model.yaml and a running SQL warehouse. This page does not execute a Databricks query or present fabricated results.
Connect Databricks SQL
DOSI Databricks connector docsUse a SQL warehouse as the execution endpoint.
DOSI connects through the Databricks Statement Execution API. The connection needs a workspace host, SQL warehouse ID, token and appropriate catalog access.
- 01
Choose a SQL warehouse
Copy the warehouse ID from Databricks connection details. Grant the token holder access to the warehouse and the catalog or schema your model references.
- 02
Keep credentials out of the model
Set the workspace host, warehouse ID and token in a DOSI connection profile. Reference the token through an environment variable rather than committing it.
- 03
Validate and run a metric
Review the model, run the sample request with --connection databricks, and inspect the SQL and result. A stopped warehouse may need time to start.
DOSI connects to your Databricks SQL warehouse through the Statement Execution API. Queries run with the permissions of the configured token.
Useful for teams
One meaning, several consumers.
The goal is to make reviewed business definitions available where people and agents ask questions.
Agent-assisted analysis
An agent can ask DOSI for a named metric instead of inferring its meaning from table and column names alone.
Shared reporting logic
Applications and analysts can refer to the same model while Databricks remains the query engine and source of results.
FAQ
Databricks semantic layer FAQ
The scope and setup details that matter before you connect.
Does DOSI replace Databricks SQL?
No. DOSI reads a compatible semantic model, compiles a metric request and sends SQL to the Databricks SQL warehouse you configure. Databricks executes the query.
How does DOSI differ from Genie Ontology?
Genie Ontology combines governed Unity Catalog semantics with inferred context for Genie experiences and external agents via MCP. DOSI uses a separately managed, team-reviewed Apache Ossie semantic model to plan metric queries across supported SQL backends. DOSI is not an ontology product; on this page, Databricks remains the query engine.
Can I reuse existing Databricks metric definitions?
Yes—start with exported definitions. Metric View YAML contains sources, dimensions or fields, measures, filters and joins. DESCRIBE TABLE EXTENDED <catalog.schema.metric_view> AS JSON exposes the YAML in view_text. SQL files and exported dashboard .lvdash.json files can also supply calculation evidence. Give these files and the relevant source schema to Datus Agent as context for drafting an Apache Ossie semantic model. Supported Metric View definitions can also use the Apache Ossie Databricks converter. Review mappings, calculation grain and unsupported features, then validate the model with DOSI and compare representative results in Databricks. This is a file-based conversion workflow, not automatic live synchronization.
Is Unity Catalog metadata already a semantic layer?
Catalog metadata and semantic calculations have different roles. Unity Catalog metric views add reusable measures and dimensions; table names and descriptions alone do not define every KPI. DOSI uses a separate compatible Apache Ossie model to make metric calculations explicit before compiling a query for Databricks SQL.
Does a semantic layer replace my Gold layer or dbt transformations?
No. Curated Gold tables and views can remain the source data. The semantic model defines business calculations, dimensions and relationships over that foundation. Decide which transformations belong in prepared data and which calculations should remain reusable across query groupings.
Will catalog descriptions alone give my AI agent the right metrics?
Descriptions help interpretation, but the agent also needs an explicit calculation and a supported way to request it. In a DOSI workflow, expose the reviewed model through the supported agent interface and inspect the generated SQL. A warehouse connection alone does not make an agent use that model or guarantee a correct answer.
Start with a metric your team already trusts.
Turn that definition into a reviewable model, then connect DOSI to your Databricks SQL warehouse.