Dosi

by DatusAI — a component of Datus Studio

An OSI-native semantic layer engine for metrics. Dosi compiles Apache Ossie (OSI) YAML into warehouse SQL for
metrics, dimensions, entity joins, filters and time grains, and exposes that compilation through a CLI, a REST API,
native MCP and a Python SDK. Tagline: “Define once. Use everywhere.”

Overview

Dosi is a compiler, not an authoring tool or a catalog. The OSI file is the contract: Dosi takes it as input and
produces executable SQL, so the same metric definition can be used by a person, a CI pipeline or an AI agent.
Launched in August 2026 and described by its maker as the first engine of its kind.

Core model

  • Compile to one SQL statement per request — metrics, dimensions, entity joins, filters and time grains
    are compiled together
  • Deterministic output — the same input gives the same SQL, so teams can gate CI on whether an OSI file
    compiles for each SQL dialect they run
  • Structured errors — returns error codes that let an agent correct a grain, filter or dimension mistake
    instead of guessing
  • Four interfaces — CLI, REST with Apache Arrow, native MCP, and a Python SDK

Release history

  • 0.1.13 — Level of Detail (LOD) metrics: include, exclude and fixed grains, borrowed from Tableau’s LOD expressions
  • 0.1.14 — LOD works across fact tables and on top of composed metrics
  • 0.2.0 (October 2026) — derived subtypes, defined as versioned business rules such as “Premier Customer”; native execution on Athena, Redshift and SAP HANA; MCP hot reloading of models

Platforms and integrations

  • Warehouse dialects — 15+ dialects; named examples are Snowflake, BigQuery, Databricks, PostgreSQL, Spark
    and Trino, plus related variants. The full, current list is on the Dosi documentation site.
  • Datus Agent — through the datus-semantic-dosi adapter, which runs OSI metrics without MetricFlow
  • Datus Knowledge — handles schema and metric retrieval alongside Dosi
  • MCP clients — such as Claude Desktop and custom copilots
  • CI and orchestration — through the CLI and Python SDK

Licence and status

Dosi is not open source. It is distributed under the Elastic License 2.0, as stated on the Datus website.
Dosi is available; the announcement does not describe a waitlist.

Scope

  • Does not replace the whole semantic stack. Metric authoring can stay in MetricFlow or Cube.
  • Does not migrate legacy LookML or proprietary YAML unless an OSI export path exists.
  • Is not a catalog or memory layer. Those roles belong to Datus Knowledge.

Tiers

Dosi does not appear in any tier on the Datus pricing page. The tiers there cover Datus Agent (Open Source) and Datus Studio (Cloud Personal and Enterprise).

Sources