Every enterprise has a data agent mandate; very few can ship one. Datus turns the warehouse you already own into a governed data context layer — an Apache Ossie semantic layer and ontology, not a general-purpose memory. The context your data agents can trust, live in days and evolving with every query.
The blocker is not the model. It is decades of messy, siloed data no agent can safely navigate on its own — so the pilot stalls and the bill arrives anyway. Datus gives agents a governed path through it.
One standards-based stack: a commercial Studio on top, an open-source agent in the middle, the Apache Ossie standard and the Dosi engine underneath.
Commercial · where humans, BI and agents interact
Open source · generates the models, runs the work
The standard · defines and compiles the semantics
Datus reads what you already have and compiles it into a governed Apache Ossie context layer — then keeps refining it from real usage.
Start with one governed domain, prove it in an afternoon, then expand domain by domain.
CLI, REST with Apache Arrow, native MCP and a Python SDK. One data context layer, every surface — humans, BI and agents get the same governed answer.
Point Datus at your warehouse, catalog, semantic layer and BI. The Dosi engine compiles OSI to native SQL for 16 warehouse dialects, so nothing has to migrate.
The data context layer, Apache Ossie, the Dosi engine, open core and how Datus compares.
A data context layer is a governed semantic layer and ontology built directly on your warehouse: it teaches AI what your entities, metrics and relationships mean, and gives it clean, governed paths to measure them. Unlike a general-purpose memory store, it is grounded in the data itself — the context an enterprise data agent needs to answer correctly instead of guessing at joins and definitions.
The blocker is rarely the model. It is decades of messy, siloed data no LLM can safely navigate on its own, so the demo works on sample data and dies on the real thing. Datus connects to the warehouse you already own and compiles a governed context layer in days — agents get clear, governed paths through the business, so the pilot reaches production.
OSI (Open Semantic Interchange), now Apache Ossie, is an open standard for defining ontology and semantic models — backed by 50+ organizations including Snowflake, Databricks, Salesforce and dbt Labs. Datus defines your models in OSI specs rather than a proprietary format, so your context layer is portable and vendor-neutral. The spec is open to everyone; our engine and our position in it are the product.
Dosi is a Rust engine that compiles OSI YAML to native SQL for 16 warehouse dialects, with zero-copy Arrow streaming. It is on its way to open source and already free to download and use — the same engine and the same answers behind the CLI, REST, MCP and the SDK.
The open-source Datus agent, the Dosi engine and the Apache Ossie models are the core, free to run on your own warehouse. Datus Studio is the commercial layer on top: a context-aware co-pilot where humans, BI and agents drive the whole stack in natural language, with SSO, RBAC and private / VPC deployment for enterprises.
Snowflake Semantic Views feed Cortex Analyst and Databricks Metric Views feed Genie, but neither reads the other — context inside one vendor stops there. Datus sits on neutral ground: an Apache Ossie semantic layer and ontology on the warehouse you already own, compiled to native SQL across 16 dialects, that agents call over MCP.
Open Datus Studio free in your browser, or self-host the open-source agent and Dosi engine on the warehouse you already own.
