Scan Snowflake Cortex Agents and Semantic Views Into Dawiso and Generate Them Back via OSI

How Cortex agents consume your data

A Snowflake Cortex agent answers a business question by orchestrating tools. For structured data it calls Cortex Analyst, and Cortex Analyst reads a semantic view rather than your raw tables. The semantic view is where the business logic lives: defined metrics, relationships and join paths, and verified query examples sitting on top of the underlying tables.

When you attach a semantic view to an agent as a tool, the semantic view name becomes the tool name in the agent configuration. So the chain that produces every answer is consistent and traceable: agent reads a semantic view, and the semantic view reads tables. Snowflake itself recommends semantic views for Cortex Analyst because they carry native role-based access control and governance, not just a query shortcut.

That chain is exactly what you need to govern. If you cannot see which semantic views an agent reads, and which tables those views touch, you cannot answer the one question every data and compliance team eventually asks: what data can this AI agent actually see?

Scanning agents and semantic views into Dawiso

Dawiso now scans both Cortex agents and semantic views directly from Snowflake. You connect Snowflake once, and after the scan completes Dawiso builds the interactive data lineage automatically. The agent appears as a node, the semantic views it consumes feed into it, and those views connect down to the real database tables underneath.

The screenshot below shows a scanned HR_BUSINESS_PARTNER Cortex agent with the semantic views it consumes feeding into it on the left. In the full lineage those views connect onward to the tables they read. The detail panel carries the agent's scanned description, "HR analytics for headcount, payroll and performance," and its response instructions, "Be careful with sensitive data. Use aggregates, never name individuals unless explicitly asked." Every part of that, the agent, its instructions, its views, and the tables they read, sits in one lineage you can open and trace.

Because the lineage is generated, not hand-drawn, it stays accurate as the agent and its semantic views change. The link between an AI agent and the data it touches stops being tribal knowledge and becomes a governed asset.

One agent can read many semantic views

Most enterprises do not run a single agent. They run a set of "talk to your data" chatbots, often one per department: a sales agent, an HR agent, a customer care agent, and more. Each one is scoped to a segment of the business and reads its own set of semantic views.

The number of semantic views grows quickly once several departments build their own agents. That scale is the real governance problem. Without lineage, no one can confidently answer whether the customer care agent can reach payroll data, or which views would be affected if a table changes. With every agent and every semantic view scanned into one place, those questions become a lookup instead of an investigation.

Generating semantic views back via OSI

Scanning gives you visibility. The second half of the workflow goes the other way and creates new semantic views from Dawiso. This is possible because of the Open Semantic Interchange (OSI), an open initiative Snowflake announced in September 2025 alongside Salesforce, dbt Labs, BlackRock and RelationalAI. OSI defines a vendor-neutral standard for semantic models, so semantic definitions can move between tools instead of being locked into one platform.

A Dawiso data product already carries everything a semantic view needs: the business definitions, the glossary terms, and the mapping to the underlying data. Dawiso exports that data product through OSI and generates a new semantic view inside Snowflake. You then point a Cortex agent at that semantic view and tell it to use it for answering questions. The definitions the agent relies on come from a governed source rather than being rebuilt by hand in Snowflake.

Full governance, owned in Dawiso

A semantic view contains both data and business meaning. That makes it the natural place for governance to live, and it is why managing semantic views from Dawiso closes the loop. You scan in for visibility and generate out for enablement, and both directions run through one governed source of truth.

Because Dawiso already holds the business glossary terms next to the data mapping, generating a semantic view is not only a job for data engineers. A business owner who understands what "active customer" or "headcount" means can produce a governed semantic view without writing the definition twice. The result is AI enablement in Snowflake that starts from agreed business terms, with ownership, definitions, and lineage already in place.

Where Dawiso fits

Dawiso connects to Snowflake, scans your Cortex agents and semantic views into interactive lineage, and shows precisely which agent reads which data down to the table. When you want to enable a new agent, Dawiso generates a governed semantic view back into Snowflake through OSI, built from a data product that already carries your business terms.

That two-way connection is what turns Snowflake Cortex from a set of disconnected chatbots into a governed part of your data estate. Visibility on the way in, and AI-ready context on the way out, both managed from the Context Layer that grounds your AI governance.