Context Layer for Snowflake & Horizon Catalog

A context layer for Snowflake is the governed layer of business meaning - definitions, relationships, lineage, and policy - that turns the data in your Snowflake account into something AI agents can understand and use trustworthily, not just query. Snowflake itself has invested heavily here with its Horizon Catalog, the platform's built-in governance and discovery layer, which Snowflake explicitly positions as a "governed context layer for AI, BI, and apps." Understanding what Horizon provides - and where a context layer needs to reach beyond a single platform - is key to grounding enterprise AI on Snowflake data.

It matters because Snowflake is where a huge share of enterprise data now lives, and it is increasingly where AI runs too, via Snowflake Cortex. But the same truth applies on Snowflake as everywhere: an AI model that can query a Snowflake table still does not know what its columns mean in business terms, whether the data is trustworthy, or how it relates to data elsewhere. A context layer supplies that - and the most useful one spans not just Snowflake but the whole estate the business actually runs on.

A Context Layer for Snowflake

Within Snowflake, a context layer means surrounding your tables and views with the metadata that gives them meaning and trust: business definitions for each metric and term, lineage showing how data flowed into each table, classification marking what is sensitive, and policies controlling who (and which agent) may use what. With that layer in place, an AI agent querying Snowflake can interpret the data correctly and respect governance - rather than guessing from column names and risking a confident wrong answer.

Snowflake has made this a first-class capability rather than something bolted on, which is why any discussion of a context layer for Snowflake has to start with Horizon Catalog.

Snowflake Horizon Catalog

Snowflake Horizon Catalog is Snowflake's built-in governance and discovery layer, spanning the account with capabilities that map closely onto what a context layer requires:

• Automated classification & protection. Horizon discovers and classifies columns and can detect, redact, and block PII and PHI from outputs - including from AI agent responses.
• End-to-end lineage. Native lineage so every result is traceable to its source.
• Data quality monitoring. Continuous monitoring so a query - human or AI - returns fresh, accurate data.
• AI & agent governance. Governance policies execute at the query engine layer, so they apply automatically to every caller - analyst, BI tool, or AI agent - with no separate AI configuration.
• Open-format reach. Catalog-Linked Databases sync Horizon governance with Apache Iceberg objects.

This is a genuinely strong native foundation - with one decisive qualifier: it governs the Snowflake perimeter. Everything Horizon does stops at the edge of the account. That boundary is exactly where the enterprise problem starts, because AI agents and the data they reason over almost never live entirely inside one platform.

Horizon, Cortex & Context

The reason Snowflake built Horizon into a context layer is Cortex - its suite of AI features, including Cortex Analyst and Cortex agents that answer natural-language questions over Snowflake data. These AI features are only as reliable as the context behind them, which is why governed semantics matter: Snowflake semantic views define metrics and relationships so Cortex Analyst turns plain-English questions into accurate, governed SQL. Horizon supplies the governance, semantic views supply the meaning, and together they ground Snowflake-native AI. The pattern is exactly the context-layer thesis: AI needs governed meaning, not just data.

The Gap: Beyond Snowflake

A typical enterprise runs 15-30 tools in its data stack, and each holds a piece of the context picture. Because Horizon governs only the Snowflake perimeter, four concrete gaps open up the moment your data - and your AI - reach beyond it:

• Governance doesn't propagate. A classification or policy applied in Horizon governs Snowflake objects - it does not automatically travel with the data into Tableau, Power BI, or Looker. The "restricted" tag stops at the account edge, so downstream tools lose the governance the moment data leaves Snowflake.
• Business context lives elsewhere. The institutional knowledge AI needs - definitions, decisions, caveats - sits in business glossaries, Confluence, Slack, and team wikis, not in Snowflake. A warehouse-native catalog has no way to pull that context in and govern it.
• Cross-platform lineage is shallow. Real pipelines run source systems → dbt → Snowflake → BI. Native cross-system lineage is limited and does not capture the column-level, transformation-aware path across the whole chain - exactly what impact analysis and AI trust depend on.
• Conflicting definitions & non-native agents. Metrics defined in Snowflake, in Databricks, and in BI tools drift apart; and AI agents that aren't Snowflake-native need governed context through an open standard - the Model Context Protocol - not a platform-specific interface.

These are not flaws in Horizon; they are the structural limits of any single-platform catalog. Left unaddressed they create context islands - the AI-era version of the data silos that plagued the 2000s - where each tool knows its own slice and no one, human or AI, can see the whole. The value of a context layer is precisely that it spans those islands.

How Dawiso Fits

Dawiso is the cross-platform context layer that closes exactly the four gaps above - with Snowflake as a first-class source within it, not a competitor to it. It connects to Snowflake alongside 40+ other platforms and unifies their context into one governed layer, so:

• Governance spans platforms. Classification, ownership, and policy live in one place and apply across the estate - they don't evaporate when data leaves Snowflake for a BI tool.
• Institutional knowledge is captured. The business glossary defines each term once - for Snowflake and non-Snowflake data alike - turning the context trapped in wikis and people's heads into governed knowledge.
• Lineage is end-to-end. Interactive data lineage traces the full path - source systems → dbt → Snowflake → BI - the cross-platform view a single-platform catalog cannot give.
• Cortex agents, governed both ways. Dawiso scans Snowflake Cortex agents and their semantic views into interactive lineage - so you can answer "what data can this AI agent actually see?" - and can generate governed semantic views back into Snowflake via the vendor-neutral Open Semantic Interchange (OSI) standard, keeping the definitions Cortex consumes owned and governed in Dawiso.
• Any agent, via open MCP. The Context Layer serves all of it to any AI agent through the open MCP Server - governed context spanning your entire estate, not just the Snowflake slice.

Horizon keeps governing Snowflake natively; Dawiso gives your AI the governed context that the rest of the business - and the rest of the stack - actually requires.

Conclusion

A context layer is what turns Snowflake data into something AI can use trustworthily, and Snowflake has built a strong native one in Horizon Catalog - governance, lineage, quality, and AI guardrails enforced right at the query engine, paired with Cortex and semantic views. For Snowflake data, that is a powerful governed context layer. The remaining step is scope: businesses run on more than one platform, and AI agents run in more than one place. The complete context layer unifies Snowflake with the rest of the estate and serves governed context to any agent through open MCP - so AI understands your whole business, not just the part that lives in Snowflake.