Snowflake-native text-to-SQL REST API that answers only through a semantic view's logical tables, metrics, join paths and verified example queries.
Platform and application teams already standardised on Snowflake who want natural-language querying as an API call rather than another seat-based BI tool - you call it from your own app, a Slack bot or Cortex Agents, and nothing leaves the account on the default model path. It is the right shape when you have, or will build, semantic views: logical tables, dimensions, facts, metrics, relationships and predefined join paths are what the model is restricted to. It also suits teams who want a regression harness for free, since you can run a semantic view's verified queries against Cortex Analyst and measure SQL accuracy to detect drift.
Snowflake-only. There is no path to BigQuery, Databricks, Postgres or anything else, so this is a lock-in decision, not a tool choice. The billing is awkward in both directions: the standalone Analyst API is 67 Platform Credits per 1,000 messages and is labelled 'Platform Credit (legacy)' on Snowflake's own pricing page, which recommends invoking through Cortex Agents instead, where you pay AI Credits per token at $2.00 per AI Credit for global routing or $2.20 for regional - and on either path, executing the generated SQL incurs standard virtual warehouse compute charges on top. The documented limitations are real product gaps: it has no access to results from previous SQL queries so it cannot refer back to an earlier result set, it answers only what SQL can resolve rather than producing insight or trend narrative, and long or intent-shifting conversations degrade to the point the docs tell you to reset. And the 90%+ accuracy claim is Snowflake's own internal benchmark - the engineering blog states neither the suite size nor the scoring rule, while a companion post describes replacing standard execution accuracy with a more lenient column-matching threshold judged in part by humans.
Pros
Cons
Yes, but it is internal. Snowflake's engineering blog states more than 90%+ SQL accuracy on real-world use cases and close to 2x single-shot GPT-4o, and a separate post reports 92% accuracy on standard star and snowflake schema joins versus around 73% for competing products and 55% for single-shot GPT-4o. Neither post states the suite size, and competitors are unnamed.
By the semantic view. It defines logical tables, dimensions, facts, metrics and relationships with predefined join paths, descriptions and synonyms, and the docs state Cortex Analyst uses only that metadata for SQL generation. Verified queries - sample questions paired with their correct SQL - guide generation, and you can re-run a view's verified queries against Cortex Analyst to measure SQL accuracy and catch regressions.
Two meters. The standalone Analyst API is 67 Platform Credits per 1,000 messages, which Snowflake's pricing page labels Platform Credit (legacy). Via Cortex Agents, the path Snowflake recommends, you pay AI Credits per token at $2.00 per AI Credit for global routing or $2.20 for regional. On both paths, executing the generated SQL incurs standard virtual warehouse compute charges on top.
Four documented limits. It has no access to results from previous SQL queries, so it cannot refer back to an earlier result set such as a product list it just returned. It answers only questions SQL can resolve, not general insights or trend analysis. Long conversations or frequent shifts in intent make follow-ups hard to interpret, and the docs advise resetting. And it requires Snowflake.
Facts sourced from the vendor’s own site, last verified 2026-10-10.
The Weekly Source
One email a week. New listings, repriced tools, and the ones we would skip - including this one if its pricing changes.
