Snowflake

Connect AI agents to Snowflake through the SQL API v2. Agents run parameterized SQL, load results into tables, poll long-running statements, and discover databases, schemas, and tables.

8 actions available

A scheduled flow needs the latest figures from the warehouse

Agent runs Execute SQL with bound parameters to compute the metric

If the statement outruns the synchronous window, agent polls Get Statement Results by handle

Agent reads the returned rows as plain objects keyed by column name

Agent posts the computed figures to the metrics channel via Slack

Agent loads the reconciled results back into a reporting table with Execute SQL

Any INSERT, MERGE, or DDL that writes to a production database pauses for approval

What This Integration Enables

Agents read analytics out of Snowflake and use the rows directly in flow logic, load records produced by other steps into warehouse tables with INSERT or MERGE, and discover the account structure before touching it. Statements are submitted synchronously; when one does not finish inside the API's synchronous window, Execute SQL returns a statement handle and the agent polls Get Statement Results until the rows arrive, or calls Cancel Statement to stop it. Large result sets are split into partitions the agent can page through, so a wide analytical query does not have to fit in a single response.

Without FlowRunner

Warehouse locked behind BI Numbers live in dashboards but cannot drive an automated workflow

Ad hoc export scripts Analysts export query results to spreadsheets to feed the next step by hand

Long queries block runs A slow statement stalls the whole automation with no way to poll or cancel

With FlowRunner

Warehouse in the flow Query results feed flow logic directly as a governed step

Results move themselves The agent computes a metric and routes it onward with no manual export

Async by design Long-running statements are polled by handle, and can be cancelled cleanly

Use Case Scenarios

Warehouse-to-channel reporting

On a schedule, an agent runs Execute SQL to compute the day's key figures against a virtual warehouse, reads the returned rows, and posts a formatted summary to a metrics channel with Slack. The team gets the numbers where they already work, and the figures come straight from the governed warehouse rather than a hand-built export.

Cross-system load into the warehouse

An agent pulls fresh records from an operational connector, then calls Execute SQL with an INSERT or MERGE to stage them into a Snowflake table. Because parameters are bound rather than interpolated, the load is safe against malformed values, and MERGE keeps the target table idempotent when the flow re-runs.

Anomaly detection with a human gate

An agent runs a query that surfaces an anomaly, such as a metric that crossed a control threshold. Before it writes a correction back into a production database, it does not act on its own. It routes the finding and the proposed write for review, and only runs the MERGE after a person confirms the change.

Human-in-Loop Highlight

Reading from a warehouse is safe. Writing to one, especially a MERGE or a DDL statement against a production database, is where a mistake becomes expensive and hard to reverse. FlowRunner's answer is human-in-the-loop, an execution pattern where the agent pauses on its own, assembles the context and the choices, routes to a human on their preferred channel, and resumes the moment they respond. When an Execute SQL statement would INSERT, MERGE, or run DDL against a database flagged as production, the agent pauses before submitting and asks through Slack: "Ready to MERGE 1,240 rows into `ANALYTICS.PROD.REVENUE`. Here is the statement and the source. Approve or cancel?" The write runs only after a person confirms, and the approver and timestamp are captured in the run log. A connector can run any SQL; an orchestration layer knows which statements should stop and ask first.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via Slack

Human decides

Agent resumes with decision

Agent Capabilities

8 actions

SQL

Execute SQL Runs any SQL statement (SELECT, INSERT, UPDATE, DELETE, MERGE, DDL, CALL, SHOW) with `?` placeholders bound via the Parameters array, and returns rows as plain objects plus the total row count and a statement handle. Use it for every read and write against the warehouse.
Get Statement Results Retrieves the status and results of a statement by handle, including individual partitions of a large result set. Use it to poll a long-running query that Execute SQL returned as in-progress.
Cancel Statement Cancels a running statement by handle, so an agent can abort a query that is taking too long or was started in error.

Metadata

List Databases Runs `SHOW DATABASES` for the databases visible to the configured role, for discovery before reading or writing.
List Schemas Runs `SHOW SCHEMAS IN DATABASE` for a database, listing the schemas an agent can query.
List Tables Runs `SHOW TABLES IN SCHEMA` with row counts and sizes, so an agent can find the right table before a query.
List Warehouses Runs `SHOW WAREHOUSES` with state and size, so an agent can confirm which virtual warehouse is available to provide compute.
Get Table Schema Runs `DESCRIBE TABLE` and returns column names, types, nullability, defaults, and key flags for a table.

Start building with Snowflake

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us