Agent Snowflake can choose a tool for SQL by semantic representation or search by documents. A useful answer preserves the origin of every fact.
One question has two sources.
Suppose the manager wants to know how many returns were last week and when they are checked by an employee. The number is taken from the table, the rule from the instruction. In Cortex Agents, you can connect separate tools for these two tasks. Cortex Analyst builds a query on a customized semantic view, and Cortex Search finds text fragments. When accepting, check which sources were used for each part of the answer.
Tutorial reconciliation with two versions of the instructions
Create a small returns dataset and two documents with different review deadlines. Identify the current version in a source the agent can access. Ask the question, save the SQL, and read the retrieved passage. Check the week boundaries, time zone, and treatment of duplicates. If the count is correct but the rule is outdated, score the two results separately. The data and retrieval may need different fixes.
A question without a source
In a separate run, remove access to the document or part of the table. The answer should be clear even when only a number or only a rule is available. Match the roles for each tool, as well as the rules for conveying context and accounting for work. When updating a semantic view or documents, repeat the check. So you can see which responses changed after the source update, and which changed along with the behavior of the agent. Write down the version of the rules and the moment of updating the index.
Suitable tasks
- Compare the number from the table with the condition from the instruction
- Check the mixed response from two sources
Strengths
- Supports Cortex Analyst and Cortex Search as separate tools
- There is an official setup guide and REST API
Limitations
- Prepare semantic views and the search index separately
- Results depend on permissions and whether the sources are current
Selection and conditions
- Free access
- Not confirmed
- Trial period
- Not confirmed
- No sign-up required
- Not confirmed
- Russian interface
- Not confirmed
- Local installation
- Not confirmed
- API
- Yes
“Not confirmed” means there is not enough information. The feature may still be available; check the developer’s terms.
- Input formats
- Data, Document, Text
- Result
- Text, Data
Terms
Charges depend on the Snowflake components you use, including models and compute. Verify the usage rules for your setup. Prepare semantic views and the search index separately.From task to first result
Practical example
- Task
- Compare the number from the table with the condition from the instruction
- What to prepare
- A training table of returns and two editions of the internal instructions.
Task brief or prompt
Show the number of returns by week and separately specify the current verification rule. Give me a source for each part.
Expected result
The answer is with the SQL being checked and a fragment of the current instruction found.
How to check the result
- Check the number with a control request.
- Open a fragment of the document and check its wording and rights.
An editorial teaching example of a task brief, not the result of hands-on testing of the service. Check that the required mode is available, that you have rights to the source material and that the intended use of the result is permitted.
Check details
Official sources
This description is based on the developer’s material. Features, prices and availability may change; check the primary source before use.