Skip to main content
A semantic function is a reusable, parameterized query created from a question you’ve already validated. Once created, you call it by name with parameters — from the SDK in your app, or from the Builder tools while developing. Because the query is built ahead of time, execution is deterministic and fast (typically under 500ms), with no LLM step at runtime. This is what makes semantic functions suitable for production dashboards and apps.

​
How it differs from Asker mode

Asker modeSemantic function
InputA natural language questionA function name + parameters
When the query is builtEvery time you askOnce, when the function is created
RuntimeJedify interprets and generates the queryRuns the pre-built query directly
ResultAnswer, chart, explanationRows of data

​
The lifecycle

1

Discover

Check what already exists with list_semantic_functions. Most dashboards can be built from existing functions — only create new ones when there’s a gap.
2

Create (if needed)

Validate a question in Asker mode, then turn that result into a function with create_semantic_function. You name the values you want to expose as parameters.
3

Call

Use the function by name — jedify.call('_FUNCTION_NAME', params) from the SDK at runtime, or call_semantic_function to test while developing.
See Examples for a full walkthrough, and the Builder tools reference for each tool.

​
What you can parameterize

When you create a function, you list the values to expose as parameters. You can parameterize values that appear in the query — for example:
  • A dimension you filter on (region, status, country, category)
  • A date or datetime range
  • A numeric threshold or range
  • A list of values (an IN list)
  • A boolean toggle
You cannot parameterize the structure of the query itself — table names, column names, which dimension you group by, or the SQL keywords. A function has one fixed shape with swappable value slots. If you need a different shape — a different grouping, a different table, or a filter the original question didn’t include — create a separate function from a question that includes it.

​
Parameter types

Each parameter has a type, which controls how its value is formatted into the query:
TypeExample value
string"active"
number100
booleantrue
date"2025-01-01"
datetime"2025-01-01T00:00:00Z"
array<string>["US", "Canada"]
array<number>[101, 102, 103]
Each parameter is either required or has a default taken from the original query, so a function runs even when you don’t pass every parameter.

​
Naming

Function names start with an underscore and use UPPER_CASE_WITH_UNDERSCORES — for example _REVENUE_BY_PRODUCT or _CHURN_BY_REGION.

​
Validation

When a function is created, Jedify validates it by running it once with its default parameters:
  • Active — the function ran successfully and is ready to call.
  • Failed validation — the function couldn’t run; it’s saved with the error but can’t be called until the issue is resolved.
Query results come back with UPPERCASE column names. Access fields as row.TOTAL_REVENUE, not row.total_revenue. See the SDK reference for details.