| Tool | What your agent uses it for |
|---|---|
list_semantic_functions | Find functions that already exist |
get_function_schema | Read a function’s parameters and output |
call_semantic_function | Try a function out while building |
create_semantic_function | Save a validated question as a reusable function |
get_sdk_info | Look up how to call functions from app code |
process_verification | Self-check the work before reporting done |
Discovering and inspecting
list_semantic_functions
Your agent lists the functions already defined for your account. Reuse comes first — most dashboards can be built largely from functions that already exist, so a good agent always checks before creating new ones.Steer it: “First check what semantic functions we already have for revenue before building anything new.”Filters (all optional):
search, category, limit (default 100, max 1000), offset.
get_function_schema
Reads one function’s full schema — its parameters (types, required/optional, defaults) and its output columns — so your agent knows exactly how to call it and what it returns.
Steer it: “Show me the parameters and output of _REVENUE_BY_REGION before you wire it up.”
Parameter: function_name (required).
call_semantic_function
Runs a function and shows the result while building, so you and your agent can confirm it returns what you expect.
Steer it: “Call _COMPLETION_BY_DEPARTMENT for R&D and show me the numbers so we can sanity-check it.”
Parameters: function_name (required), parameters (the filter values).
Creating
create_semantic_function
Turns a question your agent has already asked and verified into a reusable function, exposing the filters you asked for as parameters.Steer it: “That completion-by-department result looks right — save it as a function with department and date range as optional filters.”What your agent provides: the source question it validated, a name, a description, and the list of values to expose as filters. See semantic functions for what can and can’t be a filter (and the one-breakdown-per-function rule).
Building the app
get_sdk_info
Returns the SDK package name, install command, and usage examples so the code your agent generates calls functions correctly.You rarely need to steer this — your agent uses it automatically when writing the app. See the SDK reference for what it produces.
process_verification
Before your agent tells you “the dashboard is ready,” it runs this self-check — confirming it reused existing functions, used the right field names, added loading and error states, and kept the runtime code SDK-only (no build-time tools shipped into the app).Think of it as your agent’s pre-flight checklist. If something here fails, a good agent fixes it before reporting done — you can also ask “did you run the verification check?”