async call(tool_id: str, parameters: Dict[str, Any], search_id: str | None = None, session_id: str | None = None, max_response_size: int | None = None, respond_with: str | None = None, compatibility_mode: Literal['strict', 'legacy_optional_fields'] = 'strict', timeout: float | None = None, correlation_id: str | None = None, model: str | None = None, sub_user_id: str | None = None) → ToolExecutionResponse

Call a specific capability.

• Parameters:
  • tool_id – Tool identifier returned by discover(…).
  • parameters – JSON-serializable parameters for the tool.
  • search_id – Search ID returned by discover(…) (recommended for traceability).
  • session_id – Optional correlation id.
  • max_response_size – Optional auto-delivery inline limit measured in UTF-8 bytes. -1 is unlimited; explicit respond_with=”full” takes precedence over a finite value.
  • respond_with – Optional server-side projection. Omit for compatibility auto-delivery; full forces complete inline data, while summary and fields:<JSONPath,…> select compact projections. Hard-limit failures use response_too_large.
  • compatibility_mode – Strict mode never resubmits a paid call. The deprecated legacy mode may replay once without an unsupported optional field.
  • timeout – HTTP request timeout in seconds; credential acquisition is separate.
  • correlation_id – Non-sensitive reference forwarded only to the credential provider.
  • model – Model that selected and parameterized this capability call.
  • sub_user_id – End-user identity for provider OAuth; use the same value as Probe.
• Returns: ToolExecutionResponse with success, result, and metadata.