Skip to content

Model Call Block

Model Call invokes Pioneer's shared model provider for each row or once for the whole batch.

Configuration

  • model records the requested model name;
  • role and prompt define the configured message content;
  • mode selects per-row or once-per-batch execution;
  • output_field names the row field receiving text;
  • temperature controls sampling when supplied; and
  • context and retrieval options bound optional supporting context.

Prompt templates can interpolate row fields. Treat prompts and input rows as sensitive content: configured block inspection reports bounded metadata for role and prompt rather than returning their full text.

Evidence boundary

Configuration proves the requested model/provider options. A successful block proves that the shared provider returned a response and the block recorded completion. Current durable run evidence does not persist the actual routed provider/model identity, so do not claim that identity from the configured name alone.

Model output is generated content, not automatically verified fact. Downstream deterministic checks should validate any contract that matters to the workflow.

See Models & Providers for installed, selectable, effective, and observed state.