Running Workflows in PromptPlan: Run Modes, Step Execution, and Output Export

Before a workflow runs

PromptPlan validates required workflow inputs before starting execution.

If validation passes, PromptPlan:

1. snapshots the workflow definition
2. snapshots the run inputs
3. creates a workflow run record
4. executes enabled steps in order

Step execution behavior

For each enabled step, PromptPlan:

1. renders the prompt template
2. creates a step run record
3. calls the provider adapter
4. saves the output immediately
5. makes the output available to later steps

If one step fails, downstream steps stop. Completed earlier steps stay saved.

Run modes

The workflow UI exposes four run modes:

• fast
• balanced
• premium
• private

Current MVP behavior:

• fast: routes unpinned steps to the free OpenRouter router
• balanced: routes lighter transform/export-style steps to free routing when no model is pinned, while heavier reasoning/generation steps use the workflow default model policy
• premium: uses the account default model or explicit step model without the free-step downgrade behavior
• private: reserved for future private/local execution; today it acts as a policy placeholder and generally follows the same server-side model path as the default or pinned model

Step-level model overrides

If a step pins a specific model, that explicit step model wins over the run mode default.

This is useful when:

• one reasoning step needs a stronger model
• one transform step should stay cheap
• one export step needs predictable formatting

What you can inspect after a run

After the run finishes, you can inspect:

• saved run inputs
• per-step status
• rendered prompt
• output text
• error details
• total tokens when available
• final output

Export options

The run view supports:

• markdown export
• plain text export
• JSON export of the run snapshot and step runs

Use JSON export when you want a complete audit trail of how a workflow produced its result.