Skip to main content
Every recipe run is recorded automatically. Use recipe runs to see what happened, and recipe judge <run-id> to evaluate it.

Quick Start

1

Run a recipe

Every run is recorded automatically — no --save needed.
2

List past runs

3

Judge a run

Copy a Run ID from the table and evaluate it.

How It Works

Each run returns a RecipeResult, which is persisted to history before the result is returned.

Configuration

Persistence is on by default and controlled by one environment variable. Disable persistence:
Re-enable (default):

Common Patterns

Find and judge the last failure:
Filter by session for a JSON pipeline:
List runs with the Python API:

Best Practices

Leave PRAISONAI_RECIPE_HISTORY unset so every run is captured. Disable it only in ephemeral CI where run records add no value.
Start with praisonai recipe runs, copy a Run ID, then praisonai recipe judge <run-id>. This avoids guessing trace names.
Combine NAME, --status, and --session to narrow results instead of increasing --limit.
History captures Run IDs, statuses, and timings. Add --save to recipe run only when you need a full replay trace.

Run History CLI

Full recipe runs options and storage layout

Recipe Command

Top-level recipe commands

Recipe Workflow

Run → judge → apply cycle

LLM as Judge

Evaluate a run by its Run ID