Workflow planner

Turn a task into a plan and a prompt.

Options and examples

Preference applies to coding presets. Other categories use guidance-based settings.

Local suggestions. No model calls. Review the result.

How the suggestions work

Coding settings come from the saved DeepSWE results. Balance chooses the cheapest setting within 3 percentage points of the highest score; lower cost uses 10. Highest score selects the top published result. These are my thresholds, not OpenAI recommendations.

Other categories use starting suggestions informed by official guidance. Keyword matching can be wrong; choose the category yourself. Coding considers Sol, Luna and Astra. It does not predict your task's accuracy, speed or cost.

Test settings on the same inputs. Measure cost and time per accepted result, including retries and review. Keep the lightest setting that meets your quality bar. Use xhigh or max when your own evaluations justify them.