Prompt Optimizer

Let an LLM analyze your prompt across seven criteria, then rewrite it for clarity — while preserving every {{ variable }}.

Two phases: analyze, then improve

1
Analyze
The model scores your prompt (0–100) across seven criteria and explains its reasoning.
2
Improve
The model rewrites the prompt for clarity and effectiveness, keeping all {{ variables }} intact.

Accepting an improvement

Review the rewritten prompt. If you like it, accept it — for a prompt this creates a new version with a commit message prefixed AI Improvise -, and pushes the improved content back into the editor. The optimizer can also work on your system prompt.

  • If your prompt has no system prompt, tick “Generate a system prompt” and the model will author one for you.
  • Optimizer runs use your own provider key and are logged to usage with source="optimizer", so the spend shows up in Usage & Logs.
  • For chains, the optimizer runs per step and applies the result back to that step's draft.
Variables are always preserved
The improve step is constrained to keep every {{ variable }} placeholder, so a rewrite never silently breaks your template.