How does the optimizer improve a prompt?

Each round it sends the current prompt to the model with instructions to critique it against your goal and return an improved rewrite. The rewrite becomes the input to the next round, so improvements compound.

Where does my API key go?

It is used only for the direct browser-to-provider request and is never stored or sent to us. You bring your own OpenAI or Anthropic key, and you pay your provider directly for the calls.

How many rounds should I run?

Two or three is usually enough — gains taper off quickly, and too many rounds can over-engineer a prompt or drift from your intent. Watch the critique each round and stop when it stops finding meaningful issues.

Why does it cost money to use?

It makes real LLM calls on your behalf, billed to your provider account. The cost per round is small (typically a fraction of a cent on a mini model) but it is not free because actual inference happens.

What makes a good improvement goal?

Be specific and measurable. "Make it shorter while keeping all constraints," "enforce JSON-only output," or "add fallback behavior for missing data" give the model a target. Vague goals like "make it better" produce vague edits.

Active Prompt Optimizer

Active prompt optimizer

Writing a great prompt is iterative, and the model itself is a useful critic. The active prompt optimizer sends your prompt to an LLM with instructions to critique and rewrite it toward a goal you set, then feeds the rewrite back in for another pass. Over a few rounds the prompt tightens up — clearer constraints, better structure, explicit fallbacks. You bring your own API key and run it directly from the browser.

How it works

You provide a prompt, an improvement goal, your API key, and a round count. Each round, the tool sends the current prompt plus your goal to the model and asks for two things: a short critique and an improved rewrite. The rewrite is parsed out and becomes the input to the next round, so each pass builds on the last. Every iteration is shown so you can read the reasoning and stop early once the prompt is good enough. The key is used only for the direct provider request and never stored.

Tips and notes

Set a measurable goal. “Enforce JSON-only output” beats “make it better.”
Two or three rounds. Returns diminish fast; more rounds risk over-engineering.
Read the critiques. They often reveal a flaw you can fix by hand faster than another round.
You pay your provider. Calls bill to your own account; cost per round is small on a mini model but real.