Prompt workbench

Turn rough intent into reliable prompts

Paste a prompt or describe what you need. Standard produces a fast optimized draft; Deep adds Agent iteration and independent absolute scoring.

Standard · fast optimization, no independent score

Start from a professional case

Click to prefill
01 — Two modes

Two modes, honestly labeled

Standard · exampleNo independent score

Standard

A fast first pass. The system aligns on your intent, applies the relevant methodology, and returns an improved prompt with its reasoning.

Clearly labeled no independent score.

Deep · example resultQuality threshold met

Deep

An Agent improvement cycle. The current optimized prompt is scored intrinsically, deficiencies feed the next round, and the run stops on convergence or a clear bound.

Intent alignment0.88
high confidence
LLM executability0.91
high confidence
Failure prevention0.82
moderate confidence
CONVERGED · overall 4.20/5 · hard gates passed

Standard reports optimization only. Deep reports its independent score and threshold decision.

02 — Who it's for

Built for people who put prompts to work

Align

Understands before it optimizes

The system aligns on your task, audience, and constraints first; every change traces back to your intent.

Trace

Reasoning you can audit

No black-box rewrite. Each edit comes with why it was made and what it should improve.

Score

Evidence over adjectives

Deep mode reports intrinsic quality scores, confidence, and the Agent iteration trail.

Intent alignment0.88
high confidence

round 2 of 3 · scorer=absolute-rubric

03 — Provenance · example

Every Deep result carries its trail

Scores, Agent revisions, model versions, and the final quality decision — all traceable through the Deep run.

cycle
construct → Agent improve → absolute score → feedback
scoring
absolute rubric · model=gpt-5.6-terra
rounds
2 of 3
decision
CONVERGED · overall 0.84
Pricing

Start free — 3 optimizations a day, no card required

Deep Agent optimization from $3.99/month.

Full pricing →