Prompt Library Deduplicator

How Prompt Library Deduplicator works

Press Load model once: your browser downloads Universal Sentence Encoder (MediaPipe) (13 MB) from this site and keeps it in its cache. Then add your text and press Analyze locally. The model runs on this device, so nothing you enter is sent to a server. Until the model is loaded, or if your device cannot run it, the tool falls back to a simple keyword check and labels the result as the fallback.

Universal Sentence Encoder (MediaPipe) turns each text into a list of numbers that captures its meaning. The similarity score compares the two whole texts, rescaled to 0-100 (this model rarely scores unrelated sentences below 0.6, so 0.6 maps to 0).

90 or more reads as a near-duplicate, 75-89 as the same intent reworded, 55-74 as related and anything lower as different. The tool also splits both texts into sentences and lists the ones with no close match (under 78) on the other side, which is usually where two prompts really differ.

Worked example

The model's output for the built-in sample, recorded on a desktop browser. Your run may word things slightly differently.

Prompt A: Summarize this support ticket, identify urgency, affected product area, customer sentiment, and recommended next reply. Prompt B: Read a customer support message and return priority, sentiment, product tag, and a concise response suggestion.

Similarity

79

Verdict

Same intent, reworded

Only in the first

Nothing distinct

Only in the second

Nothing distinct

The model

The files are served from this site, versioned so a cached copy is never stale. Nothing is fetched from a third party while you use the tool.

You can switch models in the tool. Also available: Gemma 4 E2B instruction-tuned (web build) (2.0 GB). The larger language model gives more thorough answers but is a bigger download and needs a stronger device.

Limitations

• A close match in meaning is not proof. A resume line that sounds like a requirement still has to be true, and a grounded-sounding sentence still has to be supported by the source.
• Very short items (one or two words) carry little meaning, so their scores are less reliable than full sentences.
• Results describe the input you gave, not the real-world situation behind it. Review them yourself before acting.

Good uses

• Triage AI operations notes before deeper review
• Prepare consistent labels for repeat workflows
• Reduce manual sorting on small batches
• Keep sensitive drafts in the browser preview