An editor for the prompts you send to AI.

Tokenlifter reads your prompt the way a good editor reads a draft: what is missing, what costs tokens for nothing, what will go wrong. You get a score out of 100 and the fix for each point.

Start free See pricing

A free quota that refreshes every four hours. No card needed.

37out of 100

Your prompt14 tokens

Write a 100-word summary of ~~this report~~ the attached Q3 sales report for our sales team.

Clear goal 10/25: Not what the summary is for. Say what the team should decide from it.
Context 2/25: “This report” is never given. Paste it, or its key figures.
Risk 2/15: With no figures, the model will invent some. Tell it what to do when a number is missing.

Then a review, line by line

The better version is written for you, with what each line costs. Five checks show what changed and why the score moved.

prompts/sales-summary.md

reviewed by Tokenlifter37 to 72 out of 100

1-Write a 100-word summary of this report for our sales team.14 tok

Context 2/25. “This report” is never given; the model will invent figures. Attach it or paste the key numbers.

1+Summarise the attached Q3 sales report in 100 words12 tok

2+for our account managers. Lead with the three9 tok

3+fastest-growing regions and the two products that9 tok

4+missed target, one number each. If a figure is10 tok

5+missing, say so instead of estimating.8 tok

Clear goalWhat the answer is for22/25
ContextWhat the model needs to know14/25
Constraints and formatLength, tone, shape17/20
EfficiencyNo tokens spent on nothing12/15
RiskWhere it can go wrong7/15

In the browser

Paste a prompt in the playground and read the review.

In your terminal

Review a folder or your last Claude Code prompts. Secrets are masked before anything leaves your machine.

npm install -g tokenlifter

Through the API

One call per prompt, with the token cost and what is left of your quota.

POST /api/v1/evaluate