UsageTap Compress · Standalone API

Compress prompts. Cut LLM costs.

Use Compress by itself, with no UsageTap metered events, or pair it with UsageTap for usage-based billing and higher throughput.

25M input tokens free, then pay only for tokens removed →

30 requests/minute on free access · safeguard waived with any paid UsageTap plan

Before1,248 tokens
After782 tokens
Selective compression−37% tokens

Default modelTry the current compression model here. Tenant-specific fine-tuning is planned.

Savings calculator

Only pay when you save.

Process your first 25M input tokens free each month. After that, pay $0.30 per 1M tokens removed, billed at month end. There is no usage charge when no tokens are removed.

Throughput: free Compress access is limited to 30 requests/minute. Any paid UsageTap plan removes this safeguard.
1M1B
Expected reduction

Your input

25M

tokens

Removed

8.75M

tokens

Sent to GPT 5.6

16.25M

tokens

Billable removed

0M

after free tier

Original GPT 5.6 input cost
$125
Input cost after compression
$81
Estimated input-cost savings
$44
UsageTap Compress fee $0.30 per 1M billable tokens removed
$0
Estimated net savings after UsageTap Compress
$44

Estimate based on GPT 5.6 input pricing of $5/M tokens and a 35% reduction. Your first 25M input tokens each month are free. Model selection affects only the estimated downstream savings, not the UsageTap Compress price. Usage charges are billed monthly.

SDK wrappers or direct API

Add compression to the call you already make.

Wrap your OpenAI or Anthropic client, or call the compression endpoint directly from JavaScript or Python.

Control the strength: wrappers accept aggressiveness by role; the direct text API accepts one value from 0 to 1. Assistant history is skipped by default.
const response = await fetch(
  "https://compress.usagetap.com/compress",
  {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.USAGETAP_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      text: longPrompt,
      aggressiveness: 0.5,
    }),
  },
);

if (!response.ok) throw new Error("Compression failed");

const {
  compressed_text,
  original_tokens,
  compressed_tokens,
  reduction,
} = await response.json();

Compression decisions

Compress only when it helps.

UsageTap caches repeats, skips prompts that are already efficient, and compresses longer context when the expected savings are worth it.

Repeated prompt

Use cached result

Small or compact

Skip

JSON or HTML

Compact structure

Long context

Compress

Protected or code-heavy

Preserve

See what happened on every call.

Original and compressed tokens
Savings and estimated cost avoided
Cache, skip, or compress decision
Latency, warnings, and transformations

Pay only when tokens are removed

Start with 25M input tokens free each month.

Then pay $0.30 per 1M tokens removed, billed at month end. Skipped inputs, zero savings, and failed compression with original fallback are not charged.

Free access includes a 30 requests/minute safeguard. Pair Compress with any paid UsageTap plan to remove that limit.

Estimate your savings →
Tenant-specific tuning · Planned

Fine-tune a compression model for your prompts.

The planned tuning workflow will use approved samples to adapt compression to your terminology and formats. This capability is not yet generally available.

Planned workflow

Approved promptsEvaluateFine-tune

More

useful compression

Fewer

unintended removals

Questions about selective compression

Can I use Compress by itself?+

Yes. Compress is standalone. You do not need a UsageTap plan or need to send UsageTap metered events. Use UsageTap for usage-based billing, Compress for prompt compression, or both together.

How does UsageTap Compress charge?+

Your first 25 million input tokens processed each month are free. After that, UsageTap Compress costs $0.30 per 1 million input tokens removed and is billed at the end of the month.

Is there a rate limit?+

Free Compress access is limited to 30 requests per minute as a safeguard. The limit is waived when Compress is used with any paid UsageTap plan.

Is compression lossless?+

No. Compression is conservative and inspectable, but it can change text. Protect critical spans and evaluate representative prompts before production rollout.

What happens when compression is skipped?+

The original input is returned and there is no compression usage charge.

Does it add latency?+

It can. Cached results and fast skip decisions avoid unnecessary work; deeper compression is reserved for inputs where expected savings justify latency.

Is raw prompt content sent to marketing analytics?+

No. Marketing events record only product interaction metadata, never prompt content.