compresh

Simple, transparent pricing

Pay only for what you save. One line to integrate. No hidden fees.

Starter
$0 service fee

For free or local models. Top up budget when you need it.

  • No card on file — one-time top-ups
  • $10 minimum to start
    (pay $7.50 with 25% discount)
  • Full EMA compression engine
  • Injection detection (3-layer, 19 languages)
  • No savings-share on free or local models
  • Use a paid model? Savings share applies (30%). Budget runs to $0 → service stops, top up to resume.
  • Min $5 balance required for free-model access
  • 25% discount on every top-up
Get started
Most popular
Pro
Subscription plans

For production apps using paid LLM models

  • Card on file — monthly metered billing
  • Everything in Starter
  • Fixed subscription + lower savings share on paid models
  • Longer commitment = lower rate
Plan Price Savings
share
Quarterly $18 / 3 mo 20%
Semi-annual $33 / 6 mo 16%
Annual $60 / yr 12%

Heavy paid-model use? Pro pays itself off — savings share drops from 30% (Starter) down to 12% (Annual).

Start saving

Credit top-ups

All top-ups include a permanent 25% discount.
Pay less, get full credit.

$10
credit
Pay $7.50
$25
credit
Pay $18.75
$50
credit
Pay $37.50

Every plan includes

Depth-aware compression

EMA extracts semantic tags as conversations deepen. Turn 10+ sees ~80% savings.

Injection detection

3-layer defense: regex patterns, heuristic analysis, DeBERTa ML model. 19 languages.

Zero code changes

Change one line (base_url). Streaming, tools, function calling — all work unchanged.

Multi-provider support

OpenAI, Anthropic, OpenRouter (200+ models). One API key for all.

Encrypted at rest

Provider keys encrypted with Fernet (SHA-256 derived). No conversation content stored.

<2% quality variance

Benchmarked across 1000+ tests. Compression preserves meaning, not just tokens.

Early adopter bonus

First 100 users get $30 free credit

No deposit required. Full access to all features. Use it on any model — free or paid.

Claim your spot

Questions? Check the FAQ or reach out.