How billing works

Free tier

Every account gets 1,000,000 tokens processed free (counted as tokens we see before compression). No card required to start. When the limit is hit, API calls return 402 free_tier_exhausted until you activate billing in your dashboard. Activating charges nothing — it saves a card with Paddle for later usage fees, and we never see card numbers.

Add the card before the first million runs out and the free tier doubles to 2,000,000 tokens. No payment is taken and no subscription starts; fees only begin past the enlarged tier. Remove the card later and any bonus tokens you have not used go with it — the ones you already spent do not.

Hosted models with a known price

OpenAI, Anthropic, and other priced APIs. The fee is always a share of measured input-token dollar savings (cost_without_input_compression − cost_with_input_compression):

# Default (input compression only)
our_fee = 20% × input_dollars_saved

# When compress_output is true (input + output savings for you)
our_fee = 50% × input_dollars_saved

Example (input only): without CompresLM a call would cost $1.00; with CompresLM it costs $0.60. You saved $0.40. Our fee is $0.08. You keep $0.32.

Same savings with compress_output: true: our fee is $0.20 (50%). You still keep $0.20 of the input savings, and you additionally benefit from fewer completion tokens on the provider bill (those output savings are yours — we do not take a cut of output dollars).

How the dashboard counts output tokens saved

On calls with compress_output: true, the dashboard also shows output tokens saved. We take the completion tokens your provider actually billed for that call and scale them using the average reduction from our controlled tests (~60% fewer completion tokens vs a verbose baseline):

# Observed completion tokens = C (what the provider billed)
# Assumed remaining share after output compression ≈ 40%
would_be_completion = C / 0.40
output_tokens_saved = would_be_completion − C   # = C × 1.5

Example: a call returns 100 completion tokens with output compression on → dashboard credits 150 output tokens saved (as if the verbose answer would have been 250 tokens).

This figure is for your savings view only. CompresLM’s fee is not calculated from output tokens saved. Fees stay a share of measured input dollar savings (20% by default, 50% when compress_output is true) — see above.

Local / self-hosted LLMs

When there is no public token price (vLLM, Ollama, private endpoints), we cannot compute a dollar delta from the provider. Instead:

our_fee = $0.50 × (tokens_saved / 1,000,000)

This rate is shown in the dashboard and in the setup instructions whenever you choose a local upstream (X-LLM-API-Base).

What we never do

On-prem & custom plans

Want CompresLM inside your VPC, or a compression profile tuned to your traffic? We help set that up for free. Prefer a personal plan? Same form.

How payment works

Activate billing from your dashboard after the free tier. Activation is free — no payment is taken at that point. Paddle stores your card (we never see it) and usage fees start accruing from that moment.

• When your unpaid balance reaches $50, we charge the saved card automatically.
• At the end of each month we close any remaining balance of $1 or more.
• Invoices appear in your dashboard — download PDFs or pay manually if a charge fails.
• If an invoice stays unpaid for 3 days, API access is suspended until you pay.
• Fees are for usage already delivered, so they are not refundable — with defined exceptions for our own errors. See the Refund Policy.
• Prices here exclude tax. Where sales tax or VAT applies, Paddle adds it at checkout as merchant of record; we do not charge tax separately.

Two controls, both in the dashboard

• A monthly spending limit. Set a ceiling on our fee and calls return 402 spend_limit_reached once the month hits it. Off unless you set it. It caps our fee only — your provider bills you on your own key and we cannot see or cap that. And since a fee is measured after a request runs, the request that crosses the line still completes.
• Remove your card. One click, any time. Whatever balance has accrued is charged at that moment as a final invoice — after the card is gone there is nothing left to charge — and then it is removed. Your account stays open and your data stays put; you are back on the free tier.

Who you are actually paying

Paddle.com Market Limited is the merchant of record and the authorised reseller of this service. You buy from Paddle; Paddle buys from us. In practice: your statement shows Paddle, Paddle issues the tax-compliant invoice, Paddle determines and remits any VAT or sales tax for your country, and Paddle handles refunds. Card details are collected and held by Paddle and never reach us. Enter a VAT or tax ID at checkout if you have one, so the invoice is right.

If the service is down

You pay nothing. There is no subscription and no prepaid credit, so a fee can only come from a request we actually processed and measurably saved money on. No requests, no savings, no fee, nothing to refund. A request that errors, times out, or comes back uncompressed is free too.

Open your dashboard → Billing to activate, set a limit, or manage invoices.