Your prompts stay yours

Route your traffic through PromptCrunch and cut the input bill by up to 75% on long conversations, without handing anyone your data. Your provider key passes straight to the model. We keep no copy of what you send, and no analytics vendor ever sees a word of it.

01 ·
You
Your app
https
02 ·
Us
PromptCrunch
https
03 ·
Provider
OpenAI / Anthropic

Three hops: your app, us, your provider. Your key and prompt move through in memory and go straight to the model in the Authorization header.

What we store, and what we don't

Never stored

  • Prompt content. Lives in memory for the request, then gone.
  • Response content. Streams straight back. No copy kept.
  • Provider API keys. Forwarded to the provider, never to disk.
  • User prompts in logs. We log metadata, not what you sent.

What we keep

  • Token counts. Needed to bill and to prove savings.
  • Model and status. For your dashboard.
  • Your account. Email, password, API key, Stripe billing.
  • Aggregate analytics. Daily rollups. No content.

No trackers. No data brokers. No exceptions.

+ api.anthropic.com your requests, your key
+ api.openai.com your requests, your key
+ api.stripe.com billing, only on paid plans

That's the entire outbound list. Your model provider and Stripe for billing, companies you already pay and trust. No ad pixels, no analytics SaaS, no data broker reading your prompts on the side.

Zero-retention mode

No prompt state on our servers, even encrypted.

No server state

Your state travels in the response

Optimization state is encrypted with a per-account key, signed, and returned in the payload. Your client echoes it back next request.

Signed & encrypted

You can't forge it, we can't read it

AES-256-GCM authenticated encryption with a key tied to your account. Invalid or tampered blobs are safely ignored, and the request is reprocessed from scratch.

One flag to flip

Toggle in your dashboard

No code change. Once on, we keep nothing between requests. Available on paid plans.

What happens to a prompt, exactly

Follow one prompt from your app to the model and back. Here is every place it goes, and every place it doesn't. Hand this section to your security team and let them check us on it.

Every request

Forwarded to your provider

Your prompt goes to OpenAI or Anthropic under your own API key, over HTTPS. We process it in memory on the way through.

During compression

Older history summarized by a fast model

On long conversations, older turns get condensed by a fast, low-cost model so you stop paying full price for history you already sent. It runs on Anthropic, the only outside party that ever sees a line of your prompt, and a name you already trust with your traffic.

At rest

Nothing at rest, unless you want it

No prompt content is ever written to disk. Token counts and metadata power your dashboard, and you can switch even that retention off whenever you like.

The architecture is the guarantee. Prompts never touch disk, zero-retention mode holds nothing between requests, and your security team gets the exact data path to audit line by line. Every control on this page is one you can verify yourself.

Start free today, or run it all on your own network

Most teams are live on our infrastructure in two lines of setup, with prompts that never hit our disk. Want everything inside your own walls instead? Self-hosted is on the way: a Docker image in your VPC, license-validated and offline-capable, so no prompt ever leaves your network. Get on the list and you'll be first in.

Try it free Join the self-hosted list