Answers to the common questions. For the full API reference, see the docs; for anything else, email us.
From sign-up to your first optimized call in a couple of minutes.
Sign up at promptcrunch.dev/signup. You start with $5 of free trial credit and no card is required.
Your key looks like pc_live_... and is shown exactly once, at signup. For security we store only a hash of it, so it cannot be displayed again later.
Lost it? Open your dashboard and use Generate new key on the Account tab. The new key works immediately and the old one stops working; your stats, credit, and account are unaffected. Treat your key like a password.
Point your existing SDK at Prompt Crunch and add one header. Your provider key still goes to your provider as normal.
https://api.promptcrunch.devhttps://api.promptcrunch.dev/v1X-PromptCrunch-Key: pc_live_...That is the whole integration. See the full quickstart and examples for Python, Node, and curl.
Set ANTHROPIC_BASE_URL to https://api.promptcrunch.dev and add your X-PromptCrunch-Key via custom headers, then restart Claude Code. Step-by-step instructions are in the docs.
Note: Claude Code's Pro and Max subscription logins use OAuth tokens bound to anthropic.com and cannot be proxied. Prompt Crunch works with API-key billing.
Your dashboard's Overview and Usage tabs show requests, tokens saved, and dollars saved. They fill in after your first few calls. Every API response also carries the per-request numbers in the X-PromptCrunch-* headers.
Flat monthly pricing. No usage math.
Paid plans are a flat monthly subscription: one predictable price, no percentage-of-savings math. Every dollar the optimization saves is yours.
The Free tier includes a $5 trial credit and 100 optimized requests per day. Requests past the daily cap still work, they just aren't optimized. Current plan prices are on the pricing page.
New accounts get $5 of trial credit so you can try the proxy with no card on file. When it runs low you'll see a banner on the dashboard.
On the Billing tab, use Add credit for a one-time top-up, or Manage auto top-up to refill automatically when your balance reaches zero (requires a card on file).
Set a hard ceiling on the Billing tab to cap your optimized usage. Once you reach it, Prompt Crunch passes requests straight through to your provider, so your calls keep working.
Paid plans (Pro and Agency) are flat monthly subscriptions. Manage, change, or cancel your plan anytime through the Stripe billing portal, linked from the Billing tab. If you downgrade or cancel, any remaining credit balance is preserved.
Share one subscription across your team.
On the Agency plan, open the Team tab and invite teammates by email. Each teammate gets their own API key, and all usage is covered by the shared subscription.
Set an allowed email domain on your team (for example yourcompany.com). Anyone who signs in with a matching email sees a "Join team" prompt and can opt in, no individual invite needed.
A team runs on a single subscription. The Team tab shows each member's recent usage in one place, so the owner can see who is actually using the proxy.
The most common snags and how to clear them.
A 401 means either your X-PromptCrunch-Key is missing or wrong, or your provider auth header is missing. Check that your Prompt Crunch key starts with pc_live_ and that you're still sending your provider key (x-api-key for Anthropic, Authorization for OpenAI). See status codes and errors.
Usually the base URL wasn't actually changed, or (in Claude Code) an OAuth login is overriding your env vars. Run claude /logout, confirm your base URL points at api.promptcrunch.dev, then restart. More in the docs.
Check the x-promptcrunch-status response header. optimized means we reduced tokens; passthrough means we forwarded as-is (for example on tool-continuation turns to preserve prompt caching). If it reads optimized on every turn but you expected caching, your cache_control breakpoints may not be reaching us. Send a sample request's response headers to support and we'll dig in.
If optimization ever fails, Prompt Crunch falls back to forwarding your request unchanged. We never drop a request, so your app keeps working. The warning is informational; if it persists, let us know.
Prompt Crunch applies per-user rate limits on top of your provider's own limits. If you need higher limits for a production workload, get in touch.
What we do (and don't) keep.
No. Your message content is processed in memory and is not persisted to disk by default. For regulated workloads you can enable zero-retention mode. See the security page and privacy policy for the full picture.
Your provider key is forwarded to your provider on each request and discarded. We never store it. Your Prompt Crunch key is never stored in plaintext.
By default we hold a small encrypted optimization state in memory for up to an hour so repeat conversations don't reprocess from scratch. With zero-retention on, we keep nothing between requests. Toggle it on the Account tab.
Same answers, fewer input tokens.
Multi-turn conversations re-send the whole history on every call, so by turn 20 you're paying for the same context many times. Prompt Crunch reduces that redundant history before forwarding to your provider, while keeping recent turns and structured data (code, JSON, SQL) intact. You get back the same response, billed for fewer input tokens. There's a deeper walkthrough on how it works.
No, and they're complementary. Prompt caching saves money when you replay the same prefix across requests. Prompt Crunch saves money when the prefix grows over multiple turns. You can use both together.
Any model on the Anthropic or OpenAI APIs works. We've specifically tuned for the current Claude and GPT families; the docs list the benchmarked models. Not on the list? Try it anyway.