CachePilot — Eliminate duplicate AI API costs
● Infrastructure — one line change
70% of every AI API dollar
pays for identical work.
Your system prompt. Your few-shot examples. Your context window. Same bytes, same request, same response — charged at full rate every time. CachePilot sits inline, returns cached responses for exact duplicates, and forwards only unique requests to the provider.
Get API Key →
First week free. No credit card. Pay 20% of what you save.
cachepilot — proxy log
$ cachepilot proxy started — 3 active connections
─── Request log ───
✔ HIT gpt-4o $0.0000 0.3ms
✔ HIT claude-sonnet $0.0000 0.4ms
→ MISS gpt-4o $0.0023 1.2s cached
✔ HIT gpt-4o-mini $0.0000 0.2ms
✔ HIT claude-sonnet $0.0000 0.3ms
→ MISS gpt-4o $0.0023 1.1s cached
✔ HIT gpt-4o $0.0000 0.3ms
Saved today: $0.00<br>Your fee: $0.00<br>Cache rate: 71%
0%
duplicate rate (measured)
0ms
cache hit latency
cache layers
0%
our fee on savings
One line. No SDK. No config.
Change the base URL in your AI client. Every duplicate request stops costing you.
OpenAI client
Before<br>After
// Before: every request pays full provider price, duplicates included
const openai = new OpenAI({
apiKey: "sk-..."
});
// After: duplicates return from cache → $0 provider cost
const openai = new OpenAI({
baseURL: "https://olympic-veterinary-bathroom-viewpicture.trycloudflare.com/v1",
apiKey: "sk-cache-xxx-yyy"
});
Works with OpenAI, Anthropic, and OpenRouter — bring your own API key.
How the overhead is removed
You add the proxy. It finds and eliminates the duplicates. You pay for what's left.
Route through proxy
Change one string: baseURL points to CachePilot instead of the provider. Your code, SDK, and response format stay identical.
Duplicates never reach the provider
Each request is hashed and checked against our cache layers (L0 memory → L1 Redis → L1.5 normalizer → L3 prompt optimization). On hit: response returns in
You only pay for unique requests
We track what you would have paid the provider vs. what you paid through us. Our fee is 20% of that difference. No savings? No fee.
First 7 days free
No charge for any savings during your trial week. Cancel anytime.
Pricing: 20% of savings
A fraction of the overhead we eliminate. No subscription. No minimum.
20%
of calculated savings
$0 if you save $0
First 7 days free
Cancel anytime
Save $100/mo → pay $20
Save $1,000/mo → pay $200
Save $10,000/mo → pay $2,000
Your cost breakdown, in real time
Every request logged: what it would have cost, what it cost through us, and your accrued fee.
Recent requests
Open full dashboard →
Model<br>Cache<br>Savings<br>Fee
claude-sonnet-4HIT$0.0045$0.0009<br>gpt-4oHIT$0.0023$0.0005<br>claude-sonnet-4MISS$0.0000$0.0000<br>gpt-4o-miniHIT$0.0002$0.0000
FAQ
How does the free week work?
For the first 7 days after signup, your fee is waived on all savings. At the end of the trial, you're charged 20% of whatever you saved going forward. If you saved $0 during the trial, no charge applies after it either.
Do I have to change any code?
One line: replace baseURL with https://olympic-veterinary-bathroom-viewpicture.trycloudflare.com/v1. Same SDK, same authentication, same response format. Streaming works.
What if my app has no duplicate requests?
Then you pay $0 forever. We calculate savings per-request: baseline (what the provider would have charged) minus actual (what you paid through us). If every request is unique, you save nothing and owe nothing. Most apps have significant repetition — system prompts, few-shot collections, background batch jobs.
How is savings calculated?
We track each request at provider pricing (model, input tokens, output tokens). Cache hit = $0 provider cost. Cache miss = provider cost passed through. Savings = baseline − actual. Our fee = 20% of cumulative savings.
Which providers are supported?
OpenAI (GPT-4o, GPT-4, o1, o3), Anthropic (Claude Sonnet, Haiku), and any model on OpenRouter. You keep your own provider API keys — we never handle provider auth end-to-end.
Are my prompts stored?
No. We store SHA-256 hashes of message sequences for session detection — never raw content. Cache keys are tenant-isolated by customer. Send x-cachepilot-mode: passthrough to bypass caching entirely for sensitive requests.