Sluice routes every AI request to the cheapest provider that can serve it, fails over when one goes down, and itemizes what every call cost, so you can optimize for what you pay per successful task, not per token.
One base_url change · drop-in API · 1 free credit, no card
The smallest model in the right quality tier, only when you opt in with optimize: true.
Marketplaces race to the lowest price per token. But a cheap model that fails half the time costs you retries, fallbacks and engineer hours. Sluice counts all of it.
Illustrative example. Your numbers come from your own traffic and evals.
Each gate removes a different kind of waste. Together they decide where, and whether, a request runs. Gates marked LIVE work today; the rest are on our roadmap.
Same model, many providers. Sluice takes the lowest live price that's up, and fails over when it isn't.
The smallest model in the right quality tier, only when you opt in with optimize: true.
Repeated context and bloated instructions are trimmed before they're billed. Same answer, fewer tokens.
Identical repeats come back in milliseconds with no provider fee, and long repeated Claude prompts are cached automatically at a fraction of the price.
Not urgent? Send it to /v1/batches. It runs on the providers' batch capacity at half price and lands within 24 hours.
Every attempt, retry and cent is logged against the request it belongs to. Proof, not promises.
Real prices from our registry, simulated traffic. When a provider returns a 503, the request reroutes before your user notices.
Real prices from our registry. Traffic and outages here are simulated.
Keep your code and your OpenAI SDK. Swap one line.
Prepaid and transparent: provider price plus 25% of what we save you. No saving, no fee.
Every call is routed to the cheapest healthy provider and logged with what it cost and saved.
You pay the provider's price, plus 25% of whatever Sluice saves you against the direct price. If there's no saving, there's no fee, so Sluice is never more expensive than going direct. No subscription, no seats. Try it on your numbers:
You keep every month
$0.00
0% less · $0.00 a year
464k requests × 1k input + 350 output tokens, at list prices from our model registry (checked 2026-11-02). Our side includes our fee: 25% of the saving, nothing if there is none.
Point one service at Sluice, send real traffic, and read the receipt on every call before you switch the rest.