Introducing Recovea 1.0
We're launching Recovea 1.0. It's an OpenAI-compatible gateway for the AI traffic you already run. You point your SDK at one URL, keep your own provider keys, and get one place to see, govern, and prove your AI spend across every provider — so you can do more per dollar, attested. The part nobody else does: every request comes back with its receipt, recorded in an auditable cost ledger you own.
Most cost tools estimate. They show you a dashboard number you have to take on faith, and that number gets discounted the moment a finance team or an auditor asks "how do you know?" Recovea is built the other way around. Every request is metered against a fixed reference price list at the moment we serve it, and the result is written to an append-only, hash-chained cost ledger that lives in your account. That auditable basis is what a verified-savings claim will stand on once the proof gate ships. They estimate; we settle.
What Recovea 1.0 does
You change one line, the base URL, and keep everything else: your SDK, your prompts, and your own provider keys. Recovea works with the models you already use. OpenAI and Anthropic are native, and the OpenRouter long tail (Gemini, Llama, Mistral, DeepSeek, and hundreds more) is live on day one. Native Google, Bedrock, and Azure surfaces are on the way, and until each one ships we say so plainly rather than imply support we don't have yet.
Recovea sits in front of your provider and meters every request, giving you cross-provider cost visibility and an auditable basis for every dollar. Nothing is stored that you didn't send. If anything is ever uncertain, the request passes straight through to your provider untouched. We fail open, never closed.
What you get back that you don't get anywhere else is the receipt: a per-request record of what the call cost against a fixed reference, accumulated into a cost ledger you can export as CSV or JSON, with every row carrying the hash chain you can re-derive from genesis yourself, and hand to your finance team without a footnote.
Where the levers fit, by workload
Cost levers depend on the shape of your traffic, and they're the same across every provider we support. The table below is honest about what's live today, what's on the roadmap, and how each one is verified. We'd rather show you less and prove all of it than quote a headline we can't put in your ledger.
| Workload | Live today | On the roadmap (quality-gated) | How it's verified |
|---|---|---|---|
| Repeated or cacheable (RAG, classification, retries) | Exact cache and request dedup (byte-identical, concierge-verified) | More cache and routing levers | Identical calls — a saving we can prove without quality risk; concierge-verified at launch, not a self-serve billed dollar |
| Batch or high-duplication | Request dedup (concierge-verified) | Batch routing | Each call metered against a fixed reference; the verified-savings basis is proof pending |
| Reasoning or long-output | Not yet | Output-effort tuning | A lever reaches your traffic only after it clears non-inferiority on your routes (the eval gate is planned) |
| Streaming chat or live tool calls | Not eligible, by design | Not yet | We don't claim savings here, because we can't yet show them — we sell control and visibility instead |
That last row is deliberate. Streaming and live tool calls don't take our current levers, so we say so: on the page, in the product, and in the ledger. A number we can't settle isn't a number we'll quote.
What you control today, and what you'll be able to prove
Control is what you get on day one: one endpoint across every provider, real-time cross-provider cost visibility, and an auditable cost ledger you can export and re-derive. A few things shape how much a future saving would be worth:
- How repetitive your traffic is. The more identical or near-identical calls you make (retries, shared system prompts, classification over similar inputs, hot RAG queries), the more exact cache and dedup have to work with.
- How much of your spend is cacheable rather than live. Streaming chat and live tool calls don't take today's levers; batchable and repeated calls do. The mix matters more than the total.
- Your model and context size. On bigger, pricier calls, each avoided or deduplicated request is worth more.
- What you've already optimized. If you've hand-built caching, there's less for us to add. If you haven't, there's more.
Rather than guess at these for you, the free scan reads your own traffic in your own tenant and shows you your spend and where the gap is, before you commit to anything.
Proof is the product
Optimization is a commodity. Proof is the product. The ledger is where it lives: every request carries its own basis, captured the instant we serve it, computed against a fixed reference price list. Entries are not back-filled later, not modeled from an average, and not adjustable after the fact. The ledger is append-only and hash-chained, so entries can be added but never quietly edited, and anyone you choose can check that the chain is intact. The full method (how the baseline is frozen, how each lever is attributed, and how the chain is built) is documented in How Recovea works.
This matters because estimates get discounted in diligence and traceable numbers don't. A cost figure you can trace, line by line, to the traffic that produced it is worth more to a finance team than a bigger number they have to trust. That traceability is the whole point of Recovea, and it's why we'd rather under-promise on the headline and over-deliver on the proof. It's the same measurement-and-verification discipline performance contractors use to prove energy savings: a frozen baseline, an auditable trail, not an estimate.
Quality is the gate
When a savings lever does reach your traffic, it will only do so after it clears a quality bar on your own routes. We don't claim "zero quality loss," because that isn't a claim anyone can honestly make. The plan is to prove non-inferiority, route by route, so a lever that can't clear the bar simply doesn't run — and any saving we count is net of quality by construction. That eval gate is planned, not yet live, which is exactly why the verified-savings claim (and the share that would ride it) is off · proof pending today.
What's live today, and what's next
We're being precise about the difference between live and designed. Today, what's live is control and visibility: one cross-provider endpoint, real-time spend, and an auditable cost ledger. Byte-identical exact cache and request dedup are available concierge-verified — a saving we can prove without quality risk — not yet a self-serve billed dollar. The eval-gated levers and the verified-savings share are off · proof pending, and they turn on only once the proof gate is sound. We're early, and we'd rather tell you that plainly than inflate a launch number.
The roadmap lights up the Standard control tier (budget caps, kill-switch, spend alerts, attribution) right after launch, then widens the levers we can verify, then hardens the proof itself, so a Recovea number stands on its own in an audit. We'll write about each piece as it ships, and only once it ships.
Availability
Recovea 1.0 is available today. Bring your own key for OpenAI, Anthropic, or the OpenRouter long tail, point one URL, and you're running. Start free: the Free tier meters your traffic in your own tenant, at any spend and with no card, and gives you cross-provider spend visibility. Professional ($249/mo) adds the in-path gateway, the auditable cost ledger, and concierge-verified cache/dedup. A control tier (Standard, $79/mo) is coming soon; Scale & Enterprise is a conversation — talk to us. The verified-savings share is off · proof pending, and it never bills before the ledger can stand behind it.
What's live today is cross-provider cost visibility and an auditable cost ledger. Exact cache and request dedup are concierge-verified, byte-identical, and not a self-serve billed dollar. Eval-gated quality verification and the verified-savings share are off · proof pending. Every request is metered against a fixed reference list in an append-only ledger you own. Results vary. We measure yours.