Run a free scan
● Control first · proof when proven

Prove and govern your AI spend — do more per dollar, attested.

Recovea is an OpenAI-compatible gateway: change one line and route every provider through it to see your spend, set hard budget caps and a kill-switch, and attribute every dollar by key, team, and feature, in an auditable, basis-labeled cost ledger. Verified savings — net of quality, the 25% share — are off · proof pending until the eval gate ships.

See and govern your spendHow we measure

Or size it first: the estimator below is instant, ungated, and runs entirely in your browser. Verified savings settle later, on your own traffic — off · proof pending today.

OpenAI-compatible · one line—OpenAI · Anthropic · the OpenRouter long tail live today · coming soon: native Gemini · Bedrock · Azure
RECOVEA · COST LEDGERSample dataacct ████ · period 2026-05
leverbasisstatus
prompt-cache prefix hygieneplannedplanned
Batch API migrationplannedplanned
reasoning / effort trimplannedplanned
dedup / single-flightconciergeconcierge
exact cacheconciergeconcierge
model routing / cascade—proof pending
context / RAG trim (gateway)—proof pending
spend visibilitylive · every provider
cost ledger basislabeled · auditable
verified savingsoff · proof pending
The cost ledger is live and basis-labeled. Verified savings — net of quality, adjustments stripped — turn on only once the eval gate ships at volume; the 25% share stays off until then.

OpenAI-compatible · one line

OpenAIAnthropicOpenRoutercoming soonGoogle Gemini (native)AWS BedrockAzure OpenAI
Your estimate

Roughly what could be on the table?

Three inputs, an honest band, computed from cohort data per workload shape, not a sales fantasy. It’s a sizing estimate, not a settled saving — verified savings are off · proof pending and only settle on your own traffic once the eval gate ships.

savings · settled on your traffic · nothing leaves this pagev2026-06
$
What does most of your traffic look like?
How optimized are you already?

Fill in the three inputs to see what happens next.

No email. No signup. Computed in your browser.

Control is live today: one base_url change routes every provider through a fail-open gateway with cross-provider cost visibility and an auditable, basis-labeled cost ledger. Verified savings are proof pending — the eval-gated verification and the 25% savings share are off until the proof gate ships at volume. Concierge-verified cache/dedup savings (byte-identical, zero quality risk) are available hands-on, never billed as a live self-serve dollar. We don’t claim savings we can’t show you in the ledger, and we never claim "zero quality loss". We prove non-inferiority per route.

How it works

Change one line. See and govern every dollar; prove savings only when we can.

1

Point your base URL at Recovea

OpenAI, Anthropic, and the OpenRouter long tail today (more planned); your keys pass through. Reversible in one line. Fail-open to baseline: before the first token we fall back cleanly to your provider; mid-stream you get a clean error to retry.
2

We record your baseline

Every request is costed twice (what it cost, and what it would have cost untouched) against a frozen price list.
3

Control now, savings gated

Caps, kill-switch, and attribution are control you turn on. Cache and dedup savings are concierge-verified, never a self-serve billed dollar; routing and context slimming wait behind shadow tests on your real traffic.
4

The cost ledger is auditable

Spend attributed per key, team, and lever, basis-labeled. Verified savings — baseline minus realized, net of quality — are off · proof pending until the eval gate ships at volume.
the one-line change
one line in, one line out
from openai import OpenAI
client = OpenAI(
-     base_url="https://api.openai.com/v1",
+     base_url="https://api.recovea.ai/v1",  # your keys pass through · fail-open
    api_key=OPENAI_API_KEY,
)
risky levers climb a ladder
shadow1%5%20%50%100%

gate must hold at each rung · one-config rollback any time

The full mechanism
Proof, not promises

When we do claim a saving, it’s settled — not estimated

Verified savings are off · proof pending today. Quality is the gate, not a footnote, and the credibility move is what we refuse to claim: your traffic growth, provider price cuts, and your own prompt edits are stripped out of the number before we ever bill a cent of share.

per-route quality verdict
non-inferiority testSample data PASS
routesupport-classify
candidategpt-4o-mini· baseline gpt-4o
margin δ2.0 pts· observed +0.4
oraclegolden set + x-family judge
→ promote to 5% canary · rollback armed
what we strip out before billing
  • Your traffic growth: normalized, never billed as “savings”
  • Provider price cuts: they flow to you, via the frozen price list
  • Your own prompt edits: your win, not ours
  • Our own shadow-testing costs: our expense, removed from your numbers

Billing runs on cost-per-successful-output: a cheaper answer that failed its quality bar counts as zero saving. The cheap-but-worse loophole doesn't exist here.

The honest comparison

The optimizer is table stakes. The proof isn't.

Caching, routing, and batching ship for $0–500/mo across LiteLLM, Portkey, Helicone, Cloudflare AI Gateway, OpenRouter. We use some of them ourselves and we'll tell you so. What we found nobody else selling is the contract-grade proof that your bill dropped without your product getting worse, measured across every method, including tools you already run.

gateways optimizereducers cut tokensplatforms report spendRecovea governs spend; proves savings, net of quality
Read this first

When Recovea won't save you money

The fastest way to waste your time is for us to pretend everyone's a fit. These three aren't, and the scan will tell you which one you are before any money moves.

You spend under ~$15K/mo on inference

Our fee would likely exceed your savings, so don't pay us. Take the free scan and docs, and if you're growing, install the free Ledger: it watches at zero cost, and we'll tell you the day the math flips in your favor.

You've already built routing + evals in-house

If you run a custom router with a real eval harness, the remaining gap is usually thin. The scan will tell you honestly if it's not worth it.

An inline proxy is a hard no in your org

We sit in the request path (fail-open, your provider keys pass through; self-host is planned), but if that's a policy blocker, we're not a fit and we won't fight it.

Security & trust

We sit inline on your traffic. We treat that seriously.

  • Fail-open to baseline: on our outage, traffic falls back to your provider before the first token; mid-stream errors clean
  • Your provider keys pass through, encrypted at rest (AES-256-GCM), no token resale
  • No prompt bodies logged by default; metadata-only mode available
  • Per-tenant salted caches (planned, ships with the cache lever): designed so cross-tenant hits are impossible
  • Self-host in your VPC (planned) for trust-sensitive deployments
  • DPA, subprocessors, posture: /security
    • Objections, answered

    The questions a skeptical CTO asks

    See and govern your spend before you integrate anything.

    Start free in observe-only mode — no card. A spend scan on a sample of your logs comes back in 48 hours. No proxy, and an honest 'not worth it' if that's the truth.

    See and govern your spendBook a 20-minute call
    Auditable cost ledgerNon-inferiority per routeFail-open by designPass-through keys · no token resaleFree scan · 48h report