Recovea API

Recovea is an OpenAI-compatible API (with a native Anthropic surface, and the OpenRouter long tail on /v1) that lets you see, govern, and prove your AI spend across every provider — one neutral, BYO-key endpoint, so you do more per dollar, attested. Keep your OpenAI-style code unchanged, point it at Recovea, and every request is metered against a frozen reference price list and written to an auditable cost ledger, fail-open. Cache/dedup savings are concierge-verified today; eval-gated quality verification and the verified-savings share are off · proof pending — shown as labeled estimates, never billed. They estimate; we settle.

Two lines change. Set your base URL to https://api.recovea.ai/v1 and your key to a Recovea key (rcv_live_…). Streaming, tools, and errors all behave exactly like the OpenAI API.

Providers

"OpenAI-compatible" describes the interface, not the backend. You write OpenAI-style requests against the /v1 surface and Recovea routes them by model id: bare OpenAI ids go native to OpenAI on your connected key, and vendor/model slug ids (exactly as listed by GET /v1/models) ride your OpenRouter key for the long tail — Gemini, Llama, Mistral, DeepSeek, and more. For Anthropic, point the Anthropic SDK at the /anthropic surface (below) — same key, same ledger, Anthropic's own wire format. Recovea takes no margin on provider spend.

Provider	Surface	Models	Status
OpenAI	`/v1`	GPT-4o, GPT-5-class, o-series (bare ids)	Live
Anthropic	`/anthropic`	Claude Opus / Sonnet / Haiku, native wire format	Live
OpenRouter long tail	`/v1`	Gemini, Llama, Mistral, DeepSeek + more, as `vendor/model` ids	Live
Google (native) · AWS Bedrock · Azure OpenAI	—	Native enterprise surfaces	Planned

On /v1, a bare OpenAI id passes through to OpenAI and a vendor/model slug routes via OpenRouter; a bare non-OpenAI id (claude-…, gemini-…) returns an honest 404 that names the slug to use. Routing a single request across providers — cascading to the cheapest model that still passes the quality gate — is planned, not yet live. Your provider keys stay yours: pasted into the dashboard once, encrypted at rest (AES-256-GCM; KMS envelopes planned), used pass-through.

Already on a provider's native SDK? You don't have to switch to the OpenAI SDK. Recovea also speaks Anthropic's protocol: point the Anthropic SDK's base URL at https://api.recovea.ai/anthropic and keep using it exactly as Anthropic documents it. We keep the OpenAI-compatible surface as the stable, recommended path, so most teams write it once and never touch a provider SDK again.

Why Recovea

Drop-in. Parity is 100% server-side: no SDK fork, no client code change. If your code works against OpenAI, it works against Recovea.
Net of quality. Any optimization that could change output is gated — it must pass a cross-family non-inferiority test on your real traffic before it serves a user. (The gate's shadow collection is live; its scoring engine is rolling out — see How it works.)
Provable. An auditable cost ledger — every basis labeled, baseline vs. realized, per route — that you can export and re-derive. Verified savings (and the share that rides them) are off · proof pending; today the ledger settles spend, not a savings claim. Finance can take the same view to a board — a self-select lane, not the headline.
Fail-open. If anything in our layer fails, your request flows straight to your provider on your own key — never a 5xx. Reverting is one line: point your base URL back.

Quickstart

# OpenAI via Recovea
from openai import OpenAI  # Recovea is OpenAI-compatible: any OpenAI client works

client = OpenAI(
    base_url="https://api.sandbox.recovea.ai/v1",  # point the client at Recovea
    api_key="rcv_live_…",                  # your Recovea key, not a provider key
)

resp = client.chat.completions.create(
    model="gpt-4o",  # OpenAI on the /v1 surface
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

That's the whole integration. Continue to the Quickstart for the env-var setup and a curl smoke test.

How it fits together

your app ──► api.recovea.ai/v1 ──► (govern · meter · attribute) ──► your provider
   │                                                                  ▲
   └──────────────── fail-open: revert base_url, traffic flows direct ┘

Your provider keys live in Recovea pass-through, encrypted at rest with AES-256-GCM (customer-managed KMS envelopes are planned). We take no margin on provider spend. The platform fee is flat; the verified-savings share is off · proof pending and bills nothing today.

Next steps

Quickstart: your first request in under five minutes
Authentication: Recovea API keys (rcv_live_ / rcv_test_)
Chat Completions: the core endpoint
Streaming: server-sent events, unchanged
How Recovea works: the gateway, the meter, and the cost ledger
Pricing: a flat platform fee; the verified-savings share is off · proof pending

Next Quickstart