Skip to content
Rapid Router
Enterprise

Every model, one balance, and the controls to hand it to a whole team.

Budgets your apps cannot spend past, keys your backend mints on its own, and an invoice on every top-up.

    
Fee, charged once at top-up
500
Requests per batch job, 24-hour window
1, 7, 30
Day retention, or store nothing
USD
Billed in dollars, no local tax added
Spend control

A budget nobody can code around

prod-web$500 / month
batch-jobsstopped at its cap
dev$50 / month

Caps are enforced in the gateway, so a loop in one service cannot spend the budget of another. The account carries its own daily, weekly and monthly ceilings on top.

Per key and per account

Ceilings that reset daily, weekly, monthly or never.

Anomaly alerts

An hour that spikes against your own seven-day baseline names the model and key, with a link to kill it.

Signed webhooks

Budget, limit, balance and anomaly events, HMAC-signed. Email and WhatsApp too.

Keys at scale

One key per customer, without a human

sk-rr-mgmt
your backend
customer-041$10 / mo
customer-042$10 / mo
customer-043revoked
customer-044$25 / mo

Your backend mints a capped key per customer and revokes it when they leave. The management key cannot call a model, so it is safe to keep on a server that only administers.

Model allowlists

Restrict a key to named models or a whole provider, and deny models account-wide.

Presets

Model, fallbacks and prompt saved under a name. Change it in the dashboard, every service follows, no deploy.

Teams

Owner, admin and member roles, checked on the server for every route.

Reliability

One provider having a bad day is not your outage

your request
one model id
Anthropic503
Bedrockanswered, billed here
Vertexstandby

A provider that starts failing is taken out of routing until a probe says otherwise, and you are billed at the route that actually answered, not the one you asked for.

Routing you control

Pin, exclude or order providers, or sort by price, latency or throughput.

Health from real traffic

Providers are judged by your requests, not a synthetic ping.

Published uptime

Hourly availability per model for the last 72 hours.

Data

We keep the numbers, not the words

Off by default

Prompts and replies are not stored unless you switch capture on, per key.

Retention enforced

Captured bodies are deleted after 1, 7 or 30 days by a sweep, not by policy.

Redaction first

Email, PAN, Aadhaar, cards and mobile numbers stripped before the request leaves.

Keys as hashes

We store a hash, never the key. Provider keys are encrypted at rest.

Cost

Ways the bill gets smaller

Free tier

Sponsored :free models from a daily pool, plus free requests every day on any free-eligible model with :free appended.

Caching

Prompt caching passed through, plus an opt-in exact and semantic response cache.

Bring your own keys

Tokens bill your provider account; we charge — on what it would have cost here.

Billing

Billing that your finance team accepts

A numbered dollar invoice

Every top-up gets an invoice naming your company, billed in US dollars as an export of service: zero-rated, so no tax is added.

Usage exports

Daily CSV by key, model, provider and status, up to 92 days.

What we do not have yet

You would find these during an evaluation, so here they are first.

SAML, OIDC and SCIM
Sign-in is Google, GitHub or password.
Data-residency pinning
You pin providers, not their region.
An audit-log screen
Recorded and readable over the API only.
Distributed rate limits
60 rpm per key by default, raised on request.
SOC 2 or ISO 27001
Said plainly on the Security page too.

Tell us your volume

We will size committed-use credits to it, and tell you honestly where Rapid Router is a fit and where it is not yet.

Talk to us