Bring your own key
The gateway holds no model-provider credentials. Each request carries your own provider API key, which is forwarded for that one request and never stored or logged.
About
About Revl AI
Revl AI is an early-stage, founder-led project. It builds an open-source, bring-your-own-key gateway that sits between your application and model providers (Anthropic and OpenAI).
Why it exists
What the gateway is for
The problem
When an application calls a model API directly:
- A repeated, identical request is a new upstream call each time.
- Email addresses, secrets and other private values in a prompt reach the provider as written.
- Rate limits and retry settings are kept separately in each service.
The approach
A thin gateway that speaks each provider's native format. You keep using the provider's SDK with your own API key, and point its base URL at the gateway. Caching, masking and retries are then opt-in per request, each with a single header.
Principles
The rules the gateway is built on
Native wire formats
No translation layer. OpenAI-format requests go to OpenAI and Anthropic-format requests go to Anthropic, with the body unchanged unless you turn on masking.
Tenant isolation
Cache entries and rate limits are scoped to your gateway key (the tenant). A cache entry is also bound to a hash of your provider key, so a cached response is never served to a different provider account.
Private by default
Request and response bodies are never logged. The gateway logs one line of metadata per request, with no bodies, no headers and no key material.
API keys only
OAuth and consumer subscription tokens are rejected. The gateway is for your own pay-as-you-go provider API account, not for pooling or sharing subscriptions.
Open-source core
The gateway core is open source. Anyone can self-host it on their own Cloudflare account.
Bring your own key
How provider access works
- No resale of model access
- Revl AI does not resell, sublicense, pool or share model-provider API access.
- Your key, your provider account
- Every request is authenticated to Anthropic or OpenAI with your own API key, under your own provider account and billing, and under that provider's terms and usage policies.
- Keys pass through
- A provider key is forwarded for the single request it arrives with. It is never stored and never logged.
- API keys only
- OAuth tokens and consumer subscription tokens are rejected. Only provider API keys are accepted.
- Charges are for the gateway only
- Revl charges only for the gateway, never for model usage. Model usage stays on your own provider account and is billed by that provider.
Revl AI is an independent project. It is not affiliated with, endorsed by or sponsored by Anthropic or OpenAI.
Status
Where the project is today
Available now
In the open-source core today.
- BYOK passthrough for the OpenAI Chat Completions API and the Anthropic Messages API
- Gateway access keys
- Exact-match response caching
- PII and secret masking with rehydration, for non-streaming requests
- Retries
- Per-tenant rate limiting
- Streaming passthrough
- Metadata-only logging
Masking is pattern based. It will miss things and is not a compliance control by itself.
Roadmap
Not built yet, and in no committed order.
- Semantic caching
- Streaming cache and streaming rehydration
- Cross-provider fallback
- Usage dashboard and analytics
- Per-key budgets and per-key rate limits
- SSO and SCIM
- Custom masking detectors
Hosted gateway
The hosted gateway at https://api.proxyrevlvay.com is invite-only and is being rolled out to the waitlist. It is not generally available.
Self-hosting
You can self-host the core on your own Cloudflare account today. The source is published on GitHub so that you can read what the gateway does with a request before you run it.
Contact
Who is behind it
Revl AI is an early-stage, founder-led project. To get in touch, write to founder@proxyrevlvay.com.
Check it for yourself
Read how the gateway behaves, watch a simulated dashboard, or ask for access to the hosted gateway.