Every model, on your own keys.
A gateway that speaks OpenAI and Anthropic natively, running on the provider keys you already have.
Keys, names and limits, all in one place.
Three things to set up, once. After that the gateway is the only address your code needs to know.
Connect the providers you already pay for
Keys are encrypted at rest and stay yours. Every call is billed by the provider that answered it, at the rates you negotiated with them.
Give the models names your code can keep
Point smart, fast or cheap at whatever is winning this month, with a fallback behind it. Switching is a dropdown, not a deploy.
Put a limit over both
Per workspace, per key or per member. A quota that runs out pauses that scope until the next window. It never bills through.
Everything the gateway does for you.
The console holds the pieces that usually live in a config file nobody wants to own: keys, providers, aliases, limits, and the record of what actually happened.
Both wire formats
OpenAI and Anthropic requests answer at one address. Existing SDKs keep working after a base URL change.
Bring your own keys
Provider keys stay yours, encrypted at rest, and every token is billed by that provider at your rates.
Named models
Point smart, fast or cheap at whatever is winning this month. Your code never learns a model id.
Quotas that pause
Limits per workspace, key or member. When one runs out the work stops instead of billing through.
Workspaces
Separate environments, separate keys, separate limits, and members who only see what they should.
Usage you can audit
Every request timed and attributed: gateway overhead, which provider answered, and where quota went.
Start with your keys, keep the whole bill.
The gateway, the chat and the computer on one account. Free is enough to evaluate the gateway and the chat on your own keys, and your provider keeps billing the tokens exactly as it does today.