Two lines to switch
Change the base URL and the key. Every existing call keeps working — same request shape, same response shape, same streaming, same tool calling. Nothing to install and no wrapper to adopt.
Infrastructure for modern AI applications
One API for every model you need
Point the client you already use at one endpoint and every model you need is behind it — nothing to rewrite, nothing to install, no new SDK to learn. Tokens are bought up front, so your balance is the whole of what you can spend: no subscription, no card on file, no invoice at the end of the month, and no way to run up a bill you did not plan for.
- Two lines to switch
- Prepaid, never overspent
- Billed on real cost
Two lines changed
Claude CLI
export ANTHROPIC_BASE_URL="https://api.claudcli.com" # no /v1 here
export ANTHROPIC_AUTH_TOKEN="sk-your-key-here"
claudePrepaid, never overspent
Buy a block of tokens up front and watch it draw down in the dashboard. Your balance is a ceiling rather than a meter — nothing can take you past it, so a retry loop or a stuck agent cannot run up a bill overnight. No subscription, nothing to cancel, and no invoice that arrives after the fact.
Billed on what it cost to serve
A long prompt that repeats context the model has already seen is served far more cheaply than its size suggests, and that saving is passed on rather than kept. A large agent turn can meter tens of thousands of tokens and bill a fraction of them. Every request shows you both figures.
One account, not one per model
Every model a modern application needs, behind a single key and a single balance. No separate signup for each one, no second billing relationship to keep alive, and no integration work when you move from one to another.
One key
Create, label and revoke keys yourself, and use the same one against every model in the catalogue. Nothing to provision per vendor and nothing to rotate in several places at once.
One balance
A single prepaid balance covers everything you run. There is no reconciling several invoices at month end, and no vendor that can keep billing you after you stop using it.
One line to change model
The model is a string in the request. Moving a workload from a small model to a large one, or back when the traffic profile changes, is an edit — not another SDK, another key, and another integration tested from scratch.
Models
The catalogue below is live. Call GET /v1/models with your
key for the same list in machine-readable form.
| Model | Context tokens | Max output tokens |
|---|---|---|
|
Haiku 4.5
Haiku 4.5
|
200,000 | 64,000 |
|
Opus 5
Opus 5
|
1,000,000 | 128,000 |
|
Sonnet 5
Sonnet 5
|
1,000,000 | 128,000 |
How it works
-
Create an account
An email address and a password. No card, nothing to cancel.
-
Fund it
Choose the package that fits and add the tokens. Your balance is visible from the moment it lands, and it does not expire while your account is open.
-
Point your client at it
Swap the base URL and the key. Keys are yours to create, label and revoke from the dashboard.
The questions people ask first
Short answers, before you sign up for anything.
Do I have to change my code?
Two lines: the base URL and the key. The request and response shapes are unchanged, so streaming, tool calling and error handling all keep working as they are. There is nothing to install and no wrapper to adopt — the snippet at the top of this page is the whole migration.
Can this surprise me with a bill?
No, and that is the design rather than a setting. Your balance is a ceiling: nothing can take you past it, so a retry loop, a stuck agent or a batch job with a mistyped zero cannot become an invoice. There is no card on file and nothing recurring to cancel.
Is there a monthly fee or a minimum term?
Neither. You buy tokens once and spend them at your own pace. The balance does not expire while your account is open, and buying more later adds to it rather than replacing it.
What exactly am I charged for?
What the request cost to serve — not how large it was. A long prompt that repeats context the model has already seen is served far more cheaply than its size suggests, and that saving is passed on rather than kept. Every request is itemised on your usage page with both figures side by side.
Does my existing OpenAI client work?
Yes. Only base_url and api_key differ from
the defaults. The Claude CLI is served too, unmodified, with no model
to configure — the compatibility layer resolves the model it ships
with to the matching tier.
What if I need more than the largest package?
Ask. Anything larger, or an unusual shape of traffic, is priced on what you actually need rather than rounded up to the nearest tier — and the balance you already hold carries over untouched.
One price per package, paid up front
No monthly fee, no seat count, and no metering that keeps running while you are not looking. You buy tokens, you spend them at your own pace, and nothing is charged for a request that failed.
Start with a balance, not a subscription
An account takes a moment and needs no card. Every model on one endpoint, a balance you control, and itemised usage for every request — and nothing to cancel, because there is nothing recurring.