One endpoint, every model your stack depends on
Point the OpenAI or Anthropic SDK at shuntio and switch models by changing a single field. Per-token billing from one shared balance — no subscription, no minimum spend.
- Wire protocols — OpenAI and Anthropic
- 2Wire protocols — OpenAI and Anthropic
- Key and balance across every model
- 1Key and balance across every model
- Billed on usage; an idle account costs nothing
- Per tokenBilled on usage; an idle account costs nothing
- Change base_url, keep your existing code
- Drop-inChange base_url, keep your existing code
Built for people who read the docs
No proprietary SDK and no abstraction layer — the same request shapes you already use, with billing you can audit line by line.
Both wire protocols, one key
POST /chat/completions speaks OpenAI; POST /messages speaks Anthropic. The same key and balance serve both, so a Claude-oriented client can reach models the OpenAI catalog does not carry.
Per-token billing, no tiers
Credit is deducted for the tokens you actually consume, with input and output counted separately. There is no plan to renew and no floor under the balance.
Every request on the record
The console log lists time, model, key, token counts and credit deducted for each call — enough to reconcile a bill without asking anyone.
Keys you can revoke
Create one key per project or environment. All of them draw from the same balance, and revoking one leaves the rest untouched.
Questions worth answering up front
Everything you need to know about the product and billing.
What is shuntio?
shuntio is an API gateway that provides a single OpenAI- and Anthropic-compatible endpoint over a shared model catalog. You keep your existing SDK and change the base_url.
Do I have to rewrite my client?
No. The request and response shapes match the official APIs. With the OpenAI SDK you change base_url and api_key; with the Anthropic SDK you set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN.
How is billing calculated?
Per token, with input and output counted separately. The exact deduction for every call is in the console request log.
Is there a minimum spend or a subscription?
No. Credit is deducted only for calls you make, and an idle account costs nothing.
Which models can I call?
The models page lists the current catalog with per-token rates. The same list is published as machine-readable JSON for scripts and AI agents.