One endpoint. Both protocols.
Point any OpenAI or Anthropic client at the gateway and keep your model ids. Streaming, tool calls and vision pass through untouched — billing is metered from the provider's own usage block.
https://xrouters.xyz/v1https://xrouters.xyzanthropic uses the hostWire it up in one line.
The gateway speaks both OpenAI and Anthropic wire formats. Change the base URL, keep your SDK, keep your types. Everything below assumes a key from the dashboard.
Quickstart
Create a key under API keys, then pick whichever client you already use.
curl https://xrouters.xyz/v1/chat/completions \
-H "Authorization: Bearer $XROUTERS_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v3.2",
"messages": [{"role":"user","content":"hello"}],
"stream": true
}'Authentication
A single bearer token authenticates every endpoint. There is no project id, no organisation header, no second signature.
Authorization: Bearer xr_live_xxxxxxxxxxxxxxxxxxxx
Keys are hashed at rest. If you lose one, revoke it and create another — the original cannot be recovered, only replaced.
OpenAI protocol
Fully compatible with the Chat Completions shape, including stream, tools, temperature and max_tokens.
POST https://xrouters.xyz/v1/chat/completions
Content-Type: application/json
{
"model": "claude-sonnet-4.5",
"messages": [{ "role": "user", "content": "hi" }],
"max_tokens": 512
}Python SDK: base_url="https://xrouters.xyz/v1". Node SDK: baseURL: "https://xrouters.xyz/v1".
Anthropic protocol
The Messages API is served at the root path. Send x-api-key and an anthropic-version header, or reuse your bearer token.
POST https://xrouters.xyz/v1/messages
x-api-key: xr_live_xxxxxxxxxxxx
anthropic-version: 2023-06-01
{
"model": "claude-opus-4.1",
"max_tokens": 1024,
"messages": [{ "role": "user", "content": "hi" }]
}Streaming
Add "stream": true and read server-sent events as they arrive. The proxy forwards chunks unbuffered in both protocols, so first-token latency is the provider's own.
for await (const part of stream) {
process.stdout.write(part.choices[0]?.delta?.content ?? "");
}Usage & billing
Each response carries the provider's own usage block. Quota is debited from that block multiplied by the model's multiplier, and the request is written to your log with tokens, latency and status.
- 1× models debit one quota token per token used.
- 2× models debit two — the multiplier is printed in the catalog.
- Upstream failures are refunded: nothing is debited when no usage block is returned.
- Quota never expires and is shared across every key on the account.
Errors
| status | meaning | what to do |
|---|---|---|
| 401 | missing or revoked key | create a new key in the dashboard |
| 402 | quota exhausted | top up — the pool is shared per account |
| 404 | unknown model id | check the catalog for the exact id |
| 429 | rate limit for your package | back off and retry, or upgrade the package |
| 502 | upstream provider error | retry; nothing was debited |
| 503 | maintenance mode | wait for the switch to clear |
Rate limits
Limits are per account, counted in requests per minute, and scale with the package you bought. They apply to both protocols together — there is no separate budget per endpoint.
Need a key first?
Registration includes starter quota — no card needed to test.