Gateway

One key for Claude, GPT, Grok and DeepSeek.

Point any OpenAI or Anthropic client at api.oneaiapikey.com. You pay per token from prepaid credits, at the rates on the models page, and every request shows up in your usage log.

Get API key Read the quickstart

OpenAI formatAnthropic format

curl https://api.oneaiapikey.com/v1/chat/completions \
  -H "Authorization: Bearer $RELAYKIT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6.1-sol",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

curl https://api.oneaiapikey.com/v1/messages \
  -H "x-api-key: $RELAYKIT_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 512,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Protocols: OpenAI + Anthropic
Models: 11 across 4 families
Lowest input rate: $0.24 / 1M tokens
Smallest top-up: $3

Request flow

What happens between your code and the model

The gateway is a thin, strict layer. It checks the key and its limits, forwards your request unchanged in meaning, streams the answer back and writes one line to your usage log. Nothing else sits in the path.

Every check runs before a token is spent. Billing uses the token counts the provider reports for that request.Pause motion

Protocols

Both request formats, same key

Claude models answer on both formats. GPT, Grok and DeepSeek models answer on the OpenAI format. Keep the SDK you already use and change two values: the base URL and the key.

Full API reference

Endpoints
Endpoint	Format	Used for
POST /v1/chat/completions	OpenAI	Chat with any model. Cursor, Cline, Kilo Code, OpenCode, Aider, Zed and the OpenAI SDKs.
POST /v1/responses	OpenAI	Responses API, used by Codex CLI with GPT models.
POST /v1/messages	Anthropic	Messages API for Claude Opus 5.5, Claude Sonnet 5, Claude Haiku 4.5 and Claude Sonnet 5.5. Claude Code and the Anthropic SDKs.
GET /v1/models	OpenAI	Lists the model ids your key is allowed to call.

Base URLs

OpenAI SDKs and OpenAI-compatible tools
https://api.oneaiapikey.com/v1
Include /v1.
Anthropic SDKs and Claude Code
https://api.oneaiapikey.com
No /v1: the client appends /v1/messages.

Authentication on both: Authorization: Bearer sk-…. The Anthropic route also accepts x-api-key.

Streaming

Tokens as they are written

Set "stream": true on either format and read text/event-stream. The SDKs parse it for you.
On the OpenAI format, stream_options.include_usage adds a final chunk with the token counts you are billed for.
Coding agents stream by default, which is why long agent turns do not time out.

When to stream, and when not to

PythonOn the wire

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.oneaiapikey.com/v1", api_key=os.environ["RELAYKIT_API_KEY"])

stream = client.chat.completions.create(
    model="claude-sonnet-5",
    messages=[{"role": "user", "content": "Explain SSE in one paragraph."}],
    stream=True,
    stream_options={"include_usage": True},  # token counts in the last chunk
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

data: {"choices":[{"index":0,"delta":{"role":"assistant","content":""}}]}

data: {"choices":[{"index":0,"delta":{"content":"Server-sent"}}]}

data: {"choices":[{"index":0,"delta":{"content":" events"}}]}

data: {"choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: {"choices":[],"usage":{"prompt_tokens":14,"completion_tokens":96,"total_tokens":110}}

data: [DONE]

Keys and limits

A key per project, each with its own fence

Create as many keys as you need in the console. Each one carries its own scope, so a mistake in one project stays in that project.

Scope is set per key, so a leaked key for one project cannot spend beyond its own limit, models or IP range.Pause motion

Allowed models: Restrict a key to the model ids it needs. Anything else is refused before it reaches a provider.
Spend limit: Cap how many credits a key can use. When it is spent, the key stops; your balance and other keys are untouched.
Expiry: Give short-lived keys to contractors, demos and CI jobs. After the date they fail with 401.
IP allow-list: Accept a key only from your servers’ addresses, so a copied key is useless elsewhere.

How to split keys by project

Billing

Prepaid credits, metered per token

No subscription required, no overage bill. At zero balance requests stop; top up and the same keys work again.
1 credit = $1 of usage at the published per-model rates. Packs start at 10 credits.
Every request is a line in your usage log: model, input and output tokens, cost.

Worked example: one Claude Sonnet 5 request at current rates ($2.20 in, $11 out per 1M tokens), starting from a 10-credit top-up.Pause motion

Credit packs Estimate a monthly bill

Reliability

What to expect, plainly

01
Provider incidents reach you.
When a provider is down or throttling a model, requests to that model fail with 429 or 5xx. We do not answer with a different model than the one you asked for.
02
Retry 429 and 503 with backoff.
The official OpenAI and Anthropic SDKs already do this. Requests the gateway rejects (invalid key, no balance, unknown model) are not charged.
03
Stream long outputs.
Proxies on the internet close connections that stay silent for about 100 seconds. A non-streaming request for a long answer can fail even while the model is still working; with stream: true tokens keep the connection alive.
04
No uptime percentage, on purpose.
We do not publish an uptime figure or an SLA until we have a measured history to back one. Current state and incidents are on the status page.
05
Models change.
Providers retire models on their own schedule. Removals and new ids are announced in Product Updates with dates.

System status Errors and retries

Try it from a $3 top-up.

Create an account, make a key, and send the request above. The first line in your usage log tells you exactly what it cost.

Get API key Set up a coding agent