Ship your first requestin one minute.

Get a key on the Get a key page, then point any OpenAI SDK at MuxKey:

1from openai import OpenAI
2
3client = OpenAI(base_url="https://www.muxkey.xyz/v1", api_key="mk-live-...")
4
5reply = client.chat.completions.create(
6    model="meta-llama/llama-3.3-70b-instruct",
7    messages=[{"role": "user", "content": "ping"}],
8)
9print(reply.choices[0].message.content)

That is the entire integration. Every endpoint below follows the OpenAI shape.

Send your key as a bearer token. Keys start with mk-live-. There are no accounts, so the key is the identity — keep it secret.

1Authorization: Bearer mk-live-...

Lost a key? Open your permanent dashboard link to rotate it. Without the dashboard link, there is no recovery.

Model IDs are provider/model, e.g. deepseek/deepseek-chat. Browse the full list with prices on the Models page, or list them programmatically:

1curl https://www.muxkey.xyz/v1/models -H "Authorization: Bearer mk-live-..."

Need the Anthropic format instead? POST /v1/messagesaccepts Anthropic Messages requests (that's how Claude Code connects) with the same key and billing.

Pass stream: true and read server-sent events, exactly as with OpenAI. Tokens arrive as they are generated.

1const stream = await client.chat.completions.create({
2  model: "deepseek/deepseek-chat",
3  messages: [{ role: "user", content: "Write a haiku" }],
4  stream: true,
5});
6for await (const chunk of stream) process.stdout.write(chunk.choices[0]?.delta?.content ?? "");

If an upstream provider errors or times out, MuxKey retries the same model on another provider. Where an equivalent model is configured it can fail over to it — the response's model field always tells you what actually answered.

Before a request runs, its worst-case cost is held from your balance; the unused part is returned the moment it finishes.

Credit is prepaid in crypto from $0.50 and never expires. You pay the provider's per-token price with a 0% routing fee. See Pricing.

Prompt and completion content is not stored by default. We keep token counts and timestamps per key for billing. Upstream providers apply their own data policies.