Anthropic compatibility

The Anthropic SDKs and Claude-style clients can call Noket through the Messages API. The routes are served at two equivalent paths, so you can pick the base URL your client expects to append to.

Send a message

curl https://api.noket.dev/anthropic/v1/messages \
  -H "x-api-key: YOUR_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "alice/llama-3.1-8b",
    "max_tokens": 256,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.noket.dev/anthropic",
    api_key="YOUR_API_KEY",
)

message = client.messages.create(
    model="alice/llama-3.1-8b",
    max_tokens=256,
    messages=[{"role": "user", "content": "Hello"}],
)
print(message.content[0].text)

A bearer token is also accepted where an Anthropic client sends x-api-key; if both headers are present, x-api-key wins. See Authentication.

Request handling

Noket requires model and a non-empty messages array, and forwards the rest of the body to the model unchanged: system, max_tokens, temperature, tools, stream, and the other Messages fields behave as the provider implements them. The same routing and credential fields are stripped as on the OpenAI routes, so a request cannot choose a provider or supply a key.

The response is the provider’s Messages response. Noket replaces the internal model id with the public id and removes internal metadata; usage keeps its token counts but not Noket’s cost.

Streaming

Send stream: true. Noket passes the Messages event stream through: message_start, content_block_delta, message_stop, and the rest. A model configured without streaming support returns 400 with stream_not_supported.

An error after the stream has started ends the stream; the HTTP status is already 200, so read the error event instead of the status code.

Count tokens

curl https://api.noket.dev/anthropic/v1/messages/count_tokens \
  -H "x-api-key: YOUR_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "alice/llama-3.1-8b",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Token counting runs no generation and costs nothing against a budget. The reply is the provider’s {"input_tokens": ...} body.

List models

curl https://api.noket.dev/anthropic/v1/models \
  -H "x-api-key: YOUR_API_KEY"

{
  "data": [
    {
      "type": "model",
      "id": "alice/llama-3.1-8b",
      "display_name": "Llama 3.1 8B",
      "created_at": "2025-03-04T09:12:00.000Z"
    }
  ],
  "has_more": false,
  "first_id": "alice/llama-3.1-8b",
  "last_id": "alice/llama-3.1-8b"
}

The catalog fits in one page, so has_more is always false. display_name is the model’s name in the console — the name its owner gave it, not the provider’s model name.

Errors

Anthropic routes use the Anthropic error envelope:

{
  "type": "error",
  "error": {
    "type": "rate_limit_error",
    "message": "Your traffic quota has been exceeded."
  }
}

See Errors and limits for the statuses and messages both dialects can return.

Next

• OpenAI compatibility
• Errors and limits
• Examples