Messages API (Anthropic wire dialect)

/api/client/v1/messages speaks Anthropic's Messages API request/response shape natively, alongside the OpenAI-shaped Chat Completions endpoint. Both run through the same model routing, guardrails, and provider gateway — pick whichever wire format your client already speaks.

New in September 2026

This endpoint shipped alongside the guardrail hook plane rewrite. If your integration currently targets Anthropic's API directly, pointing it at this endpoint instead gets you Console's model routing, guardrails, and tracing without a rewrite.

Endpoint

POST /api/client/v1/messages
Authorization: Bearer <token>
Content-Type: application/json

Request

{
  "model": "claude-sonnet",
  "system": "You are a helpful assistant.",
  "messages": [
    { "role": "user", "content": "What is the capital of France?" }
  ],
  "max_tokens": 1000,
  "temperature": 0.7,
  "stream": false
}

Field	Type	Required	Description
model	string	Yes	Model key configured in the dashboard
messages	array	Yes	Array of { role: "user" | "assistant", content } turns
system	string	No	System prompt, kept separate from messages as Anthropic's shape expects
max_tokens	number	Yes	Maximum tokens to generate
temperature	number	No	Sampling temperature (0-1)
stream	boolean	No	Enable SSE streaming (default: false)
tools	array	No	Tool definitions, in Anthropic's tool-use shape

content accepts either a plain string or an array of content blocks ({ type: "text", text: "..." }, plus tool-use/tool-result blocks when tools is set).

Response (Non-Streaming)

{
  "id": "msg_abc123",
  "type": "message",
  "role": "assistant",
  "model": "claude-sonnet",
  "content": [
    { "type": "text", "text": "The capital of France is Paris." }
  ],
  "stop_reason": "end_turn",
  "usage": {
    "input_tokens": 25,
    "output_tokens": 8
  }
}

Response (Streaming)

When stream: true, the response is a Server-Sent Events stream using Anthropic's event types (message_start, content_block_start, content_block_delta, content_block_stop, message_delta, message_stop) rather than Chat Completions' chat.completion.chunk shape.

Guardrails

Guardrails attached to the target model apply the same way they do on Chat Completions — the input.pre / output.pre hooks run regardless of which wire dialect the request arrived through. A blocked request returns the same guardrail_block error shape.

Errors

Status	Description
400	Missing model, messages, or max_tokens; guardrail block
401	Invalid API token
500	Model not found or not configured for chat completions
429	Rate limit or quota exceeded

Examples

cURL

curl -X POST https://gateway.example.com/api/client/v1/messages \
  -H "Authorization: Bearer cpeer_your_token" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet",
    "max_tokens": 1000,
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Python (Anthropic SDK)

Point Anthropic's own SDK at the gateway by overriding its base URL:

from anthropic import Anthropic

client = Anthropic(
    api_key="cpeer_your_token",
    base_url="https://gateway.example.com/api/client/v1",
)
response = client.messages.create(
    model="claude-sonnet",
    max_tokens=1000,
    messages=[{"role": "user", "content": "Hello"}],
)