Messages - airmux

POST /inf/v1/messages

The route always returns Messages-shaped output and errors, independent of the selected provider family.

Request

curl https://llm.example.com/inf/v1/messages \
  -H "Authorization: Bearer $AIRMUX_INFERENCE_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "anthropic/claude-sonnet-4-6",
    "max_tokens": 128,
    "system": "Be concise.",
    "messages": [{"role": "user", "content": "Say hello"}]
  }'

Fields

Field	Notes
`model`	Required catalog model ID
`messages`	User and assistant Messages content
`system`	String or supported text-block array
`max_tokens`	Required; minimum 1; maps to canonical `max_output_tokens`
`stream`	Boolean
`temperature`, `top_p`	Sampling fields
`stop_sequences`	Maps to the gateway’s internal stop sequences
`tools`, `tool_choice`	Tool definitions and selection
`thinking`	Reasoning type, token budget, and display settings
`output_config`	Reasoning effort and structured-output format

Other top-level fields, such as top_k, metadata, and service_tier, travel as provider extras when the selected profile accepts them.

Response

{
  "id": "019...",
  "type": "message",
  "role": "assistant",
  "model": "anthropic/claude-sonnet-4-6",
  "content": [{"type": "text", "text": "Hello!"}],
  "stop_reason": "end_turn",
  "usage": {"input_tokens": 8, "output_tokens": 2},
  "gateway": {"finish_reason": "stop", "adjustments": []}
}

Reasoning signatures are opaque provider values. Replay the complete thinking block unchanged on a later turn when the SDK includes it.

Streaming

The response uses named Messages events: message_start, ping, content block start/delta/stop events, message_delta, and message_stop. The terminal message_delta carries final usage and the additive gateway object. See Use an SDK for an Anthropic client example.