Messages

Request#

curl https://api.ahurasense.com/v1/messages \
  -H "x-api-key: $AHURA_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-5",
    "max_tokens": 256,
    "system": "You are a terse assistant.",
    "messages": [
      {"role": "user", "content": "Explain rate limiting in two sentences."}
    ]
  }'

modelstringrequired

A catalog id. Any chat model works, not only Anthropic’s.

max_tokensintegerrequired

Upper bound on generated tokens, reasoning included on models that reason.

messagesarrayrequired

Alternating user and assistant turns. content is a string or an array of blocks: text, image (base64 source), tool_use and tool_result.

systemstring or arrayoptional

The system prompt, as a string or text blocks.

streambooleanoptionaldefault false

Anthropic-style event stream. See below.

temperaturenumber, 0 to 1optional

Anthropic’s range. Mapped onto the model unchanged.

top_pnumber, 0 to 1optional

Nucleus sampling.

top_kintegeroptional

Passed through where the model supports it.

stop_sequencesarray of stringsoptional

Sequences at which generation stops.

toolsarrayoptional

Accepted but not translated, so tool calling does not work on this route. See below.

tool_choiceobjectoptional

Accepted but not translated. See below.

metadata.user_idstringoptional

Your identifier for the end user.

Response#

A native Anthropic message object:

{
  "id": "msg_…",
  "type": "message",
  "role": "assistant",
  "model": "anthropic/claude-sonnet-5",
  "content": [{ "type": "text", "text": "Rate limiting caps how many requests …" }],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 31,
    "output_tokens": 48,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0
  }
}

Errors use Anthropic’s envelope, {"type":"error","error":{"type":"…","message":"…"}}, with the same status codes as the rest of the API. X-Ahura-Request-Id is on every response.

Streaming#

With stream: true the response is a stream of named events in Anthropic’s order:

The Anthropic SDKs’ streaming helpers consume this without changes.

Differences from Anthropic’s API#

• Tool use does not work on this route. Anthropic tool definitions are forwarded without being converted to the wire format the backends expect: hosted models reject the request with 400, and partner models answer as though no tool were offered. A tool call the model does make is not returned as a tool_use block either. Use Chat completions for anything agentic; tool calling there is complete, on every model.
• thinking and output_config.effort are honoured on hosted models: {"type": "disabled"} turns reasoning off, budget_tokens caps it, and effort picks a level. See Reasoning effort. On partner-served models these fields are not forwarded.
• cache_control hints on content blocks are accepted and ignored.
• Images must be base64 sources; URL sources are not translated.
• Container and code-execution tools are not available.

For new integrations, Chat completions is the fuller endpoint. This route exists so that existing Anthropic code can move without a rewrite.