Messages

The Anthropic-compatible endpoint. Every model on xKiro is reachable here, not only Claude.

POST/v1/messages

Base URL has no /v1

The Anthropic SDKs append /v1/messages themselves, so set baseURL: "https://api.xkiro.com". Adding /v1 yourself produces /v1/v1/messages and a 404.

curl https://api.xkiro.com/v1/messages \
  -H "x-api-key: $XKIRO_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-opus-5",
    "max_tokens": 1024,
    "system": "You are a concise assistant.",
    "messages": [
      { "role": "user", "content": "Name three uses for a paperclip." }
    ]
  }'

Request body#

Core fields

Field	Type	Description
`model`required	string	Full model ID including the vendor prefix.
`max_tokens`required	integer	Required by this dialect, unlike Chat Completions. Upper bound on generated tokens.
`messages`required	array	Alternating `user` and `assistant` turns. System instructions go in the top-level `system` field, not in this array.
`system`	string \| array	System prompt. A string, or an array of text blocks that xKiro joins in order.
`stream`	boolean	Emit server-sent events. See Streaming.
`temperature`	number	Randomness, 0–1 in this dialect.
`top_p`	number	Nucleus sampling.
`top_k`	integer	Sample from the K most likely tokens.
`stop_sequences`	string[]	Sequences that end generation.
`tools`	array	Tool definitions using Anthropic's `input_schema` shape. See Tool calling.
`tool_choice`	object	`{ "type": "auto" \| "any" \| "none" }` or `{ "type": "tool", "name": "..." }`.
`thinking`	object	Extended thinking. See Reasoning for the exact shapes and how they map to non-Anthropic models.

Response#

200 OK

{
  "id": "msg_01ABCdefGHIjklMNO",
  "type": "message",
  "role": "assistant",
  "model": "anthropic/claude-opus-5",
  "content": [
    { "type": "text", "text": "..." }
  ],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 24,
    "output_tokens": 88
  }
}

stop_reason

Field	Type	Description
`end_turn`	string	The model finished naturally.
`max_tokens`	string	Hit the max_tokens ceiling; the answer is truncated.
`stop_sequence`	string	One of your stop sequences matched.
`tool_use`	string	The model is calling a tool; run it and reply with a tool_result block.

Content blocks#

Responses are an array of blocks rather than a single string. Handle each type you care about and ignore the rest — new block types can appear over time.

text — visible output.
thinking— the model's reasoning, present when thinking is enabled.
tool_use — a tool call with id, name and input.

Counting tokens#

POST/v1/messages/count_tokens

Estimate the input size of a request before sending it — useful for staying inside a context window or predicting cost.

curl https://api.xkiro.com/v1/messages/count_tokens \
  -H "x-api-key: $XKIRO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-opus-5",
    "messages": [{ "role": "user", "content": "Hello there" }]
  }'

200 OK

{ "input_tokens": 11 }

Differences from Chat Completions#

Same capability, different spelling

Field	Type	Description
`system prompt`	—	Top-level `system` field here; a `system` role message in Chat Completions.
`max_tokens`	—	Required here, optional in Chat Completions.
`response`	—	`content` block array here; `choices[].message` in Chat Completions.
`usage`	—	`input_tokens`/`output_tokens` here; `prompt_tokens`/`completion_tokens` in Chat Completions.
`reasoning`	—	`thinking` object here; `reasoning_effort` string in Chat Completions. Both reach the same control.