> ## Documentation Index
> Fetch the complete documentation index at: https://docs.haitoken.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Claude format
> Create Claude messages compatible with the Anthropic Messages protocol, supporting streaming and non-streaming responses
This endpoint is compatible with the Anthropic Claude Messages protocol for creating conversation messages.
## Features
* Supports both streaming (SSE) and non-streaming response modes
* Supports multimodal content (text, images, etc.)
* Supports tool use
* Supports thinking/reasoning mode configuration
* Supports cache control optimization
## Use cases
Suitable for scenarios requiring interaction with Claude models, including:
* Single-turn or multi-turn conversations
* Structured output (JSON Schema)
* Tool calling and function execution
* Code execution and container reuse
## Authentication
Include the `Authorization` header in the format `Bearer YOUR_API_KEY`.
The Claude format endpoint also requires the `anthropic-version` header, e.g. `2023-06-01`.
## Quick example
```python Python theme={null}
import requests
url = "https://api.haitoken.ai/v1/messages"
headers = {
"Authorization": "Bearer YOUR_API_KEY",
"Content-Type": "application/json",
"anthropic-version": "2023-06-01"
}
data = {
"model": "claude-sonnet-4-0",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Hello, please introduce yourself"}
]
}
response = requests.post(url, json=data, headers=headers)
print(response.json())
```
```javascript Node.js theme={null}
const response = await fetch('https://api.haitoken.ai/v1/messages', {
method: 'POST',
headers: {
'Authorization': 'Bearer YOUR_API_KEY',
'Content-Type': 'application/json',
'anthropic-version': '2023-06-01'
},
body: JSON.stringify({
model: 'claude-sonnet-4-0',
max_tokens: 1024,
messages: [
{ role: 'user', content: 'Hello, please introduce yourself' }
]
})
});
const data = await response.json();
console.log(data);
```
```curl cURL theme={null}
curl -X POST 'https://api.haitoken.ai/v1/messages' \
-H 'Authorization: Bearer YOUR_API_KEY' \
-H 'Content-Type: application/json' \
-H 'anthropic-version: 2023-06-01' \
-d '{
"model": "claude-sonnet-4-0",
"max_tokens": 1024,
"messages": [
{"role": "user", "content": "Hello, please introduce yourself"}
]
}'
```
## OpenAPI
````yaml en/api-reference/chat/claude/openapi.json POST /v1/messages
openapi: 3.0.1
info:
title: Default Module
description: ''
version: 1.0.0
servers: []
security: []
tags: []
paths:
/v1/messages:
post:
tags: []
summary: Create Message
description: Supports both streaming (SSE) and non-streaming response modes
parameters:
- name: anthropic-version
in: header
description: ''
required: true
schema:
type: string
example: '2023-06-01'
- name: X-Api-Key
in: header
description: ''
required: true
schema:
type: string
example: YOUR_API_KEY
requestBody:
content:
application/json:
schema:
$ref: '#/components/schemas/ClaudeMessageRequest'
description: ''
examples: {}
responses:
'200':
description: ''
content:
application/json:
schema:
$ref: '#/components/schemas/ClaudeMessageResponse'
description: >-
Claude Messages API non-streaming response
Reference:
https://platform.claude.com/docs/en/api/messages/create
headers: {}
deprecated: false
security: []
components:
schemas:
ClaudeMessageRequest:
type: object
properties:
max_tokens:
type: integer
description: Maximum number of output tokens (required)
top_p:
type: number
description: Nucleus sampling, 0-1
top_k:
type: integer
description: Top-K sampling
stop_sequences:
type: array
items:
type: string
description: Stop sequences
tool_choice:
type: object
properties: {}
description: >-
Tool choice configuration, can be a String or ClaudeToolChoice
object
cache_control:
$ref: '#/components/schemas/ClaudeCacheControl'
description: >-
Top-level cache control configuration
Automatically applies a cache_control marker to the last cacheable
block in the request
inference_geo:
type: string
description: >-
Inference geographic region
Specifies the geographic region for inference processing. If not
specified, uses the workspace's default_inference_geo
output_config:
$ref: '#/components/schemas/ClaudeOutputConfig'
description: >-
Output configuration
Configures model output format options, such as effort level and
structured output format
service_tier:
type: string
description: >-
Service tier
Determines whether to use priority capacity (if available) or
standard capacity
Possible values: "auto" / "standard_only"
model:
type: string
description: Model name, e.g. claude-sonnet-4-0, claude-opus-4-0, etc.
messages:
type: array
items:
$ref: '#/components/schemas/ClaudeMessage'
description: >-
Claude message object
The content field can be a String or List (multimodal content
blocks)
description: List of messages
system:
type: object
properties: {}
description: System instruction, can be a String or List
temperature:
type: number
description: Sampling temperature, 0.0-1.0
stream:
type: boolean
description: Whether to enable streaming output
thinking:
$ref: '#/components/schemas/ClaudeThinkingConfig'
description: Thinking/reasoning mode configuration
tools:
type: array
items:
$ref: '#/components/schemas/ClaudeTool'
description: |-
Claude tool definition (Tool / ToolUnion)
Uses input_schema instead of OpenAI's function.parameters
Reference: https://platform.claude.com/docs/en/api/messages/create
description: List of tool definitions
metadata:
$ref: '#/components/schemas/ClaudeMetadata'
description: Metadata
container:
type: string
description: >-
Container identifier, used for cross-request container reuse (code
execution tool)
required:
- max_tokens
- model
- messages
ClaudeMessageResponse:
type: object
properties:
stop_reason:
type: string
description: >-
Stop reason: end_turn / max_tokens / stop_sequence / tool_use /
pause_turn / refusal
Always output in the message_start event (even when null)
stop_sequence:
type: string
description: |-
Stop sequence that was matched (can be null)
Always output in the message_start event (even when null)
stop_details:
$ref: '#/components/schemas/ClaudeStopDetails'
description: >-
Refusal stop details
Structured information returned when the model refuses due to safety
policy
id:
type: string
description: Message ID (msg_ prefix)
type:
type: string
description: Object type, always "message"
role:
type: string
description: Role, always "assistant"
content:
type: array
items:
$ref: '#/components/schemas/ClaudeContentBlock'
description: >-
Claude response polymorphic content block
Uses Jackson @JsonTypeInfo to distinguish subtypes by the type
field
description: List of response content blocks
model:
type: string
description: Model name (the actual model that completed the task)
container:
$ref: '#/components/schemas/ClaudeContainer'
description: >-
Container information
Container identifier for the code execution tool, used for
cross-request reuse
usage:
$ref: '#/components/schemas/ClaudeUsage'
description: Usage statistics
ClaudeCacheControl:
type: object
properties:
type:
type: string
description: Cache type, always "ephemeral"
ttl:
type: string
description: |-
Cache time-to-live (TTL)
Possible values: "5m" (5 minutes) or "1h" (1 hour)
Default: "5m"
ClaudeOutputConfig:
type: object
properties:
effort:
type: string
description: |-
Reasoning effort level
Possible values: "low" / "medium" / "high" / "xhigh" / "max"
format:
$ref: '#/components/schemas/JsonOutputFormat'
description: |-
JSON output format configuration (structured output)
Used to specify the JSON Schema for Claude's output
ClaudeMessage:
type: object
properties:
role:
type: string
description: 'Message role: "user", "assistant", or "system"'
enum:
- USER
- ASSISTANT
- SYSTEM
content:
type: object
properties: {}
description: >-
Message content, can be a String or List of polymorphic content
blocks
ClaudeThinkingConfig:
type: object
properties:
budget_tokens:
type: integer
description: Thinking budget token count (required only when type="enabled")
type:
type: string
description: 'Thinking mode type: enabled / disabled / adaptive'
display:
type: string
description: >-
Controls how thinking content is displayed in the response:
"summarized" or "omitted"
When set to summarized, thinking is restored to normal. When set to
omitted, thinking content is redacted but a signature is returned
for multi-turn continuity. Defaults to summarized.
ClaudeTool:
type: object
properties:
input_schema:
type: array
items:
type: array
items:
type: array
items:
type: object
properties: {}
description: Tool input parameter definition in JSON Schema format
allowed_callers:
type: array
items:
type: string
description: >-
List of allowed callers
Possible values: "direct" / "code_execution_20250825" /
"code_execution_20260120"
cache_control:
$ref: '#/components/schemas/ClaudeCacheControl'
description: Cache control configuration
defer_loading:
type: boolean
description: >-
Whether to defer loading
When set to true, the tool is not included in the initial system
prompt,
and is only loaded when a tool_reference is returned via tool_search
eager_input_streaming:
type: boolean
description: >-
Whether to enable eager input streaming
When true, tool input parameters are streamed incrementally and
types are dynamically inferred
When false, the complete JSON output is buffered even if
fine-grained tool streaming is enabled
When null (default), the default behavior based on the beta header
is used
input_examples:
type: array
items:
$ref: '#/components/schemas/MapObject'
description: |-
List of tool input examples
Helps the model understand how to use the tool
name:
type: string
description: Tool name
description:
type: string
description: Tool description
type:
type: string
description: >-
Tool type, only client tools need to be set to "custom"
Server-side tools have their own type identifiers (e.g.
"bash_20250124", "web_search_20250305", etc.)
strict:
type: boolean
description: >-
Strict mode
When set to true, guarantees schema validation of tool name and
input
ClaudeMetadata:
type: object
properties:
user_id:
type: string
description: ''
ClaudeStopDetails:
type: object
properties:
type:
type: string
description: Object type, always "refusal"
category:
type: string
description: |-
Policy category that triggered the refusal
Possible values: "cyber" / "bio" / "reasoning_extraction"
Null when the refusal does not correspond to a named category
explanation:
type: string
description: >-
Human-readable explanation of the refusal
This text is not guaranteed to be stable. Null when no explanation
is available for the category
ClaudeContentBlock:
type: object
properties: {}
ClaudeContainer:
type: object
properties:
expires_at:
type: string
description: Container expiration time (ISO 8601 format)
id:
type: string
description: Container identifier
ClaudeUsage:
type: object
properties:
input_tokens:
type: integer
description: ''
output_tokens:
type: integer
description: ''
cache_creation_input_tokens:
type: integer
description: ''
cache_read_input_tokens:
type: integer
description: ''
cache_creation:
$ref: '#/components/schemas/ClaudeCacheCreation'
description: Cache creation details, categorized by TTL
inference_geo:
type: string
description: Inference geographic region
output_tokens_details:
$ref: '#/components/schemas/ClaudeOutputTokensDetails'
description: >-
Output token details, broken down by category (read-only, for
observability)
server_tool_use:
$ref: '#/components/schemas/ClaudeServerToolUsage'
description: Server-side tool usage statistics
service_tier:
type: string
description: |-
Service tier
Possible values: "standard" / "priority" / "batch"
JsonOutputFormat:
type: object
properties:
schema:
type: array
items:
type: array
items:
type: array
items:
type: object
properties: {}
description: JSON Schema definition
type:
type: string
description: Format type, always "json_schema"
MapObject:
type: object
properties:
key:
$ref: '#/components/schemas/key'
ClaudeCacheCreation:
type: object
properties:
ephemeral_5m_input_tokens:
type: integer
description: Number of input tokens used to create 5-minute cache entries
ephemeral_1h_input_tokens:
type: integer
description: Number of input tokens used to create 1-hour cache entries
ClaudeOutputTokensDetails:
type: object
properties:
thinking_tokens:
type: integer
description: >-
Number of output tokens generated by the model as internal reasoning
Includes thinking block delimiter tokens
Always ≤ output_tokens; output_tokens - thinking_tokens ≈
non-reasoning output
ClaudeServerToolUsage:
type: object
properties:
web_search_requests:
type: integer
description: Number of web search tool requests
web_fetch_requests:
type: integer
description: Number of web fetch tool requests
key:
type: object
properties: {}
````