Schemas

ChatCompletionRequest

Fields not listed here are rejected with 400 unsupported_parameter, except store and metadata, which are ignored, and fields sent as null.

model

string · enum · required

A chat model id from GET /models.

Enum values:

gpt-6.1-terra

muse-spark-1.3-contributor

deepseek-v4.1-flash-1017

gemini-3.8-flash

glm-5.3-flashx

Message[] · minItems: 1 · maxItems: 2048 · required

The conversation so far, oldest first.

max_tokens

integer · min: 1

The most tokens the reply may use, reasoning included. Defaults to 8,192, or the model's max_output_tokens if lower. A value above the model's maximum is rejected. Also accepted as max_completion_tokens, which wins if both are sent.

max_completion_tokens

integer · min: 1

Same as max_tokens.

temperature

number · min: 0 · max: 2

top_p

number · min: 0 · max: 1

Up to 4 sequences where the model stops generating.

seed

integer

presence_penalty

number · min: -2 · max: 2

frequency_penalty

number · min: -2 · max: 2

Tool[] · minItems: 1 · maxItems: 128

none, auto, required, or a named function. Only allowed together with tools.

parallel_tool_calls

boolean

Only allowed together with tools.

ResponseFormat

reasoning_effort

string · enum

How much the model reasons before answering. none is refused by models that always reason (see reasoning_can_be_disabled). Omit it to use the model's default.

Enum values:

none

minimal

low

medium

high

xhigh

stream

boolean

Default: false

object

Only allowed with stream: true.

user

string · maxLength: 256

An id for your end user. It is hashed before it reaches the model provider and helps attribute abuse to one user instead of your whole account.

n

integer · enum

Only one choice is supported.

Enum values:

logprobs

boolean · enum

Log probabilities are not supported.

Enum values:

Message

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = object · requires: role, content
	type = object · requires: role, content
	type = object · requires: role
	type = object · requires: role, tool_call_id, content

Properties for System or developer message:

SystemMessage

role

string · enum · required

developer is treated as system.

Enum values:

system

developer

TextContent · required

name

string

SystemMessage

role

string · enum · required

developer is treated as system.

Enum values:

system

developer

TextContent · required

name

string

UserMessage

role

string · enum · required

Enum values:

user

required

name

string

AssistantMessage

Needs `content`, `tool_calls`, or both.

role

string · enum · required

Enum values:

assistant

ToolCall[]

object[]

The reasoning_details the model returned, sent back unchanged.

name

string

ToolMessage

role

string · enum · required

Enum values:

tool

tool_call_id

string · required

TextContent · required

TextContent

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = string
	type = array

Properties for Variant 1:

string

TextPart

type

string · enum · required

Enum values:

text

text

string · required

ImagePart

type

string · enum · required

Enum values:

image_url

object · required

Tool

type

string · enum · required

Enum values:

function

object · required

ToolCall

id

string · required

type

string · enum · required

Enum values:

function

object · required

ResponseFormat

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = object · requires: type
	type = object · requires: type
	type = object · requires: type, json_schema

Properties for Text:

type

string · enum · required

Enum values:

text

ChatCompletion

id

string · required

object

string · enum · required

Enum values:

chat.completion

created

integer · required

model

string · required

object[] · minItems: 1 · maxItems: 1 · required

Usage · required

system_fingerprint

null

ReplyMessage

role

string · enum · required

Enum values:

assistant

content

string | null · required

refusal

string | null

reasoning_content

string

The model's readable reasoning, when it shares it.

object[]

Opaque reasoning blocks to send back on the next request.

ToolCall[]

FinishReason

string · enum

Enum values:

stop

length

tool_calls

content_filter

ChatCompletionChunk

id

string · required

object

string · enum · required

Enum values:

chat.completion.chunk

created

integer · required

model

string · required

object[] · required

Empty in the final usage chunk.

system_fingerprint

null

Usage

prompt_tokens

integer · required

completion_tokens

integer · required

Output tokens, reasoning included.

total_tokens

integer · required

object · required

cost_credits

number · required

The exact cost of this request in credits.

ModelList

object

string · enum · required

Enum values:

list

Model[] · required

Model

id

string · required

object

string · enum · required

Enum values:

model

created

integer · required

owned_by

string · required

name

string · required

context_length

integer · required

The most input tokens the model accepts.

max_output_tokens

integer · required

The largest max_tokens this model accepts.

input_modalities

string[] · required

Enum values:

text

image

reasoning_can_be_disabled

boolean · required

Whether reasoning_effort accepts none.

supports_json_schema

boolean · required

Whether response_format accepts json_schema.

supported_parameters

string[] · required

object · required

notes

string · required

DecisionRequest

Fields not listed here are rejected with 400 unsupported_parameter, except fields sent as null.

model

string · enum · required

A decision model id.

Enum values:

jev-1.14

required

The content to evaluate, as a string or a JSON object or array.

object · minProps: 1 · maxProps: 64 · required

Questions keyed by a name you choose, 1 to 128 characters. Each answer comes back under the same name.

object · minProps: 1 · maxProps: 16

Images keyed by the reference string you write wherever the image belongs in state, instructions or criteria. Every asset must be referenced at least once.

user

string · maxLength: 256

An id for your end user. It is hashed before it reaches the model provider.

Guidance

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = string
	type = object
	type = array

Properties for Variant 1:

string

DecisionQuestion

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = object · requires: type, instructions
	type = object · requires: type, instructions, criteria
	type = object · requires: type, instructions, criteria
	type = object · requires: type, instructions

Properties for Noul:

NoulQuestion

type

string · enum · required

Enum values:

noul

Guidance · required

A plain string, or a JSON object or array of structured guidance.

object

What makes the answer yes and what makes it no.

NoulQuestion

type

string · enum · required

Enum values:

noul

Guidance · required

A plain string, or a JSON object or array of structured guidance.

object

What makes the answer yes and what makes it no.

ChoiceQuestion

type

string · enum · required

Enum values:

choice

Guidance · required

A plain string, or a JSON object or array of structured guidance.

object · minProps: 1 · required

Each option, mapped to what it means.

ScoreQuestion

type

string · enum · required

Enum values:

score

Guidance · required

A plain string, or a JSON object or array of structured guidance.

Guidance[] · minItems: 1 · required

The levels of the scale, lowest first.

BoxQuestion

type

string · enum · required

Enum values:

box

Guidance · required

A plain string, or a JSON object or array of structured guidance.

ImageAsset

format

string · enum · required

Enum values:

png

jpg

jpeg

data_base64

string · required

The file in standard base64, without a data URL prefix.

type

string · enum

Enum values:

image

Decision

id

string · required

model

string · required

object · required

One answer per question, keyed by the question's name.

object · required

DecisionAnswer

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = object · requires: type, noul
	type = object · requires: type, choice
	type = object · requires: type, score
	type = object · requires: type, box

Properties for Noul answer:

NoulAnswer

type

string · enum · required

Enum values:

noul

noul

number · required

The probability that the statement is true, 0 to 1.

NoulAnswer

type

string · enum · required

Enum values:

noul

noul

number · required

The probability that the statement is true, 0 to 1.

ChoiceAnswer

type

string · enum · required

Enum values:

choice

choice

string · required

confidence

number

object

ScoreAnswer

type

string · enum · required

Enum values:

score

score

number · required

The expected level, from 0 for the first level.

confidence

number

legend

object

Each level's index mapped to its criterion.

object

BoxAnswer

type

string · enum · required

Enum values:

box

object · required

Edges as fractions of the image's height and width.

confidence

number | null

Error

object · required

Hoosha Chat API

Schemas

ChatCompletionRequest

Fields not listed here are rejected with 400 unsupported_parameter, except store and metadata, which are ignored, and fields sent as null.

model

string · enum · required

A chat model id from GET /models.

Enum values:

gpt-6.1-terra

muse-spark-1.3-contributor

deepseek-v4.1-flash-1017

gemini-3.8-flash

glm-5.3-flashx

Message[] · minItems: 1 · maxItems: 2048 · required

The conversation so far, oldest first.

max_tokens

integer · min: 1

max_completion_tokens

integer · min: 1

Same as max_tokens.

temperature

number · min: 0 · max: 2

top_p

number · min: 0 · max: 1

Up to 4 sequences where the model stops generating.

seed

integer

presence_penalty

number · min: -2 · max: 2

frequency_penalty

number · min: -2 · max: 2

Tool[] · minItems: 1 · maxItems: 128

none, auto, required, or a named function. Only allowed together with tools.

parallel_tool_calls

boolean

Only allowed together with tools.

ResponseFormat

reasoning_effort

string · enum

How much the model reasons before answering. none is refused by models that always reason (see reasoning_can_be_disabled). Omit it to use the model's default.

Enum values:

none

minimal

low

medium

high

xhigh

stream

boolean

Default: false

object

Only allowed with stream: true.

user

string · maxLength: 256

An id for your end user. It is hashed before it reaches the model provider and helps attribute abuse to one user instead of your whole account.

n

integer · enum

Only one choice is supported.

Enum values:

logprobs

boolean · enum

Log probabilities are not supported.

Enum values:

Message

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = object · requires: role, content
	type = object · requires: role, content
	type = object · requires: role
	type = object · requires: role, tool_call_id, content

Properties for System or developer message:

SystemMessage

role

string · enum · required

developer is treated as system.

Enum values:

system

developer

TextContent · required

name

string

SystemMessage

role

string · enum · required

developer is treated as system.

Enum values:

system

developer

TextContent · required

name

string

UserMessage

role

string · enum · required

Enum values:

user

required

name

string

AssistantMessage

Needs `content`, `tool_calls`, or both.

role

string · enum · required

Enum values:

assistant

ToolCall[]

object[]

The reasoning_details the model returned, sent back unchanged.

name

string

ToolMessage

role

string · enum · required

Enum values:

tool

tool_call_id

string · required

TextContent · required

TextContent

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = string
	type = array

Properties for Variant 1:

string

TextPart

type

string · enum · required

Enum values:

text

text

string · required

ImagePart

type

string · enum · required

Enum values:

image_url

object · required

Tool

type

string · enum · required

Enum values:

function

object · required

ToolCall

id

string · required

type

string · enum · required

Enum values:

function

object · required

ResponseFormat

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = object · requires: type
	type = object · requires: type
	type = object · requires: type, json_schema

Properties for Text:

type

string · enum · required

Enum values:

text

ChatCompletion

id

string · required

object

string · enum · required

Enum values:

chat.completion

created

integer · required

model

string · required

object[] · minItems: 1 · maxItems: 1 · required

Usage · required

system_fingerprint

null

ReplyMessage

role

string · enum · required

Enum values:

assistant

content

string | null · required

refusal

string | null

reasoning_content

string

The model's readable reasoning, when it shares it.

object[]

Opaque reasoning blocks to send back on the next request.

ToolCall[]

FinishReason

string · enum

Enum values:

stop

length

tool_calls

content_filter

ChatCompletionChunk

id

string · required

object

string · enum · required

Enum values:

chat.completion.chunk

created

integer · required

model

string · required

object[] · required

Empty in the final usage chunk.

system_fingerprint

null

Usage

prompt_tokens

integer · required

completion_tokens

integer · required

Output tokens, reasoning included.

total_tokens

integer · required

object · required

cost_credits

number · required

The exact cost of this request in credits.

ModelList

object

string · enum · required

Enum values:

list

Model[] · required

Model

id

string · required

object

string · enum · required

Enum values:

model

created

integer · required

owned_by

string · required

name

string · required

context_length

integer · required

The most input tokens the model accepts.

max_output_tokens

integer · required

The largest max_tokens this model accepts.

input_modalities

string[] · required

Enum values:

text

image

reasoning_can_be_disabled

boolean · required

Whether reasoning_effort accepts none.

supports_json_schema

boolean · required

Whether response_format accepts json_schema.

supported_parameters

string[] · required

object · required

notes

string · required

DecisionRequest

Fields not listed here are rejected with 400 unsupported_parameter, except fields sent as null.

model

string · enum · required

A decision model id.

Enum values:

jev-1.14

required

The content to evaluate, as a string or a JSON object or array.

object · minProps: 1 · maxProps: 64 · required

Questions keyed by a name you choose, 1 to 128 characters. Each answer comes back under the same name.

object · minProps: 1 · maxProps: 16

Images keyed by the reference string you write wherever the image belongs in state, instructions or criteria. Every asset must be referenced at least once.

user

string · maxLength: 256

An id for your end user. It is hashed before it reaches the model provider.

Guidance

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = string
	type = object
	type = array

Properties for Variant 1:

string

DecisionQuestion

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = object · requires: type, instructions
	type = object · requires: type, instructions, criteria
	type = object · requires: type, instructions, criteria
	type = object · requires: type, instructions

Properties for Noul:

NoulQuestion

type

string · enum · required

Enum values:

noul

Guidance · required

A plain string, or a JSON object or array of structured guidance.

object

What makes the answer yes and what makes it no.

NoulQuestion

type

string · enum · required

Enum values:

noul

Guidance · required

A plain string, or a JSON object or array of structured guidance.

object

What makes the answer yes and what makes it no.

ChoiceQuestion

type

string · enum · required

Enum values:

choice

Guidance · required

A plain string, or a JSON object or array of structured guidance.

object · minProps: 1 · required

Each option, mapped to what it means.

ScoreQuestion

type

string · enum · required

Enum values:

score

Guidance · required

A plain string, or a JSON object or array of structured guidance.

Guidance[] · minItems: 1 · required

The levels of the scale, lowest first.

BoxQuestion

type

string · enum · required

Enum values:

box

Guidance · required

A plain string, or a JSON object or array of structured guidance.

ImageAsset

format

string · enum · required

Enum values:

png

jpg

jpeg

data_base64

string · required

The file in standard base64, without a data URL prefix.

type

string · enum

Enum values:

image

Decision

id

string · required

model

string · required

object · required

One answer per question, keyed by the question's name.

object · required

DecisionAnswer

oneOf

Exactly one variant must match.

Decision Table

Variant	Matching Criteria
	type = object · requires: type, noul
	type = object · requires: type, choice
	type = object · requires: type, score
	type = object · requires: type, box

Properties for Noul answer:

NoulAnswer

type

string · enum · required

Enum values:

noul

noul

number · required

The probability that the statement is true, 0 to 1.

NoulAnswer

type

string · enum · required

Enum values:

noul

noul

number · required

The probability that the statement is true, 0 to 1.

ChoiceAnswer

type

string · enum · required

Enum values:

choice

choice

string · required

confidence

number

object

ScoreAnswer

type

string · enum · required

Enum values:

score

score

number · required

The expected level, from 0 for the first level.

confidence

number

legend

object

Each level's index mapped to its criterion.

object

BoxAnswer

type

string · enum · required

Enum values:

box

object · required

Edges as fractions of the image's height and width.

confidence

number | null

Error

object · required