generated: '2026-08-02'
method: derived
source: openapi/taalas-inference-api-openapi.yml, openapi/taalas-inference-v1-api-openapi.yml
note: >-
  Taalas is a stateless inference API — it exposes no persistent resources, no identifiers and no
  CRUD surface, so there is no entity graph in the usual sense. The model below is the message
  graph: the request/response schemas declared in the two OpenAPI documents and the $ref links
  between them. Success responses on the inference operations are declared with an empty schema
  ({}) in both published specs, so no response entity can be derived without fabricating it.
domains:
- name: Inference
  entities: [GenerateRequest, V1CompletionsRequest, V1ChatCompletionsRequest, StreamOptions]
- name: Messages
  entities: [Message, TextContent, ImageContent]
- name: Tools
  entities: [Tool, FunctionDefinition, ParametersDefinition]
- name: Monitoring
  entities: [HealthResponse]
- name: Errors
  entities: [HTTPValidationError, ValidationError]
entities:
- name: GenerateRequest
  spec: openapi/taalas-inference-api-openapi.yml
  used_by: [generate_generate_post]
  required: [prompt]
  additional_properties: true
  key_fields:
  - {name: prompt, type: array<object>}
  - {name: model, type: string, default: llama3.1-8B}
  - {name: min_output_tokens, type: integer, default: 0}
  - {name: max_output_tokens, type: integer, default: 0}
  - {name: topk, type: integer, min: 1, max: 8, default: 1}
  - {name: temperature, type: number, min: 0, max: 2, default: 1}
  - {name: stream, type: boolean, default: false}
  - {name: tools, type: array<object>}
  - {name: tool_choice, type: string}
  - {name: tool_name, type: string}
  - {name: context_for_tool, type: string}
  - {name: guided_json, type: object}
  - {name: guided_regex, type: string}
  - {name: response_format, type: object}
  - {name: system_prompt_version, type: string, default: offline}
  - {name: logprobs, type: boolean, default: false}
  - {name: top_logprobs, type: integer, min: 0, max: 20, default: 0}
  - {name: stop, type: array<string>|string, note: up to 4 sequences}
  - {name: echo, type: boolean, default: false}
  - {name: ignore_eos, type: boolean, default: false}
  - {name: metadata, type: object}
- name: V1CompletionsRequest
  spec: openapi/taalas-inference-v1-api-openapi.yml
  used_by: [completions_completions_post]
  required: [model, prompt]
  additional_properties: true
  key_fields:
  - {name: model, type: string}
  - {name: prompt, type: string|array<string>|array<integer>|array<array<integer>>}
  - {name: max_tokens, type: integer, default: 0}
  - {name: temperature, type: number, min: 0, max: 2, default: 1}
  - {name: top_p, type: number, min: 0, max: 1, default: 1.0}
  - {name: logprobs, type: integer, min: 0, max: 5}
  - {name: stream, type: boolean, default: false}
  - {name: echo, type: boolean, default: false}
  - {name: stop, type: array<string>|string}
  - {name: ignore_eos, type: boolean, default: false}
- name: V1ChatCompletionsRequest
  spec: openapi/taalas-inference-v1-api-openapi.yml
  used_by: [chat_completions_chat_completions_post]
  required: [model]
  additional_properties: true
  key_fields:
  - {name: model, type: string}
  - {name: messages, type: array<Message>}
  - {name: temperature, type: number, min: 0, max: 2, default: 1}
  - {name: top_p, type: number, min: 0, max: 1, default: 1}
  - {name: n, type: integer, min: 1, default: 1}
  - {name: max_tokens, type: integer, default: 0}
  - {name: max_completion_tokens, type: integer, default: 0}
  - {name: presence_penalty, type: number, min: -2, max: 2}
  - {name: frequency_penalty, type: number, min: -2, max: 2}
  - {name: logit_bias, type: map<string,number>}
  - {name: user, type: string}
  - {name: tools, type: array<Tool>}
  - {name: tool_choice, type: string}
  - {name: logprobs, type: boolean, default: false}
  - {name: top_logprobs, type: integer, min: 0, max: 20, default: 0}
  - {name: stream, type: boolean, default: false}
  - {name: ignore_eos, type: boolean, default: false}
- name: Message
  spec: openapi/taalas-inference-v1-api-openapi.yml
  required: [role]
  key_fields:
  - {name: role, type: string}
  - {name: content, type: 'string | array<TextContent|ImageContent|object>'}
- name: TextContent
  required: [text]
  key_fields:
  - {name: type, type: string, default: text}
  - {name: text, type: string}
- name: ImageContent
  required: [image_url]
  key_fields:
  - {name: type, type: string, default: image_url}
  - {name: image_url, type: map<string,string>}
- name: Tool
  key_fields:
  - {name: type, type: string, default: function}
  - {name: function, type: FunctionDefinition}
- name: FunctionDefinition
  key_fields:
  - {name: name, type: string}
  - {name: description, type: string}
  - {name: parameters, type: ParametersDefinition}
- name: ParametersDefinition
  key_fields:
  - {name: type, type: string, default: object}
  - {name: description, type: string, default: ''}
  - {name: properties, type: object}
  - {name: required, type: array<string>}
- name: StreamOptions
  key_fields:
  - {name: include_usage, type: boolean, default: false}
  - {name: include_obfuscation, type: boolean, default: false}
- name: HealthResponse
  spec: openapi/taalas-inference-api-openapi.yml
  used_by: [health_health_get]
  required: [status, queue_size, current_adapter]
  key_fields:
  - {name: status, type: string, note: always "healthy" per the operation description}
  - {name: queue_size, type: integer, note: items in the inference queue}
  - {name: current_adapter, type: string, note: currently loaded LoRA adapter, or "none"}
- name: HTTPValidationError
  key_fields:
  - {name: detail, type: array<ValidationError>}
- name: ValidationError
  required: [loc, msg, type]
  key_fields:
  - {name: loc, type: array<string|integer>}
  - {name: msg, type: string}
  - {name: type, type: string}
relationships:
- from: V1ChatCompletionsRequest
  to: Message
  kind: has_many
  via: messages
- from: V1ChatCompletionsRequest
  to: Tool
  kind: has_many
  via: tools
- from: V1ChatCompletionsRequest
  to: StreamOptions
  kind: has_one
  via: stream_options
- from: V1CompletionsRequest
  to: StreamOptions
  kind: has_one
  via: stream_options
- from: GenerateRequest
  to: StreamOptions
  kind: has_one
  via: stream_options
- from: Message
  to: TextContent
  kind: has_many
  via: content
- from: Message
  to: ImageContent
  kind: has_many
  via: content
- from: Tool
  to: FunctionDefinition
  kind: has_one
  via: function
- from: FunctionDefinition
  to: ParametersDefinition
  kind: has_one
  via: parameters
- from: HTTPValidationError
  to: ValidationError
  kind: has_many
  via: detail
persistent_resources: []
identifiers: []
id_prefixes: []
gaps:
- No response schema is declared for /generate, /models, /v1/models, /v1/completions or
  /v1/chat/completions — the 200 responses use an empty schema, so the output entity (choices,
  usage, tokens/second, etc.) cannot be derived from the published contract.
- GenerateRequest.prompt is typed as array<object> with no item schema, so the Taalas-native prompt
  shape is not machine-readable.