openapi: 3.1.0
info:
  title: siema_ai API
  version: "1.0.0"
  description: |
    The siema_ai API serves Polish-first open-weight models (Bielik, Qwen, BGE-M3)
    behind an OpenAI-compatible interface. Point an OpenAI SDK at
    `https://api.siema-ai.pl/v1` with an `sk-siema-` API key and the chat
    completions, embeddings and models endpoints work unchanged.

    Billing is prepaid in PLN. Prices are published per model in micro-PLN
    (1 PLN = 1 000 000 micro-PLN) per million tokens; `GET /v1/models` is the
    source of truth for the price in force right now. No prompt, completion or
    embedding input is ever stored, logged or traced.
  contact:
    name: siema_ai support
    email: support@siema-ai.pl
    url: https://docs.siema-ai.pl
servers:
  - url: https://api.siema-ai.pl
    description: Production (EU)
tags:
  - name: Chat
    description: Chat completions, streaming and non-streaming.
  - name: Embeddings
    description: Text embeddings with BGE-M3 (1024 dimensions).
  - name: Models
    description: Model aliases, capabilities, limits and current prices.
security:
  - ApiKeyAuth: []

paths:
  /v1/chat/completions:
    post:
      operationId: createChatCompletion
      tags: [Chat]
      summary: Create a chat completion
      description: |
        Generates a model response for a conversation. Set `stream: true` to
        receive server-sent events.

        Any request field not listed in the schema (for example `tools`,
        `tool_choice`, `response_format`, `seed`, `logprobs`, `presence_penalty`,
        `best_of`) is forwarded to the model server unmodified. The gateway
        rewrites `model` to the upstream model name, clamps `max_tokens` and
        `max_completion_tokens` to the model's `max_output_tokens`, and always
        requests usage from the upstream so it can bill the request; the usage
        chunk is returned to you only when `stream_options.include_usage` is
        true.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ChatCompletionRequest'
            examples:
              nonStreaming:
                summary: Non-streaming Polish request
                value:
                  model: polish-pro
                  messages:
                    - role: system
                      content: Jesteś pomocnym asystentem. Odpowiadaj po polsku.
                    - role: user
                      content: Streść w dwóch zdaniach, czym jest faktura VAT.
                  max_tokens: 200
              streaming:
                summary: Streaming with usage in the last chunk
                value:
                  model: polish-pro
                  stream: true
                  stream_options:
                    include_usage: true
                  messages:
                    - role: user
                      content: Napisz krótki wiersz o Warszawie.
      responses:
        '200':
          description: |
            With `stream: false` the body is one `ChatCompletion` JSON object.
            With `stream: true` the body is `text/event-stream`: each event is
            `data: <ChatCompletionChunk JSON>` followed by a blank line, and the
            stream ends with `data: [DONE]`. When `stream_options.include_usage`
            is true, the last chunk before `[DONE]` has an **empty** `choices`
            array and a `usage` object.

            If the stream fails after the response has started (HTTP 200 was
            already sent), it ends without `data: [DONE]` and without any
            in-band error event; treat a stream that ends without `[DONE]`, or
            whose last choice never carried a `finish_reason`, as incomplete.

            Example streaming transcript:

            ```
            data: {"id":"chatcmpl-0192a1b2c3d4","object":"chat.completion.chunk","created":1758700000,"model":"polish-pro","choices":[{"index":0,"delta":{"role":"assistant","content":""},"finish_reason":null}]}

            data: {"id":"chatcmpl-0192a1b2c3d4","object":"chat.completion.chunk","created":1758700000,"model":"polish-pro","choices":[{"index":0,"delta":{"content":"Warszawo, "},"finish_reason":null}]}

            data: {"id":"chatcmpl-0192a1b2c3d4","object":"chat.completion.chunk","created":1758700000,"model":"polish-pro","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

            data: {"id":"chatcmpl-0192a1b2c3d4","object":"chat.completion.chunk","created":1758700000,"model":"polish-pro","choices":[],"usage":{"prompt_tokens":12,"completion_tokens":40,"total_tokens":52}}

            data: [DONE]
            ```
          headers:
            X-Request-Id:
              $ref: '#/components/headers/X-Request-Id'
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ChatCompletion'
              example:
                id: chatcmpl-0192a1b2c3d4
                object: chat.completion
                created: 1758700000
                model: polish-pro
                choices:
                  - index: 0
                    message:
                      role: assistant
                      content: Faktura VAT to dokument potwierdzający sprzedaż towaru lub usługi, zawierający kwotę netto, stawkę i kwotę podatku VAT oraz kwotę brutto. Wystawia ją sprzedawca będący podatnikiem VAT.
                    finish_reason: stop
                usage:
                  prompt_tokens: 41
                  completion_tokens: 58
                  total_tokens: 99
            text/event-stream:
              schema:
                $ref: '#/components/schemas/ChatCompletionChunk'
        '400':
          $ref: '#/components/responses/BadRequest'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/InsufficientQuota'
        '403':
          $ref: '#/components/responses/Forbidden'
        '404':
          $ref: '#/components/responses/NotFound'
        '429':
          $ref: '#/components/responses/RateLimited'
        '500':
          $ref: '#/components/responses/ServerError'
        '502':
          $ref: '#/components/responses/UpstreamError'
        '503':
          $ref: '#/components/responses/ServerError'

  /v1/embeddings:
    post:
      operationId: createEmbedding
      tags: [Embeddings]
      summary: Create embeddings
      description: |
        Returns dense vectors for one string or an array of strings using the
        `embed` alias (BGE-M3, 1024 dimensions, multilingual). Only input
        tokens are billed. Fields not in the schema are forwarded unmodified.
        There is no fixed cap on the number of inputs per request; the request
        body itself is capped (20 MiB by default), which returns 400
        `invalid_request`.
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/EmbeddingRequest'
            example:
              model: embed
              input:
                - Umowa o pracę na czas nieokreślony
                - Umowa zlecenie
      responses:
        '200':
          description: Embedding vectors in input order.
          headers:
            X-Request-Id:
              $ref: '#/components/headers/X-Request-Id'
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/EmbeddingResponse'
              example:
                object: list
                model: embed
                data:
                  - object: embedding
                    index: 0
                    embedding: [0.0123, -0.0456, 0.0789]
                  - object: embedding
                    index: 1
                    embedding: [0.0231, -0.0112, 0.0655]
                usage:
                  prompt_tokens: 14
                  total_tokens: 14
        '400':
          $ref: '#/components/responses/BadRequest'
        '401':
          $ref: '#/components/responses/Unauthorized'
        '402':
          $ref: '#/components/responses/InsufficientQuota'
        '403':
          $ref: '#/components/responses/Forbidden'
        '404':
          $ref: '#/components/responses/NotFound'
        '429':
          $ref: '#/components/responses/RateLimited'
        '500':
          $ref: '#/components/responses/ServerError'
        '502':
          $ref: '#/components/responses/UpstreamError'
        '503':
          $ref: '#/components/responses/ServerError'

  /v1/models:
    get:
      operationId: listModels
      tags: [Models]
      summary: List models
      description: |
        Lists every enabled model alias with its capabilities, limits and the
        price in force right now. This endpoint runs behind authentication
        only (no balance check, no rate limit) and is the source of truth for
        pricing; the docs only quote it. Prices and limits shown in this
        document's examples are illustrative placeholders pending the launch
        benchmark.
      responses:
        '200':
          description: Enabled models.
          headers:
            X-Request-Id:
              $ref: '#/components/headers/X-Request-Id'
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ModelList'
              example:
                object: list
                data:
                  - id: polish-pro
                    object: model
                    created: 1758000000
                    owned_by: siema
                    capabilities: [chat]
                    context_length: 8192
                    max_output_tokens: 2048
                    pricing:
                      input_micro_pln_per_1m: 1500000
                      output_micro_pln_per_1m: 3000000
                      currency: PLN
                  - id: embed
                    object: model
                    created: 1758000000
                    owned_by: siema
                    capabilities: [embeddings]
                    context_length: 8192
                    max_output_tokens: 0
                    pricing:
                      input_micro_pln_per_1m: 200000
                      output_micro_pln_per_1m: 0
                      currency: PLN
        '401':
          $ref: '#/components/responses/Unauthorized'
        '403':
          $ref: '#/components/responses/Forbidden'
        '500':
          $ref: '#/components/responses/ServerError'

  /v1/models/{id}:
    get:
      operationId: retrieveModel
      tags: [Models]
      summary: Retrieve a model
      description: Retrieves the model with the given alias. Runs behind authentication only, like `GET /v1/models`.
      parameters:
        - name: id
          in: path
          required: true
          description: Model alias, for example `polish-pro`.
          schema:
            type: string
          example: polish-pro
      responses:
        '200':
          description: The model.
          headers:
            X-Request-Id:
              $ref: '#/components/headers/X-Request-Id'
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Model'
              example:
                id: polish-pro
                object: model
                created: 1758000000
                owned_by: siema
                capabilities: [chat]
                context_length: 8192
                max_output_tokens: 2048
                pricing:
                  input_micro_pln_per_1m: 1500000
                  output_micro_pln_per_1m: 3000000
                  currency: PLN
        '401':
          $ref: '#/components/responses/Unauthorized'
        '403':
          $ref: '#/components/responses/Forbidden'
        '404':
          $ref: '#/components/responses/NotFound'
        '500':
          $ref: '#/components/responses/ServerError'

components:
  securitySchemes:
    ApiKeyAuth:
      type: http
      scheme: bearer
      bearerFormat: sk-siema-<40 characters>
      description: |
        Create keys in the console (Keys page). Send as
        `Authorization: Bearer sk-siema-...` (the `Bearer` scheme is
        case-insensitive). A malformed key is rejected with 401 before any
        lookup. Keys are org-scoped and shown once. Revoking a key deletes the
        gateway's cache entry within one second; the next request with that
        key gets 401.

  headers:
    X-Request-Id:
      description: |
        Unique id of this request, minted by the gateway. Include it in
        support requests; it also appears in the error body's `request_id`
        field. A client-sent `X-Request-Id` header is never used: it is
        always ignored, not echoed back.
      required: true
      schema:
        type: string
        example: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
    Retry-After:
      description: |
        Seconds to wait before retrying. On `rate_limit_exceeded` this is the
        time remaining in the per-key window; on `concurrency_limit_exceeded`
        it is always 1.
      schema:
        type: integer
        minimum: 1
        example: 12

  responses:
    BadRequest:
      description: Malformed JSON, missing `messages`, a parameter outside its allowed range, or a request body over the size limit (`invalid_json`, `invalid_request`, `context_length_exceeded`).
      headers:
        X-Request-Id:
          $ref: '#/components/headers/X-Request-Id'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
          example:
            error:
              message: messages must be a non-empty array
              type: invalid_request_error
              code: invalid_request
              param: messages
              request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
    Unauthorized:
      description: Missing, malformed, unknown or revoked API key (`invalid_api_key`).
      headers:
        X-Request-Id:
          $ref: '#/components/headers/X-Request-Id'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
          example:
            error:
              message: invalid or revoked API key
              type: authentication_error
              code: invalid_api_key
              param: null
              request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
    InsufficientQuota:
      description: |
        The organisation's prepaid balance is not positive (`insufficient_quota`).
        Top up in the console. This is **402**, not 429 as at OpenAI, so that
        SDKs do not retry a request that cannot succeed. Checked before the
        rate limit.
      headers:
        X-Request-Id:
          $ref: '#/components/headers/X-Request-Id'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
          example:
            error:
              message: prepaid balance is exhausted; top up in the console
              type: insufficient_quota
              code: insufficient_quota
              param: null
              request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
    Forbidden:
      description: The organisation is suspended (`org_suspended`).
      headers:
        X-Request-Id:
          $ref: '#/components/headers/X-Request-Id'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
          example:
            error:
              message: this organization is suspended
              type: permission_error
              code: org_suspended
              param: null
              request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
    NotFound:
      description: Unknown model alias (`model_not_found`), or an endpoint siema_ai does not offer (`unsupported_endpoint`).
      headers:
        X-Request-Id:
          $ref: '#/components/headers/X-Request-Id'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
          examples:
            model:
              summary: Unknown model alias
              value:
                error:
                  message: model "gpt-4o" does not exist
                  type: invalid_request_error
                  code: model_not_found
                  param: model
                  request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
            endpoint:
              summary: Endpoint not offered
              value:
                error:
                  message: /v1/images/generations is not supported by this API
                  type: invalid_request_error
                  code: unsupported_endpoint
                  param: null
                  request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
    RateLimited:
      description: Per-key requests-per-minute (`rate_limit_exceeded`) or per-organisation concurrency (`concurrency_limit_exceeded`) limit exceeded.
      headers:
        X-Request-Id:
          $ref: '#/components/headers/X-Request-Id'
        Retry-After:
          $ref: '#/components/headers/Retry-After'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
          examples:
            perKey:
              summary: Requests-per-minute exceeded
              value:
                error:
                  message: requests per minute limit exceeded for this API key
                  type: rate_limit_error
                  code: rate_limit_exceeded
                  param: null
                  request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
            concurrency:
              summary: Per-organisation concurrency exceeded
              value:
                error:
                  message: too many concurrent requests for this organization
                  type: rate_limit_error
                  code: concurrency_limit_exceeded
                  param: null
                  request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
    ServerError:
      description: |
        Either an unexpected gateway failure (`internal_error`, HTTP 500) or
        temporarily no capacity for the requested model (`no_capacity`, HTTP
        503, for example when its routing table has no deployments). Safe to
        retry with backoff.
      headers:
        X-Request-Id:
          $ref: '#/components/headers/X-Request-Id'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
          examples:
            internal:
              summary: Unexpected gateway failure
              value:
                error:
                  message: authentication backend unavailable
                  type: server_error
                  code: internal_error
                  param: null
                  request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
            noCapacity:
              summary: No deployment available
              value:
                error:
                  message: no deployment available for model "advanced"
                  type: server_error
                  code: no_capacity
                  param: null
                  request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b
    UpstreamError:
      description: |
        The model server could not complete the request (`upstream_unavailable`)
        or timed out (`upstream_timeout`). A connection failure or a missing
        first byte within 180 s is retried once on another deployment of the
        same model; a model-server 4xx other than 400, an idle gap over 60 s
        once a non-streaming reply has started, and the 600 s total budget are
        not retried. A streaming reply's HTTP status is already 200 once it
        starts, so an idle gap there ends the stream without `data: [DONE]`
        instead of a 502. If the model server produced no output before the
        failure, nothing is billed; a non-streaming request that fails after
        the model server started sending its reply is billed the estimated
        input tokens, and so is a client disconnect at any point after the
        request reached the model server. Safe to retry with backoff.
      headers:
        X-Request-Id:
          $ref: '#/components/headers/X-Request-Id'
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/ErrorResponse'
          example:
            error:
              message: could not reach the model server
              type: upstream_error
              code: upstream_unavailable
              param: null
              request_id: 0192a1b2-c3d4-7e5f-8a9b-0c1d2e3f4a5b

  schemas:
    ContentPart:
      type: object
      description: |
        One part of a multimodal message content array, forwarded to the model
        server unmodified and unvalidated by the gateway. The model server
        expects `type` to be `text` or `image_url`. Image parts (`image_url`)
        are only understood by a model whose `capabilities` include `vision`;
        none of the V1 aliases do today, so a model server that cannot accept
        images rejects the request with a 400 that the gateway surfaces as
        `invalid_request`.
      properties:
        type:
          type: string
        text:
          type: string
        image_url:
          type: object
          properties:
            url:
              type: string
              description: HTTPS URL or `data:image/...;base64,...`.
      additionalProperties: true

    ChatMessage:
      type: object
      required: [role]
      properties:
        role:
          type: string
          enum: [system, user, assistant, tool]
        content:
          description: A string, an array of content parts for multimodal input, or null (an assistant message that carries only tool_calls).
          oneOf:
            - type: string
            - type: array
              items:
                $ref: '#/components/schemas/ContentPart'
            - type: 'null'
        name:
          type: string
        tool_call_id:
          type: string
        tool_calls:
          type: array
          items:
            type: object
      additionalProperties: true

    ChatCompletionRequest:
      type: object
      required: [model, messages]
      description: |
        Any additional OpenAI parameter not listed below (for example `tools`,
        `tool_choice`, `response_format`, `seed`, `logprobs`,
        `presence_penalty`, `best_of`) is forwarded to the model server
        unmodified.
      properties:
        model:
          type: string
          description: A model alias from `GET /v1/models`, for example `polish-fast`, `polish-pro`, `advanced`.
          example: polish-pro
        messages:
          type: array
          minItems: 1
          items:
            $ref: '#/components/schemas/ChatMessage'
        stream:
          type: boolean
          default: false
        stream_options:
          type: object
          properties:
            include_usage:
              type: boolean
              description: When true, a final chunk with empty `choices` and a `usage` object is sent before `[DONE]`.
        max_tokens:
          type: integer
          minimum: 1
          description: "Upper bound on generated tokens. Values above the model's `max_output_tokens` are clamped silently; a clamped answer ends with `finish_reason: \"length\"`."
        max_completion_tokens:
          type: integer
          minimum: 1
          description: Alias for `max_tokens`; also clamped to the model's `max_output_tokens`.
        temperature:
          type: number
          description: Forwarded to the model server unchanged, which validates it. OpenAI's range (0-2) is guidance, not enforced by siema_ai.
        top_p:
          type: number
          description: Forwarded to the model server unchanged, which validates it. OpenAI's range (0-1) is guidance, not enforced by siema_ai.
        stop:
          description: Forwarded to the model server unchanged, which validates it. siema_ai applies no limit on the number of sequences.
          oneOf:
            - type: string
            - type: array
              items:
                type: string
        n:
          type: integer
          description: Forwarded unmodified, with no limit applied by siema_ai. The model server may return more than one choice; every choice's output tokens are billed.
        user:
          type: string
          description: Forwarded unmodified. Not stored by siema_ai.
      additionalProperties: true

    Usage:
      type: object
      required: [prompt_tokens, total_tokens]
      description: Token accounting relayed from the model server. `completion_tokens` is absent on embeddings usage.
      properties:
        prompt_tokens:
          type: integer
        completion_tokens:
          type: integer
        total_tokens:
          type: integer

    ChatCompletionChoice:
      type: object
      required: [index, message]
      properties:
        index:
          type: integer
        message:
          type: object
          properties:
            role:
              type: string
            content:
              type: [string, "null"]
            tool_calls:
              type: array
              items:
                type: object
          additionalProperties: true
        finish_reason:
          type: [string, "null"]
          description: "One of `stop`, `length`, `tool_calls`, `content_filter`. `length` means `max_tokens` (possibly clamped) was reached."

    ChatCompletion:
      type: object
      required: [id, object, created, model, choices]
      properties:
        id:
          type: string
        object:
          type: string
          enum: [chat.completion]
        created:
          type: integer
          description: Unix seconds.
        model:
          type: string
          description: Always the requested alias, never the upstream model name.
        choices:
          type: array
          items:
            $ref: '#/components/schemas/ChatCompletionChoice'
        usage:
          $ref: '#/components/schemas/Usage'

    ChunkDelta:
      type: object
      description: Partial update to a streamed choice's message. Only the fields that changed since the previous chunk are present.
      properties:
        role:
          type: string
        content:
          type: [string, "null"]
      additionalProperties: true

    ChatCompletionChunkChoice:
      type: object
      properties:
        index:
          type: integer
        delta:
          $ref: '#/components/schemas/ChunkDelta'
        finish_reason:
          type: [string, "null"]
      additionalProperties: true

    ChatCompletionChunk:
      type: object
      required: [id, object, created, model, choices]
      properties:
        id:
          type: string
        object:
          type: string
          enum: [chat.completion.chunk]
        created:
          type: integer
        model:
          type: string
        choices:
          type: array
          description: Empty in the final usage chunk (sent only when `stream_options.include_usage` is true). Guard `choices[0]` accesses.
          items:
            $ref: '#/components/schemas/ChatCompletionChunkChoice'
        usage:
          description: Present only on the final chunk when `stream_options.include_usage` is true.
          $ref: '#/components/schemas/Usage'

    EmbeddingRequest:
      type: object
      required: [model, input]
      properties:
        model:
          type: string
          example: embed
        input:
          oneOf:
            - type: string
            - type: array
              minItems: 1
              items:
                type: string
        encoding_format:
          type: string
          enum: [float, base64]
          description: When `base64`, each embedding is a base64 string of little-endian float32 values instead of an array of numbers.
        user:
          type: string
      additionalProperties: true

    Embedding:
      type: object
      required: [object, index, embedding]
      properties:
        object:
          type: string
          enum: [embedding]
        index:
          type: integer
        embedding:
          description: 1024 floats for `embed`, as an array of numbers, or as a base64 string of little-endian float32 values when `encoding_format` is `base64`.
          oneOf:
            - type: array
              items:
                type: number
            - type: string

    EmbeddingResponse:
      type: object
      required: [object, data, model]
      properties:
        object:
          type: string
          enum: [list]
        model:
          type: string
        data:
          type: array
          items:
            $ref: '#/components/schemas/Embedding'
        usage:
          $ref: '#/components/schemas/Usage'

    Pricing:
      type: object
      description: |
        Prices in micro-PLN (1 PLN = 1 000 000 micro-PLN) per one million
        tokens, effective right now. All prices are illustrative placeholders
        until the launch benchmark; `GET /v1/models` is the source of truth.
      required: [input_micro_pln_per_1m, output_micro_pln_per_1m, currency]
      properties:
        input_micro_pln_per_1m:
          type: integer
          format: int64
        output_micro_pln_per_1m:
          type: integer
          format: int64
          description: 0 for embedding models.
        currency:
          type: string
          enum: [PLN]

    Model:
      type: object
      required: [id, object, created, owned_by, capabilities, context_length, max_output_tokens, pricing]
      properties:
        id:
          type: string
          description: The alias to pass as `model`.
          example: polish-pro
        object:
          type: string
          enum: [model]
        created:
          type: integer
        owned_by:
          type: string
          enum: [siema]
        capabilities:
          type: array
          items:
            type: string
            enum: [chat, embeddings, vision]
        context_length:
          type: integer
          description: Maximum prompt plus completion tokens.
        max_output_tokens:
          type: integer
          description: Ceiling applied to `max_tokens`. 0 for embedding models.
        pricing:
          $ref: '#/components/schemas/Pricing'

    ModelList:
      type: object
      required: [object, data]
      properties:
        object:
          type: string
          enum: [list]
        data:
          type: array
          items:
            $ref: '#/components/schemas/Model'

    Error:
      type: object
      required: [message, type, code, param]
      properties:
        message:
          type: string
          description: Human-readable, safe to show to developers. Never contains prompt, completion or embedding text.
        type:
          type: string
          enum:
            - invalid_request_error
            - authentication_error
            - permission_error
            - insufficient_quota
            - rate_limit_error
            - server_error
            - upstream_error
        code:
          type: string
          description: Stable machine-readable code.
          enum:
            - invalid_json
            - invalid_request
            - context_length_exceeded
            - invalid_api_key
            - org_suspended
            - model_not_found
            - unsupported_endpoint
            - insufficient_quota
            - rate_limit_exceeded
            - concurrency_limit_exceeded
            - no_capacity
            - internal_error
            - upstream_unavailable
            - upstream_timeout
        param:
          type: [string, "null"]
          description: The offending request field, or null.
        request_id:
          type: string
          description: Same value as the response's `X-Request-Id` header.

    ErrorResponse:
      type: object
      required: [error]
      properties:
        error:
          $ref: '#/components/schemas/Error'