> ## Documentation Index > Fetch the complete documentation index at: https://docs.omnia-voice.com/llms.txt > Use this file to discover all available pages before exploring further. # Create chat completion > Generate a model response for a conversation. OpenAI-compatible; supports streaming, tool calls, and structured output. Supports gateway fallback models (`fallbacks`) and, when the workspace has [response caching](/billing/caching) enabled, serves identical repeat requests from cache at a discount (see the `X-Omnia-Cache` response header). ## OpenAPI ````yaml /openapi-inference.json post /chat/completions openapi: 3.1.0 info: title: Omnia Inference API description: >- OpenAI-compatible inference endpoints: chat completions, embeddings, rerank, and the Responses API. Authenticated with a workspace API key (sk_sovereign_...). version: 1.0.0 servers: - url: https://gateway.omnia-voice.com/v1 description: Production security: - bearerAuth: [] paths: /chat/completions: post: tags: - Chat summary: Create chat completion description: >- Generate a model response for a conversation. OpenAI-compatible; supports streaming, tool calls, and structured output. Supports gateway fallback models (`fallbacks`) and, when the workspace has [response caching](/billing/caching) enabled, serves identical repeat requests from cache at a discount (see the `X-Omnia-Cache` response header). operationId: createChatCompletion parameters: - name: X-Omnia-Cache-Control in: header required: false schema: type: string enum: - no-cache - no-store description: >- Per-request cache control (only meaningful when the workspace has response caching enabled — it can reduce caching, never enable it). `no-cache`: skip the cache and force a fresh model run, refreshing the stored copy. `no-store`: fresh run AND keep this response out of cache storage entirely. Either way the response carries `X-Omnia-Cache: bypass`. requestBody: required: true content: application/json: schema: type: object required: - model - messages properties: model: type: string description: >- The model id, e.g. Qwen/Qwen3-32B. A dedicated endpoint is addressed as dedicated/. example: Qwen/Qwen3-32B messages: type: array description: The conversation so far. items: type: object properties: role: type: string enum: - system - user - assistant - tool content: description: >- String, or an array of content parts (text/image) for vision. stream: type: boolean default: false description: Stream the response as server-sent events. stream_options: type: - object - 'null' description: >- Streaming options. Usage reporting is included automatically. temperature: type: number description: Sampling temperature (0–2). top_p: type: number description: Nucleus sampling probability mass. top_k: type: integer description: Top-k sampling. 'n': type: integer description: Number of completions to generate. max_tokens: type: integer description: >- Maximum output tokens. Bounded by the model's context window. max_completion_tokens: type: integer description: Alias for max_tokens. stop: description: Up to a few stop sequences (string or array of strings). seed: type: integer description: For more reproducible sampling. frequency_penalty: type: number description: Penalize repeated tokens by frequency. presence_penalty: type: number description: Penalize tokens that have appeared. repetition_penalty: type: number description: Penalize repetition. logit_bias: type: object description: Bias specific tokens. logprobs: type: boolean description: Return log probabilities. top_logprobs: type: integer description: How many top log probabilities to return per token. response_format: type: object description: >- { "type": "json_object" } or { "type": "json_schema", ... } for structured output. tools: type: array description: Function definitions the model may call. items: type: object tool_choice: description: auto, required, none, or a specific function. parallel_tool_calls: type: boolean description: Allow the model to call multiple tools in parallel. reasoning_effort: type: string description: Reasoning effort for reasoning-capable models. user: type: string description: A stable identifier for the end user. fallbacks: type: array maxItems: 2 items: type: string description: >- Up to 2 backup model ids tried in order when the primary fails with a transient error (429/timeout/5xx). Billing uses the model that actually served; the X-Omnia-Fallback-From response header names the model it fell back from. Gateway feature — never forwarded to the model. responses: '200': description: >- A chat completion. When streaming, a text/event-stream of chunks ending with data: [DONE]. headers: X-Omnia-Cache: schema: type: string enum: - hit - miss - bypass description: >- Present when the workspace has response caching enabled: `hit` means this response was served from cache (billed at the cache discount), `miss` means it was generated fresh and may be stored, `bypass` means the request's `X-Omnia-Cache-Control` header forced a fresh run. Absent when caching is off. X-Omnia-Fallback-From: schema: type: string description: >- Present only when a fallback model served the request: the primary model id it fell back from. The body's `model` field names the model that actually answered. content: application/json: schema: $ref: '#/components/schemas/ChatCompletion' '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' '402': $ref: '#/components/responses/InsufficientBalance' '404': $ref: '#/components/responses/ModelNotFound' components: schemas: ChatCompletion: type: object properties: id: type: string example: chatcmpl-... object: type: string example: chat.completion model: type: string choices: type: array items: type: object properties: index: type: integer message: type: object properties: role: type: string example: assistant content: type: string tool_calls: type: array items: type: object finish_reason: type: string example: stop usage: $ref: '#/components/schemas/Usage' Usage: type: object properties: prompt_tokens: type: integer completion_tokens: type: integer total_tokens: type: integer Error: type: object properties: error: type: object properties: message: type: string type: type: string code: type: string responses: BadRequest: description: Missing a required field (e.g. model, messages/input/query/documents). content: application/json: schema: $ref: '#/components/schemas/Error' Unauthorized: description: Missing, malformed, or revoked API key. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: Invalid API key type: invalid_request_error code: invalid_api_key InsufficientBalance: description: Your wallet can't cover the request's worst-case cost. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: Insufficient balance. Please top up your wallet. type: insufficient_quota code: insufficient_balance ModelNotFound: description: The model is not offered (or the id is wrong). content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: The model 'x' does not exist or is not available. type: invalid_request_error code: model_not_found securitySchemes: bearerAuth: type: http scheme: bearer description: >- Your workspace API key, e.g. sk_sovereign_..., sent as Authorization: Bearer . ````