> ## Documentation Index > Fetch the complete documentation index at: https://docs.fastrun.si/llms.txt > Use this file to discover all available pages before exploring further. # Create a chat completion > Generates the next assistant message for a conversation. Compatible with OpenAI's Chat Completions API. Set `stream: true` to receive the answer as server-sent events. ## OpenAPI ````yaml /openapi.yaml post /chat/completions openapi: 3.1.0 info: title: fastrun API version: '1.0' description: >- OpenAI-compatible inference API. Point any OpenAI SDK at https://api.fastrun.si/v1 and send a fastrun API key as a bearer token. servers: - url: https://api.fastrun.si/v1 security: - bearerAuth: [] tags: - name: Chat description: Conversations with chat messages. - name: Completions description: Raw text continuation. - name: Models description: The models you can use. paths: /chat/completions: post: tags: - Chat summary: Create a chat completion description: >- Generates the next assistant message for a conversation. Compatible with OpenAI's Chat Completions API. Set `stream: true` to receive the answer as server-sent events. operationId: createChatCompletion requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/ChatCompletionRequest' example: model: google/gemma-4-31b-it messages: - role: system content: You are a concise assistant. - role: user content: Explain prefix caching in one sentence. max_tokens: 200 responses: '200': description: >- The completion. With `stream: true` the body is a `text/event-stream` of `chat.completion.chunk` events ending with `data: [DONE]`. content: application/json: schema: $ref: '#/components/schemas/ChatCompletion' text/event-stream: schema: $ref: '#/components/schemas/ChatCompletionChunk' '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' '429': $ref: '#/components/responses/TooManyRequests' '502': $ref: '#/components/responses/BadGateway' '503': $ref: '#/components/responses/Unavailable' components: schemas: ChatCompletionRequest: type: object required: - model - messages properties: model: type: string description: Model ID. See [Models](/models). examples: - google/gemma-4-31b-it messages: type: array description: The conversation so far, oldest first. minItems: 1 items: $ref: '#/components/schemas/Message' stream: type: boolean default: false description: >- Send the answer as server-sent events while it is generated. See [Streaming](/guides/streaming). stream_options: type: object description: Options for streamed responses. properties: include_usage: type: boolean default: false description: >- Send one last chunk with `usage` and an empty `choices` before `data: [DONE]`. max_tokens: type: integer minimum: 1 maximum: 16384 description: Maximum number of tokens to generate. temperature: type: number minimum: 0 maximum: 2 description: Sampling temperature. Higher is more random; 0 is greedy. top_p: type: number minimum: 0 maximum: 1 description: >- Nucleus sampling. Considers the smallest set of tokens whose probabilities add up to `top_p`. top_k: type: integer minimum: 0 maximum: 1000 description: Considers only the `top_k` most likely tokens. 0 turns it off. min_p: type: number minimum: 0 maximum: 1 description: Drops tokens less likely than `min_p` times the most likely token. frequency_penalty: type: number minimum: -2 maximum: 2 description: Penalizes tokens by how often they already appeared. presence_penalty: type: number minimum: -2 maximum: 2 description: Penalizes tokens that already appeared at all. repetition_penalty: type: number minimum: 0 maximum: 2 description: Multiplicative penalty on repeated tokens. 1 turns it off. stop: description: >- Up to 4 sequences where generation stops. The stop sequence is not included in the output. oneOf: - type: string - type: array maxItems: 4 items: type: string seed: type: integer minimum: 0 maximum: 2147483647 description: Seed for sampling. logprobs: type: boolean default: false description: Return the log probability of each output token. top_logprobs: type: integer minimum: 0 maximum: 20 description: >- With `logprobs`, also return this many most likely alternatives at each position. tools: type: array description: >- Functions the model may call. See [Tool calling](/guides/tool-calling). items: $ref: '#/components/schemas/Tool' tool_choice: description: >- `auto` (default when `tools` is set) lets the model decide, `none` turns tools off, `required` forces a call, and an object names the function to call. oneOf: - type: string enum: - none - auto - required - type: object required: - type - function properties: type: type: string enum: - function function: type: object required: - name properties: name: type: string response_format: description: >- Constrain the output to JSON. See [Structured outputs](/guides/structured-outputs). oneOf: - type: object title: Text required: - type properties: type: type: string enum: - text - type: object title: JSON object required: - type properties: type: type: string enum: - json_object - type: object title: JSON schema required: - type - json_schema properties: type: type: string enum: - json_schema json_schema: type: object required: - name - schema properties: name: type: string schema: type: object description: A JSON Schema the output must match. strict: type: boolean reasoning_effort: type: string enum: - none - low - medium - high description: >- Turns the model's thinking on (`low`, `medium`, `high`) or off (`none`). Off when omitted. See [Reasoning](/guides/reasoning). chat_template_kwargs: type: object description: >- Extra variables for the model's chat template, such as `{"enable_thinking": true}`. See [Reasoning](/guides/reasoning). additionalProperties: true ChatCompletion: type: object properties: id: type: string object: type: string enum: - chat.completion created: type: integer description: Unix time in seconds. model: type: string choices: type: array items: type: object properties: index: type: integer message: type: object properties: role: type: string enum: - assistant content: type: - string - 'null' reasoning: type: - string - 'null' description: The model's thinking, when reasoning is on. tool_calls: type: array items: $ref: '#/components/schemas/ToolCall' finish_reason: type: string enum: - stop - length - tool_calls description: >- `stop` at a natural end or stop sequence, `length` at `max_tokens` or the context limit, `tool_calls` when the model called a tool. logprobs: type: - object - 'null' usage: $ref: '#/components/schemas/Usage' ChatCompletionChunk: type: object description: 'One server-sent event (`data: {...}`).' properties: id: type: string object: type: string enum: - chat.completion.chunk created: type: integer model: type: string choices: type: array items: type: object properties: index: type: integer delta: type: object properties: role: type: string content: type: - string - 'null' reasoning: type: - string - 'null' tool_calls: type: array items: type: object finish_reason: type: - string - 'null' usage: description: >- Only on the last chunk, and only with `stream_options.include_usage`. oneOf: - $ref: '#/components/schemas/Usage' - type: 'null' Message: type: object required: - role properties: role: type: string enum: - system - user - assistant - tool content: description: >- The message text, or a list of text parts. fastrun serves text only; image, audio and video parts are rejected. May be null on an assistant message that only calls tools. oneOf: - type: string - type: 'null' - type: array items: type: object required: - type - text properties: type: type: string enum: - text text: type: string tool_calls: type: array description: On assistant messages, the tool calls the model made. items: $ref: '#/components/schemas/ToolCall' tool_call_id: type: string description: On tool messages, the `id` of the call this message answers. Tool: type: object required: - type - function properties: type: type: string enum: - function function: type: object required: - name properties: name: type: string description: type: string parameters: type: object description: JSON Schema of the arguments. ToolCall: type: object required: - id - type - function properties: id: type: string type: type: string enum: - function function: type: object required: - name - arguments properties: name: type: string arguments: type: string description: The arguments as a JSON string. Usage: type: object properties: prompt_tokens: type: integer description: Input tokens, including cached ones. completion_tokens: type: integer description: Generated tokens, including reasoning tokens. total_tokens: type: integer prompt_tokens_details: type: object properties: cached_tokens: type: integer description: >- Input tokens served from the prompt cache. See [Prompt caching](/guides/prompt-caching). Error: type: object required: - error properties: error: type: object required: - message - type - code properties: message: type: string description: What went wrong, for people. type: type: string description: Error category, for code. enum: - invalid_request_error - authentication_error - rate_limit_error - upstream_error - unavailable code: type: integer description: The HTTP status. responses: BadRequest: description: >- The request is malformed: the body is not a JSON object, `model` is missing, or the model rejected a parameter (for example, a prompt longer than the context window). content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: model is required type: invalid_request_error code: 400 Unauthorized: description: >- The API key is missing, unknown or revoked, or the account has not been approved yet. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: invalid or missing API key type: authentication_error code: 401 TooManyRequests: description: >- Every GPU serving the model is busy. Retry after the number of seconds in `Retry-After`. headers: Retry-After: $ref: '#/components/headers/RetryAfter' content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: all backends for this model are at capacity type: rate_limit_error code: 429 BadGateway: description: The GPU servers failed to answer. Safe to retry. headers: Retry-After: $ref: '#/components/headers/RetryAfter' content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: all backends refused the request type: upstream_error code: 502 Unavailable: description: >- Temporarily unavailable: no GPU server for the model is up, or API keys cannot be checked right now. Retry after the number of seconds in `Retry-After`. headers: Retry-After: $ref: '#/components/headers/RetryAfter' content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: no backend is available for this model type: unavailable code: 503 headers: RetryAfter: description: Seconds to wait before retrying. schema: type: integer minimum: 1 securitySchemes: bearerAuth: type: http scheme: bearer description: A key from https://fastrun.si/dashboard/keys (starts with `fr_`). ```` This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.