Chat Completions

POST /v1/chat/completions is the general-purpose LLM endpoint. Send a list of messages, choose a model, and receive an OpenAI-shaped assistant response. LLM API resolves the model to a provider, applies project policies such as routing, caching, and guardrails, then normalizes the provider response. Set stream: true to receive the answer as server-sent events (SSE).

Quick start

curl -X POST "https://api.llmapi.ai/v1/chat/completions" \
  -H "Authorization: Bearer $LLM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [{"role": "user", "content": "Explain embeddings in one sentence."}]
  }'

The response follows the OpenAI Chat Completions shape: read the answer from choices[0].message.content and usage from usage.

{
  "id": "chatcmpl_abc123",
  "object": "chat.completion",
  "model": "gpt-4o-mini",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "Embeddings are..." },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 18,
    "completion_tokens": 12,
    "total_tokens": 30
  }
}

Request shape

model and messages are required. A message can contain plain text or multimodal content such as image URLs, depending on the selected model.

Field	Description
`model`	Model ID or a `provider/model` name for provider pinning.
`messages`	Ordered conversation messages. Use `system`, `developer`, `user`, `assistant`, and tool messages as supported by the model.
`max_tokens` / `max_completion_tokens`	Upper bound for generated output. Use the parameter accepted by the selected model.
`temperature`, `top_p`	Sampling controls; supported ranges are model-dependent.
`stream`	Return SSE deltas instead of one JSON response.
`response_format`	Request text, JSON object, or a JSON Schema response where supported.
`tools`, `tool_choice`	Enable function/tool calling where supported.
`reasoning_effort`, `no_reasoning`	Control reasoning models where supported.
`web_search`, `web_search_options`	Enable web search for models and projects that support it.
`cache_control`	Opt in to provider prompt caching; see prompt caching.

Multimodal input

Vision-capable models accept the OpenAI content-array format:

{
  "model": "gpt-4o-mini",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "What is in this image?" },
        {
          "type": "image_url",
          "image_url": { "url": "https://example.com/photo.jpg" }
        }
      ]
    }
  ]
}

Use the Models catalog to confirm vision, audio_input, video_input, tools, reasoning, and structured-output capabilities before using them.

Streaming

Streaming responses use text/event-stream. Each data: event contains a JSON chunk whose text is usually in choices[0].delta.content; the stream ends with data: [DONE].

curl -N -X POST "https://api.llmapi.ai/v1/chat/completions" \
  -H "Authorization: Bearer $LLM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "stream": true,
    "messages": [{"role":"user","content":"Write a haiku about APIs."}]
  }'

When you need usage in a stream, request it with stream_options: { "include_usage": true } if the selected model supports that option. Tool calls and reasoning content may arrive in separate delta fields rather than content.

Routing and compatibility

Use the Models page or GET /v1/models to find models and capabilities.
You can pin a provider with a provider/model model name, or configure routing and fallback.
The optional X-LLMAPI-Region header accepts eu, us, ap, or global for providers with native multi-region routing.
Vision, audio input, video input, image generation, web search, and tool calling depend on the selected model. See Vision, Audio Input, and Web Search.
Output guardrails that validate complete JSON responses cannot inspect an incomplete streaming response. Use non-streaming mode when a complete structured result must be validated.
For generated images returned inside a chat response, see Image Generation. For dedicated image generation and editing, use Images.