Skip to main content

​
OpenAI-compatible surface

Every GeneralCompute endpoint mirrors the OpenAI API shape (chat.completions, models, etc.). Use any existing OpenAI SDK or the official @generalcompute/sdk / generalcompute packages — no request body changes required.

​
Tool calling & JSON mode

Function and tool calling are fully supported. Define tools exactly as you would with OpenAI and inspect the returned tool_calls array to decide which function to execute. 
const completion = await client.chat.completions.create({
  model: "deepseek-v3.2",
  messages,
  tools: [
    {
      type: "function",
      function: {
        name: "get_weather",
        description: "Look up the weather for a city",
        parameters: {
          type: "object",
          properties: { location: { type: "string" } },
          required: ["location"],
        },
      },
    },
  ],
  response_format: { type: "json_object" }, // JSON mode
});
Tool calls stream through the same delta events you receive from OpenAI, so orchestrators such as LangChain, Vercel AI SDK, and LlamaIndex work out of the box.

​
Vision

Pass images alongside text in the messages array using the OpenAI "image_url" parts. Vision is available on models like llama-4-maverick-17b and gemma-3-12b-it. 
response = client.chat.completions.create(
    model="llama-4-maverick-17b",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Describe this image"},
                {
                    "type": "image_url",
                    "image_url": {"url": "https://example.com/photo.jpg"},
                },
            ],
        }
    ],
)

​
Reasoning models

Reasoning-first models (deepseek-v3.2, deepseek-r1-0528, etc.) are deployed with long context windows and higher thinking timeouts. Use them when you need chain-of-thought quality: 
const completion = await client.chat.completions.create({
  model: "deepseek-v3.2",
  messages,
  temperature: 0.1,
  max_completion_tokens: 2048,
});
Combine reasoning with tool calling to build autonomous agents without changing your orchestrator code.