OpenAI-Compatible Provider Support

ToolUniverse agentic tools support a generic OPENAI provider backed by the OpenAI Python SDK. Use it for OpenAI directly or for OpenAI-compatible hosted endpoints.

Configuration

Set an API key:

export OPENAI_API_KEY=your_key_here

For OpenAI-compatible endpoints, also set a base URL:

export OPENAI_BASE_URL=https://your-provider.example/v1

Then configure an agentic tool with api_type: "OPENAI":

{
  "name": "openai_compatible_reasoning_tool",
  "type": "AgenticTool",
  "prompt": "Answer this question: {question}",
  "input_arguments": ["question"],
  "api_type": "OPENAI",
  "model_id": "gpt-4o-mini",
  "parameter": {
    "type": "object",
    "properties": {
      "question": {"type": "string"}
    },
    "required": ["question"]
  }
}

Environment Variables

OPENAI_API_KEY

Required API key for the selected endpoint.

OPENAI_BASE_URL

Optional OpenAI-compatible API base URL. If unset, the OpenAI SDK default is used.

OPENAI_MAX_TOKENS_BY_MODEL

Optional JSON mapping of model ID or model prefix to default max output tokens. Example: {"gpt-4o": 4096}.

OPENAI_DEFAULT_MODEL_LIMITS

Optional JSON mapping that extends or overrides ToolUniverse’s built-in model-family defaults.

Provider Notes

Use OPENAI for OpenAI-compatible chat completions endpoints. Use OPENROUTER when you want OpenRouter-specific headers and model naming, and use VLLM for self-hosted vLLM servers.