vuluu2k/deepinfra

DeepInfra API

This skill focuses on integrating DeepInfra's REST API into any project. Do not use language-specific SDKs (like the openai NodeJS package or LangChain) unless the project already heavily depends on them.

Key Integration Rules

• Use the Repository's Native HTTP Client: Before writing code to call the DeepInfra API, you must analyze the current repository to identify how external HTTP requests are typically made (e.g., a custom useFetch wrapper, an Axios instance, Elixir Req, or standard fetch).
• Adapt to Existing Patterns: Send the DeepInfra request using the exact fetching mechanism discovered in the project. Do not introduce raw fetch commands or new libraries if a standard API wrapper already exists.
• Endpoint: The base URL for most AI inference tasks (chat completions) is POST https://api.deepinfra.com/v1/openai/chat/completions.
• Authentication: You must set the Authorization: Bearer <API_KEY> header. Always read the API key from environment variables (e.g., process.env.DEEPINFRA_API_KEY or System.get_env("DEEPINFRA_API_KEY")) rather than hardcoding.
• Headers: Always include Content-Type: application/json.

Quick Reference

Standard JSON Payload

Adapt this payload to the project's standard API caller:

{
  "model": "meta-llama/Meta-Llama-3-70B-Instruct",
  "messages": [
    {
      "role": "user",
      "content": "Hello!"
    }
  ]
}

Example Response Structure

DeepInfra yields an OpenAI-compatible JSON structure. Handle the response parsing according to the repository's typical data-transformation layers:

{
  "id": "chatcmpl-guMTxWgpFf",
  "object": "chat.completion",
  "created": 1694623155,
  "model": "meta-llama/Meta-Llama-3-70B-Instruct",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Hello! It's nice to meet you."
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 15,
    "completion_tokens": 16,
    "total_tokens": 31
  }
}

Supported Models

Always use the exact model ID from the DeepInfra catalog, such as:

• deepseek-ai/DeepSeek-V3
• meta-llama/Meta-Llama-3-70B-Instruct