Add llm-async

[sonic182]

What is this Python project?

llm_async is an async-first, lightweight Python library for interacting with modern Large Language Model (LLM) providers such as OpenAI, Google Gemini, Anthropic Claude, and OpenRouter.

It provides a unified async API for chat completions, streaming responses, tool/agent execution, and JSON-schema-validated structured outputs — all built with asyncio and designed for production integration.

Key Features

• Async-first design — built entirely around asyncio, no blocking I/O
• Unified provider interface — same API for OpenAI, Gemini, Claude, and OpenRouter
• Automatic tool execution — define tools once and use them across providers
• Pub/Sub events — real-time event emission for tool execution (start/complete/error)
• Structured outputs — enforce JSON Schema validation across supported models
• Extensible architecture — easily add new providers by inheriting from BaseProvider
• Streaming support — async iterator interface for live model responses

GitHub: https://github.com/sonic182/llm-async

---

What's the difference between this Python project and similar ones?

• Unlike synchronous SDKs (e.g. openai, anthropic), llm_async is async-first, not a wrapper.
• Unlike high-level frameworks (e.g. LangChain, LlamaIndex), it’s minimal and provider-agnostic — focused on clean async primitives rather than orchestration layers.
• Supports tool calling + structured outputs + streaming under one unified API surface.
• Designed for developers who want low-level control and high throughput without extra dependencies.

---

Anyone who agrees with this pull request could submit an Approve review to it.