agn-sdk 1.0.0

多模型统一接口 SDK - 一套 API 调用所有 AI 模型

🤖 AGN-SDK

Unified API | 5+ Providers | Async-First | Production-Ready | Type-Safe

---

🌐 Language / 语言

English | 中文

---

A unified SDK that calls all AI models through one API — whether it's text chat, image generation, video creation, or speech synthesis.

Built with async-first design, full type safety, and a pluggable adapter architecture. If you're familiar with the OpenAI API, you can use AGN-SDK immediately.

---

✨ Features

Capabilities

Capability	Description	Status
💬 Chat Completion	Multi-turn conversations with AI models	✅ Stable
🖼️ Image Generation	Text-to-image generation	✅ Stable
🎬 Video Creation	Async video generation with polling	✅ Stable
🔊 Speech Synthesis	Text-to-speech generation (Edge TTS / ElevenLabs / Cartesia)	✅ Stable
🎤 Speech Recognition	Audio transcription (Deepgram / AssemblyAI)	✅ Stable
📊 Embeddings	Text embedding vectors (OpenAI / Gemini / Agnes / aggregators)	✅ Stable

Architecture Highlights

• Unified Interface — One API to rule all AI providers (OpenAI, Azure, Anthropic, Gemini, etc.)
• Async-First Design — Full async/await support, built on httpx and anyio
• Adapter Pattern — Add new providers by implementing a single adapter class
• Type Safety — All data models defined with Pydantic v2, full type hints throughout
• Production-Ready — Built-in retry logic, error mapping, parameter normalization
• OpenAI Compatible — Use OpenAI API patterns directly, minimal learning curve

---

📦 Supported Providers

V1.0 (Stable)

Provider	Chat	Image	Video	Base URL
Agnes AI	✅	✅	✅	https://api.agnes.ai/v1
OpenAI	✅	✅	—	https://api.openai.com/v1
Azure OpenAI	✅	✅	—	Azure endpoint

V1.1+ (Coming Soon)

Provider	Chat	Image	Video
Anthropic (Claude)	✅	—	—
Google Gemini	✅	✅	—
Runway	—	—	✅
Pika	—	—	✅
Stability AI	—	✅	—
ByteDance Seedance	✅	✅	✅

Audio & Embedding Providers (Stable)

Provider	TTS	ASR	Embed	Notes
Edge TTS	✅	—	—	Free, no API key, 100+ neural voices
ElevenLabs	✅	—	—	High-quality multilingual voices
Cartesia	✅	—	—	Ultra-low-latency Sonic TTS
Deepgram	—	✅	—	Nova-2/Nova-3, fastest ASR
AssemblyAI	—	✅	—	Enterprise ASR with diarization
OpenAI	—	—	✅	text-embedding-3-small/large
Google Gemini	—	—	✅	gemini-embedding-001
Agnes AI	—	—	✅	Unified embeddings
SiliconFlow / Together / Fireworks / Cloudflare	—	—	✅	Aggregator-hosted embedding models

---

📦 Project Structure

agn-sdk/
├── agn/                              # SDK core code
│   ├── __init__.py                   # SDK entry point
│   ├── client.py                     # Unified client (API layer)
│   ├── router.py                     # Router (routing layer)
│   ├── adapters/                     # Adapter implementations
│   │   ├── base.py                   # BaseAdapter abstract class
│   │   ├── factory.py                # Adapter factory
│   │   ├── agnes.py                  # Agnes AI adapter
│   │   ├── openai.py                 # OpenAI adapter
│   │   ├── azure.py                  # Azure OpenAI adapter
│   │   └── ...                       # More adapters
│   ├── core/                         # Core utilities
│   │   ├── http_client.py            # Async HTTP client
│   │   ├── retry.py                  # Retry mechanism
│   │   ├── errors.py                 # Error definitions
│   │   ├── config.py                 # Configuration
│   │   └── utils.py                  # Utilities
│   └── models/                       # Pydantic data models
│       ├── common.py                 # Common structures
│       ├── chat.py                   # Chat models
│       ├── image.py                  # Image models
│       ├── video.py                  # Video models
│       └── options.py                # Unified options
├── docs/                             # Documentation
│   ├── 01-overview.md                # Project overview
│   ├── 02-architecture.md            # Architecture design
│   └── 03-api-reference.md           # API reference
├── tests/                            # Test suite
├── examples/                         # Usage examples
├── pyproject.toml                    # Project config
└── README.md                         # Project docs (English)

---

🚀 Quick Start

Get started in 3 steps:

Step 1: Install

# From PyPI (coming soon)
pip install agn-sdk

# Or install from source (development mode)
git clone https://github.com/your-org/agn-sdk.git
cd agn-sdk
pip install -e .

Step 2: Configure API Key

# Option A — Environment variable (Recommended)
export AGN_API_KEY='your-api-key'
export AGN_BASE_URL='https://api.agnes.ai/v1'  # Provider-specific base URL

# Option B — .env file (auto-loaded)
echo "AGN_API_KEY=your-api-key" > .env
echo "AGN_BASE_URL=https://api.agnes.ai/v1" >> .env

# Option C — Pass via code
client = Client(provider="agnes", api_key="your-key", base_url="https://api.agnes.ai/v1")

Step 3: Call AI Models

import asyncio
from agn import Client

async def main():
    # Create client
    client = Client(
        provider="agnes",
        api_key="your-api-key",
        base_url="https://api.agnes.ai/v1",
    )
    
    # 💬 Chat Completion
    response = await client.chat(
        model="claude-3-opus",
        messages=[
            {"role": "system", "content": "You are a helpful assistant."},
            {"role": "user", "content": "Hello!"}
        ],
        temperature=0.7,
    )
    print(response.choices[0].message.content)
    
    # 🖼️ Image Generation
    result = await client.image_generate(
        model="dall-e-3",
        prompt="A beautiful sunset over the ocean",
        size="1024x1024",
        quality="hd",
    )
    print(result.data[0].url)
    
    # 🎬 Video Creation (async with polling)
    task = await client.video_create(
        model="video-gen-1",
        prompt="A cat walking in the garden",
        width=1280,
        height=720,
        num_frames=121,
    )
    
    # Poll video status until complete
    while True:
        status = await client.video_poll(task.task_id)
        print(f"Status: {status.status}, Progress: {status.progress}%")
        if status.status in ("completed", "failed"):
            break
    
    print(f"Video URL: {status.video_url}")

if __name__ == "__main__":
    asyncio.run(main())

✨ That's it! You now have a unified interface to all supported AI providers.

---

📖 Complete Usage Reference

Chat Completion

response = await client.chat(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain quantum computing in simple terms."}
    ],
    temperature=0.7,        # Randomness (0.0-2.0)
    max_tokens=1000,        # Max response tokens
    top_p=1.0,              # Nucleus sampling
    frequency_penalty=0.0,   # Repetition penalty
    presence_penalty=0.0,    # Topic diversity
    stream=False,            # Streaming response
)
print(response.choices[0].message.content)

Image Generation

result = await client.image_generate(
    model="dall-e-3",
    prompt="A futuristic city with flying cars",
    size="1024x1024",       # 1024x1024, 1024x1792, 1792x1024
    quality="hd",           # standard or hd
    style="vivid",          # vivid or natural (DALL-E 3)
    n=1,                    # Number of images
)
print(result.data[0].url)   # or result.data[0].b64_json

Video Creation

# Create video task
task = await client.video_create(
    model="video-gen-1",
    prompt="A dramatic sword fight scene",
    width=1280,
    height=720,
    num_frames=121,         # Must satisfy 8n+1 (e.g., 33, 49, 81, 121, 241)
    frame_rate=24,
    seed=42,                # Optional: for reproducibility
)
print(f"Task ID: {task.task_id}")

# Poll until complete
status = await client.video_poll(task.task_id)
while status.status == "in_progress":
    await asyncio.sleep(5)
    status = await client.video_poll(task.task_id)
    
print(f"Video URL: {status.video_url}")

Speech Synthesis (TTS)

# Edge TTS — free, no API key required (install: pip install agn-sdk[edge-tts])
edge_client = Client(provider="edge-tts", api_key="")
result = await edge_client.speech(
    model="edge-tts",
    input="Hello, this is synthesized speech.",
    voice="xiaoxiao",          # short name or full ID: zh-CN-XiaoxiaoNeural
    response_format="mp3",     # mp3 / wav / ogg / pcm
    rate="+10%",               # optional: speed adjustment
)
with open("out.mp3", "wb") as f:
    f.write(result.audio_data)

# OpenAI TTS — uses alloy/echo/nova voices
result = await client.speech(
    model="tts-1",
    input="The quick brown fox jumps over the lazy dog.",
    voice="alloy",
    response_format="mp3",
    speed=1.0,
)

Speech Recognition (ASR)

# Deepgram Nova-2 (fastest) — accepts file path / URL / bytes / base64
result = await client.transcribe(
    model="nova-2",
    file="./meeting.wav",
    language="zh",             # optional: auto-detected if omitted
    smart_format=True,         # optional: punctuation + number formatting
)
print(result.text)
for seg in result.segments or []:
    print(f"[{seg.start:.2f}-{seg.end:.2f}] {seg.text}")

# AssemblyAI — enterprise ASR with speaker diarization
result = await client.transcribe(
    model="best",
    file="./interview.mp3",
    speaker_labels=True,
    sentiment_analysis=True,
)

Text Embeddings

# Single text or batch — returns unified EmbeddingResult
result = await client.embed(
    model="text-embedding-3-small",
    input=["hello world", "machine learning"],
)
vectors = result.get_embeddings()   # list[list[float]]
print(len(vectors), len(vectors[0]))

---

🏗️ Architecture Overview

┌─────────────────────────────────────────────────────────┐
│                    API Layer (Client)                   │
│            chat() / image_generate() / video_create()   │
└─────────────────────────┬───────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                  Router Layer                           │
│          Model routing, load balancing, fallback        │
└─────────────────────────┬───────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                 Adapter Layer                           │
│    BaseAdapter → AgnesAdapter / OpenAIAdapter / ...     │
└─────────────────────────┬───────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                   Core Layer                            │
│      HTTP client, retry, errors, config, utils          │
└─────────────────────────────────────────────────────────┘

• API Layer — Unified Client class, user-facing interface
• Router Layer — Model selection, routing, load balancing
• Adapter Layer — Provider-specific implementations, parameter mapping, response normalization
• Core Layer — Shared utilities (HTTP, retry, errors, config)

---

📋 Adapter Development

Adding a new AI provider is straightforward:

1. Create adapter — Inherit BaseAdapter, implement required methods
2. Register factory — Call AdapterFactory.register("provider_name", YourAdapter)
3. Declare capabilities — Set supported_capabilities list

from agn.adapters.base import BaseAdapter
from agn.adapters.factory import AdapterFactory

class NewProviderAdapter(BaseAdapter):
    provider_type = "newprovider"
    provider_name = "New Provider"
    supported_capabilities = [Capabilities.CHAT, Capabilities.IMAGE_GENERATE]
    
    async def start(self) -> None:
        # Initialize HTTP client
        ...
    
    async def chat(self, model: str, messages: list[ChatMessage], **kwargs):
        # Implement chat logic
        ...
    
    # ... implement other methods

AdapterFactory.register("newprovider", NewProviderAdapter)

---

🧪 Development

# Clone and setup
git clone https://github.com/your-org/agn-sdk.git
cd agn-sdk
python -m venv venv
source venv/bin/activate

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Code formatting
black agn/

# Linting
ruff check agn/

# Type checking
mypy agn/

---

📜 License

MIT License