AgentHub SDK

Project Overview#

AgentHub SDK is an open-source toolkit (Apache-2.0 license) developed by the Prism-Shadow organization, designed to solve the problems of fragmented LLM provider interfaces, high switching costs, and Agent debugging difficulties. By defining unified data structures like UniMessage, UniConfig, and UniEvent, it abstracts underlying model differences and allows developers to switch between models at zero cost by simply modifying the model parameter.

Core Values:

• Unified interface abstraction, shielding API differences across vendors
• Native async streaming response support
• Built-in fine-grained tracing without database dependencies
• Bilingual support (Python >=3.11, TypeScript)

Core Capabilities#

AutoLLMClient#

The core client class supports automatic routing to corresponding SDKs via the model parameter (e.g., gpt-5.2, claude-4.5, gemini-3), enabling zero-code model switching.

Supported Models#

• Gemini 3 (Official): Reasoning, tool calling, image understanding
• Claude 4.5 (Official): Reasoning, tool calling, image understanding
• GPT-5.2 (Official): Reasoning, tool calling, image understanding
• GLM-5 (Official/OpenRouter/SiliconFlow): Reasoning, tool calling
• Qwen3 (OpenRouter/SiliconFlow/vLLM): Reasoning, tool calling

Key Features#

1. Unified Interface: Consistent and intuitive interface for Agent development across different LLMs
2. Precise Processing: Automatic handling of interleaved thinking during multi-step tool calls
3. Traceability: Lightweight fine-grained tracing via trace_id parameter
4. Fine-grained Token Statistics: Distinguishes cached_tokens, prompt_tokens, thoughts_tokens, response_tokens

Installation#

Python (PyPI)#

uv add agenthub-python
# or
pip install agenthub-python

TypeScript (npm)#

npm install @prismshadow/agenthub

Quick Start Example#

import asyncio
import os
from agenthub import AutoLLMClient

os.environ["OPENAI_API_KEY"] = "your-openai-api-key"

async def main():
    client = AutoLLMClient(model="gpt-5.2")
    async for event in client.streaming_response_stateful(
        message={"role": "user", "content_items": [{"type": "text", "text": "Hello!"}]},
        config={"temperature": 1.0}
    ):
        print(event)

asyncio.run(main())

Core API#

• streaming_response: Stateless method, requires full message history each time
• streaming_response_stateful: Stateful method, automatically maintains conversation history
• clear_history(): Clear conversation history
• get_history(): Get conversation history

Integration Tools#

Tracer Web Server#

from agenthub.integration.tracer import Tracer
tracer = Tracer()
tracer.start_web_server(host="127.0.0.1", port=5000)

LLM Playground#

from agenthub.integration.playground import start_playground_server
start_playground_server(host="127.0.0.1", port=5001)

Project Structure#

├── src_py/           # Python implementation
├── src_ts/           # TypeScript implementation
└── llmsdk_docs/      # Model-specific documentation

Version Info#

• Current Version: v0.2.0
• License: Apache-2.0
• Primary Languages: Python (52.5%), TypeScript (47.2%)

Note: Model names like GPT-5.2, Claude 4.5, Gemini 3 mentioned in documentation have high version numbers; exact corresponding versions pending verification. npm package cannot be fully verified due to Cloudflare protection.