Bohrium Science Agent SDK

English | 简体中文

Transform Scientific Software into AI Assistants — 3 Steps to Intelligent Transformation

📖 Introduction

The Bohrium platform introduces the bohr-agent-sdk Scientific Agent Development Kit, enabling AI systems to truly execute professional scientific tasks and helping developers quickly build their own specialized research agents. Through a three-step process — Invoking MCP Tools, Orchestrating Agent Workflows, and Deploying Services — any scientific software can be rapidly transformed into an AI assistant.

✨ Core Features

🎯 Intelligent Task Management: Simplified Development, Standardized Output

With a decorator pattern, just a few annotations can quickly transform scientific computing programs into MCP standard services. Built-in application templates turn scattered research code into standardized, reusable intelligent components.

🔧 Multi-Backend Framework Support

Supports mainstream Agent open frameworks including Google ADK, Langraph, and Camel, providing flexible choices for developers familiar with different technology stacks.

☁️ Flexible Deployment: Local Development, Cloud Production

Dual-mode architecture supports seamless transition between development and production. Local environments enable rapid iteration and feature validation, while Bohrium's cloud GPU clusters handle production-grade computing tasks. The SDK automatically manages the complete workflow of task scheduling, status monitoring, and result collection, with built-in file transfer mechanisms for handling large-scale data uploads and downloads. Developers focus on core algorithm implementation while infrastructure management is fully automated.

🖼️ Visual Interactive Interface: Professional Presentation, Intuitive Operation

Based on the modern React framework, deploy fully-featured web applications with one click. Built-in 3D molecular visualization engine supports multiple structure formats and rendering modes for interactive molecular structure display. Real-time data synchronization ensures instant computing status updates, while multi-session management supports parallel task processing. Integrated with enterprise-grade features including file management, project switching, and permission control. Transform command-line tools into professional visual applications, significantly enhancing user experience and tool usability.

🖼️ Interface Showcase

Scientific Computing Master Console

Visual Interactive Interface

🚀 Quick Start

Installation

bash pip install bohr-agent-sdk -i https://pypi.org/simple --upgrade

Build Your Research Agent in 3 Steps

Step 1: Get Project Templates

```bash

Get calculation project template

dp-agent fetch scaffolding --type=calculation

Get device control project template

dp-agent fetch scaffolding --type=device

Get configuration file

dp-agent fetch config ```

Step 2: Develop Your Agent

Lab Mode Development Example

```python from typing import Dict, TypedDict from dp.agent.device.device import Device, action, BaseParams, SuccessResult

class TakePictureParams(BaseParams): """Picture taking parameters""" horizontal_width: str # Image horizontal width

class PictureData(TypedDict): """Picture data structure""" image_id: str

class PictureResult(SuccessResult): """Picture taking result""" data: PictureData

class MyDevice(Device): """Custom device class""" devicename = "mydevice"

@action("take_picture")
def take_picture(self, params: TakePictureParams) -> PictureResult:
    """
    Execute picture taking action

    Through the @action decorator, automatically register this method as an MCP standard service
    """
    hw = params.get("horizontal_width", "default")
    # Execute actual device control logic
    return PictureResult(
        message=f"Picture taken with {self.device_name}",
        data={"image_id": "image_123"}
    )

```

Cloud Mode Development Example

```python """ MCP protocol-based cloud device control example """ import signal import sys from dp.agent.cloud import mcp, getmqttcloudinstance from dp.agent.device.device import TescanDevice, registermcp_tools

def signalhandler(sig, frame): """Graceful shutdown handling""" print("Shutting down...") getmqttcloudinstance().stop() sys.exit(0)

def main(): """Start cloud services""" print("Starting Tescan Device Twin Cloud Services...")

# Register signal handler
signal.signal(signal.SIGINT, signal_handler)

# Create device instance
device = TescanDevice(mcp, device)

# Automatically register device tools to MCP server
# register_mcp_tools implements automatic registration through Python introspection
register_mcp_tools(device)

# Start MCP server
print("Starting MCP server...")
mcp.run(transport="sse")

if name == "main": main() ```

Step 3: Run and Deploy

```bash

Local lab environment

dp-agent run tool device

Cloud computing environment

dp-agent run tool cloud

Scientific calculation mode

dp-agent run tool calculation

Start agent (with Web UI)

dp-agent run agent --config

Debug mode

dp-agent run debug ```

🏗️ Project Structure

After running dp-agent fetch scaffolding, you'll get a standardized project structure:

your-project/ ├── lab/ # Lab mode │ ├── __init__.py │ └── tescan_device.py # Device control implementation ├── cloud/ # Cloud mode │ ├── __init__.py │ └── mcp_server.py # MCP service implementation ├── calculation/ # Calculation mode │ └── __init__.py ├── .env # Environment configuration └── main.py # Main program entry

⚙️ Configuration

Configure necessary environment variables in the .env file:

```bash

MQTT connection configuration

MQTTINSTANCEID=yourinstanceid MQTTENDPOINT=yourendpoint MQTTDEVICEID=yourdeviceid MQTTGROUPID=yourgroupid MQTTAK=youraccesskey MQTTSK=yoursecretkey

Computing resource configuration

BOHRIUMUSERNAME=yourusername BOHRIUMPASSWORD=yourpassword ```

Note: The dp-agent fetch config command automatically downloads configuration files and replaces dynamic variables (such as MQTTDEVICEID). For security reasons, this feature is only available in internal network environments.

🎯 Application Scenarios

• Materials Science Computing: Molecular dynamics simulation, first-principles calculations
• Bioinformatics Analysis: Gene sequence analysis, protein structure prediction
• Laboratory Equipment Control: Intelligent control of research equipment such as electron microscopes and X-ray diffractometers
• Data Processing Workflows: Automated data cleaning, analysis, and visualization
• Machine Learning Training: Model training, hyperparameter optimization, result evaluation

🔧 Advanced Features

File Management

```bash

Upload files to cloud

dp-agent artifact upload

Download cloud files

dp-agent artifact download ```

Task Monitoring

The SDK provides real-time task status monitoring, supporting: - Task queue management - Computing resource scheduling - Automatic result collection - Exception handling and retry mechanisms

📚 Documentation & Support

• 📖 Detailed Documentation