How to Use BioMaster

BioMaster can be used both via command-line (CLI) and through a graphical user interface (GUI). It supports fully automated multi-agent orchestration and task execution with detailed output tracking.

Using from Terminal (CLI)

1. Configure your environment

   • Ensure dependencies are installed:

     conda create -n agent python=3.12
     conda activate agent
     pip install -r requirements.txt
     

   • Set your OpenAI API key and base URL in run.py or examples/file.py.

2. Prepare your working directory

   • Move the example script (e.g. examples/file.py) into the root folder:

     mv examples/file.py ./BioMaster/
     

   • Download and place required data in the ./data/ directory:

     • Example dataset: https://drive.google.com/drive/folders/1vA3WIAVXVo4RZSqXKsItEZHVaBgIIv_E?usp=sharing

3. Set a unique task ID in your script:

   manager = Biomaster(api_key, base_url, excutor=True, id='001')
   

4. Run the script

   conda activate agent
   python run.py
   

   Or, for examples:

   python example1.py
   

5. Inspect the output

   Results are saved in the following structure:

   • Execution plan: ./output/{id}_PLAN.json

   • Execution scripts: ./output/{id}_Step_{step}.sh

   • Step logs: ./output/{id}_DEBUG_Output_{step}.json

   • All files: ./output/{id}/

Using from UI (GUI)

1. Start the GUI

   conda activate agent
   python runv.py
   

2. Open the browser

   Navigate to:

   http://127.0.0.1:7860/
   

3. Interactively configure your task

   • Set your API Key and Base URL

   • Assign a unique Task ID

   • Specify your data paths

   • Define your goal (e.g., “WGS variant calling”)

   • Click “Generate Plan”, then “Execute Plan”

4. View and manage results

   • Click “Load and Show” to inspect results

   • Click “Stop PLAN” to interrupt an ongoing task

Output Format

BioMaster generates structured outputs:

• Execution Plan: output/{id}_PLAN.json – Full step-by-step plan

• Execution Scripts: output/{id}_Step_{n}.sh – Shell script per step

• Execution Logs: output/{id}_DEBUG_Output_{n}.json – Debug output with execution status

• Data Output Directory: output/{id}/ – Folder containing all intermediate and final results

Tips & Recommendations

• Always use a unique task ID to avoid overwriting previous results.

• To modify the generated plan or script:

  • Comment out manager.execute_PLAN(…) and edit output/{id}_PLAN.json

  • Edit Step_{n}.sh to revise any command

  • Re-run python run.py to resume

• To retry a failed step, delete or set stats: false in its DEBUG_Output_{n}.json

• Before adding new tools or workflows, update the PLAN RAG or EXECUTE RAG JSON files in ./doc/, then delete ./chroma_db/ to refresh embeddings.