AGENTIC_SETUP.md

Running libEnsemble agentic workflows

Note: The agentic directory in this repo is experimental scripts and has been superceeded by the libEnsemble agent at https://github.com/Libensemble/ensemble_agent

For all workflows, you will need a key to access an LLM.

For example, you can set an OpenAPI key. Requires an OpenAI account. Make sure to check MODEL at top of agentic script and usage rates.

Set user OpenAI API Key:

export OPENAI_API_KEY="sk-your-key-here"

Or if you use Anthropic, you can set.

export ANTHROPIC_API_KEY="sk-ant-your-key-here"

Optionally, you can set the LLM_MODEL env variable to a model name.

Using Argonne inference service (optional)

If you have an ALCF account, you can use Argonne inference service instead of OpenAI.

Authenticate via Globus to obtain ALCF inference token:

pip install openai globus_sdk
wget https://raw.githubusercontent.com/argonne-lcf/inference-endpoints/main/inference_auth_token.py
python inference_auth_token.py authenticate  # Enter globus authentication when prompted.

Set environment variables for ALCF Inference service and model. Obtain API Key:

export OPENAI_BASE_URL=https://inference-api.alcf.anl.gov/resource_server/metis/api/v1
export LLM_MODEL=gpt-oss-120b
export OPENAI_API_KEY=$(python inference_auth_token.py get_access_token)

libe_agent_basic.py

Agent for running libEnsemble scripts with error recovery.

Requires:

pip install libensemble scipy mpmath langchain langchain-openai langchain-anthropic

Running

git clone https://github.com/Libensemble/script-creator.git
cd script-creator/agentic/

To run:

Provide path to directory with your libEnsemble scripts. These will be copied to a working directory generated_scripts/.

python libe_agent_basic.py --scripts tests/scripts_with_errors/

To see script fixes:

cd generated_scripts/versions/
diff 1_copied_scripts/run_example.py 2_fix_attempt_1/run_example.py

This basic agentic script is sufficiently general that it should work with any Python workflow, not just libEnsemble. Any Python scripts in the input directory will be presented to the AI in the case of error. The default run script should be of the form run_*.py.

Alternatively you can run through the web interface (locally).

Running scripts with a binary

To run test scripts_with_exe_with_errors/ , build six_hump_camel.x:

cd tests/six_hump_camel/
gcc six_hump_camel.c -o six_hump_camel.x -lm
cd ../..

and update paths to sim_app and input_file in tests/scripts_with_errors/run_example.py. Missing file paths are not yet corrected automatically.

To run:

Provide path to directory with your libEnsemble scripts. These will be copied to a working directory generated_scripts/.

python libe_agent_basic.py --scripts tests/scripts_with_exe_with_errors/

libe_agent_with_script_generator.py

Agent for libEnsemble scripts with error recovery, including optional use of the script generator MCP tool.

Requires:

pip install libensemble langchain langchain-openai mcp

Node.js - Required for MCP server.

# Linux (Ubuntu/Debian)
sudo apt install nodejs

# Or download from https://nodejs.org/

Install Node.js dependencies:

npm install

Requires this repository to be cloned to access the MCP tool. The file mcp_server.mjs should not be moved from it's original location. This file will be found if you run in the same directory or one below. To run the script elsewhere, use the --mcp-server command line option to point to this file.

Running

cd agentic/

# Use default prompt
python libe_agent_with_script_generator.py

# Custom prompt
python libe_agent_with_script_generator.py "Create my_app APOSMM scripts..."

# Custom prompt in a file
python libe_agent_with_script_generator.py --prompt-file my_prompt.txt

# Use existing scripts (skip MCP generation/tweaking)
python libe_agent_with_script_generator.py --scripts example_scripts/

# Run from anywhere
python libe_agent_with_script_generator.py --mcp-server <path/to/mcp_server.mjs>

Scripts saved to generated_scripts/ directory.

Scripts will be ran, fixes attempted on failure, and reran.