MemoryBench Frontend (Streamlit)

This frontend supports:

• Running off-policy and on-policy experiments.
• Configuring API-based model access (OpenAI or OpenAI-compatible endpoint).
• Choosing memory systems from the existing benchmark methods.
• Monitoring live experiment logs.
• Browsing results and every dialogue item, including the assembled memory context prompt.

Hint: You can use Code Agent like Claude Code to quickly configure the environment and launch the frontend. You can prompt the agent with "Help me set up the environment and launch the MemoryBench frontend for me under the instruction in frontend/README.md" or similar instructions. See For Code Agent for more details.

0. Install project dependencies

pip install -r requirements.txt
cd baselines/mem0
pip install -e .

Then install nltk data in python:

import nltk
nltk.download('punkt')
nltk.download('wordnet')
nltk.download('stopwords')

Download Huggingface Dataset to local directory (optional, but can speed up the first run):

huggingface-cli download --repo-type dataset --resume-download THUIR/MemoryBench --local-dir /path/to/MemoryBench

1. Install frontend dependency

pip install -r frontend/requirements.txt

2. Launch frontend

python -m streamlit run frontend/streamlit_app.py

3. Key notes

• The app only writes new runtime artifacts:
  • frontend/runtime_configs/ for temporary configs.
  • Your chosen output directory for experiment results.

For Code Agent

Use these commands from the repository root to set up the environment and launch the frontend:

If conda is available, it's recommended to create a new environment (ask the user whether to create a new conda environment named memorybench). If yes, then:

conda create -n memorybench python=3.10
conda activate memorybench

Then:

pip install -r requirements.txt
cd baselines/mem0
pip install -e .
cd ../..
pip install -r frontend/requirements.txt
python -c "import nltk; [nltk.download(x) for x in ('punkt', 'wordnet', 'stopwords')]"
python smoke_test.py

python smoke_test.py is smoke test. You need to fix any errors until it runs successfully. If it runs successfully, then launch the frontend:

python -m streamlit run frontend/streamlit_app.py